Crane Pose Estimation Using UWB Real-TimeLocation System

C. Zhang1; A. Hammad, M.ASCE2; and S. Rodriguez3

Abstract:Operating a crane is a complex job, which requires not only the experience of the operator, but also sufficient and appropriate real-time support to conceive and react to the environment. To help the crane operator, crane pose estimation is necessary to predict potentialcollisions. Environment perception technologies are essential to update environment information. Location data of the components of thecranes should be used to calculate the pose of the crane that can be used for collision avoidance. This paper aims to investigate how to collectand efficiently process the location data in near real time using ultra wideband (UWB) technology for providing intelligent support to craneoperators. First, the requirements of using UWB technology in construction sites to track crane movements are defined. Then, the details ofthe UWB system setting method are investigated to decide the location of sensors and the number and location of tags attached to differentcomponents of a crane. A location data processing method is proposed to improve data quality by filtering noisy data and filling in missingdata in near real time. An outdoor test is presented to demonstrate the feasibility of applying the proposed approach. Location data of a craneboom are collected and processed in near real time. The results of the test show a good potential to calculate the poses of crane booms usingUWB real-time location system (RTLS). DOI: 10.1061/(ASCE)CP.1943-5487.0000172. © 2012 American Society of Civil Engineers.

CE Database subject headings: Cranes; Construction equipment; Estimation.

Author keywords: Cranes; Pose estimation; Real-time location system; Ultra wideband.

Introduction

Previous research has indicated that machinery-related incidents werethe fourth leading cause of traumatic occupational fatalities in theconstruction industry between 1980 and 1992, resulting in1,901 deaths (2.13 deaths per 100,000 workers) (NIOSH 2007).The same research has indicated that the construction equipment mostfrequently associated with fatalities are cranes (17%), excavators(15%), tractors (15%), loaders (9%), and pavers (7%). In 2006, therewere 72 crane-related fatal occupational injuries in the United States(Bureau of Labor Statistics 2008). In Canada, there were 56 accidentsrelated to cranes in the province of British Columbia in 2006 (Work-SafeBC 2010); and during the period of 1974 to 2002, there were 23accidents with injuries, 26 accidents with death, and 13 accidents withmaterial damage related to cranes in Quebec province [Commissionde la santé et de la sécurité du travail du Québec (CSST) 2010].Furthermore, crane accident statistics are limited because typicallyonly deaths and injuries are reported. Property damage incidentsare usually not reported; however, the seriousness of a crane accidentis self-evident (Task Committee on Crane Safety on ConstructionSites 1998). It is estimated that one crane upset occurs during every

10,000 h of crane use. Approximately 3% of upsets result in death,8% in lost time, and 20% in damage to property other than the crane.According to Beavers et al. (2006), mobile cranes represented over88% of the fatal crane-related events. These data suggest that moreemphasis should be put on the operation of mobile cranes.

The present paper is part of a research program for developing amulti-agent system (Zhang et al. 2009a, b, 2010; AlBahnassi andHammad 2012) that aims to improve construction safety by providingintelligent assistance, such as giving alarm to operators and workers,and replanning the path of a crane when a potential collision is de-tected. The system integrates real-time location systems (RTLSs),path planning and replanning algorithms, and multiagent communi-cation and negotiation. The present paper focuses on the requirementsand issues related to the collection of accurate data in near real-timeand processing the data into information that can be used for collisionavoidance and path replanning. The objectives of this paper are (1) de-fining an extended set of requirements of ultra wideband (UWB)RTLSs for construction sites, mainly for tracking crane boom move-ment; (2) investigating a setting method for a UWB system, such asthe location and number of sensors and tags, to fulfil the require-ments; (3) proposing a method for processing sensed location datainto information that can be used for near real-time decision supportsystems; and (4) testing the proposed approach in a detailedcase study.

This paper is organized as follows. The next section reviewsUWB tracking technology. After that, the methodology of applyingUWBRTILS on construction sites is described, which is followed bya case study. The last section has the conclusions and future work.

Previous Research About Tracking ConstructionEquipment

Tracking Technologies

A construction site is dynamic, which requires continuous updatingof the location data of all moving objects, including equipment

1Lecturer, Dept. of Civil Engineering, Xi’an Jiaotong-Liverpool Univ.,Suzhou, Jiangsu 215123, China; formerly, Ph.D. student, Concordia Insti-tute for Information Systems Engineering, Concordia Univ., Montreal,QC H4B 1R6, Canada. E-mail: lilyzc2003@hotmail.com

2Professor, Concordia Institute for Information Systems Engineering,Concordia Univ., Montreal, QC H4B 1R6, Canada (corresponding author).E-mail: hammad@ciise.concordia.ca

3Graduate Student, Concordia Institute for Information Systems Engi-neering, Concordia Univ., Montreal, QC H4B 1R6, Canada.

Note. This manuscript was submitted on December 13, 2010; approvedon September 30, 2011; published online on October 3, 2011. Discussionperiod open until February 1, 2013; separate discussions must be submittedfor individual papers. This paper is part of the Journal of Computing inCivil Engineering, Vol. 26, No. 5, September 1, 2012. © ASCE, ISSN0887-3801/2012/5-625-637/$25.00.

JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012 / 625

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

and workers, to mitigate safety risks. The most popular trackingtechnology used on construction sites is the global positioning sys-tem (GPS), which is widely used in construction, mining, survey-ing, and infrastructure projects. For example, in earthmovingprojects, GPS and total station technology are used to accuratelyposition the blade of the excavator in real time, significantly reduc-ing material overages and dramatically improving contractors' pro-ductivity and profitability (Trimble GCS900 2010). However, GPSrequires direct line-of-sight from the satellites to the receiver, andaccurate GPS receivers are expensive to install on every movingobject on site.

On-board instrumentation (OBI) has been used to collect dataabout the equipment configuration (Navon et al. 2004). However,not all cranes are equipped with OBI, e.g., older types of cranes andcranes with small capacity. Furthermore, even when the OBIis available, it only provides the kinematic geometry of the craneswithout the location relative to other objects. Therefore, this infor-mation cannot directly be used for collision avoidance. Othertracking technologies have been applied in several research proj-ects, such as infrared, optical, ultrasound, and radio frequency iden-tification (RFID) technologies. Chae and Yoshida (2008) discussedcollecting data on site using RFID active tags to prevent collisionaccidents. BodyGuard-Vehicle Proximity Alert and CollisionAvoidance System (Orbit Communications 2008) is an RFID-basedsystem that offers continuous detection and notification of proxim-ity between moving objects and other moving or fixed objectsby setting up protection zones around a vehicle, equipment, andbuildings to offer continuous protection for valuable resources.However, RFID can give only approximate locations.

Recently, RTLSs have been applied in construction to trackmoving objects. Ward (2007) compared different location technol-ogies, such as passive RFID, electromagnetic, laser, ultrasound, in-frared (IR) proximity, conventional radio frequency (RF) timing,UWB, wireless local area network (WLAN), received signalstrength (RSS), and assisted GPS (A-GPS). This comparisonwas carried out based on the accuracy and the coverage offeredby each technology to identify the ideal technology. The resultshowed that the UWB can provide a relatively high accuracy withcoverage of approximately 100 m or more depending on the signalstrength of the tags.

UWB Technology

UWB can transmit large amounts of digital data over a wide spec-trum of frequency bands at very low power (less than 0.5 mW)(Ghavami et al. 2004). It should be noted that UWB is a specialkind of RFID. With conventional radio frequency, reflections incongested environments distort the direct path signal, making ac-curate pulse timing difficult, whereas with UWB, the direct pathsignal can be distinguished from the reflections, making pulsetiming easier. The accuracy of the UWB system applied in con-struction can reach 10 cm (Cho et al. 2010). The range of someavailable commercial UWB systems is up to 300 to 500 m.

AUWB system consists of several sensors and multiple tags. Themaster sensor receives and synchronizes the timing data from theother slave sensors. Each tag registers with its containing sensor cell,and is inserted into the schedule for that cell. When a tag emits asignal, this signal is picked up by one or more sensors in the cell.The slave sensors decode the UWB signal and send the angle ofarrival and timing information back to the master sensor throughan Ethernet connection. The master sensor accumulates all senseddata and computes the location based on trilateration. In someUWB systems, the trilateration is based only on time difference ofarrival (TDOA) and the angle of arrival (AOA) technique is not used.

Related Research Using UWB

Researchers have started to investigate the usability of UWB onconstruction sites. For example, Teizer et al. (2008) attached anUWB tag to the hook of a crane to track the position of the hook.Construction Metrology and Automation Group (CMAG) has beenmeasuring the performance of UWB tracking technology in con-struction (Saidi and Lytle 2008). Giretti et al. (2009) indicated thatUWB behavior is rather constant during most parts of the construc-tion progress. They noted that in an open area, tests confirmed anaccuracy of approximately 30 cm. They also discussed a safetymanagement system that gives an alarm when a worker is ap-proaching a static, known dangerous area. Teizer et al. (2010) pro-posed using UWB for proactive safety, which works in real time toalert personnel of the dangers occurring, and reactive safety, whichcollects data to be analyzed to determine the best practices and tomake process improvements. Carbonari et al. (2009) proposed asafety management system for tracking workers’ trajectories to pre-vent accidents. Cho et al. (2010) discussed error modeling for anuntethered UWB system for construction indoor asset tracking. Onthe basis of their experiment, elevated tags give a better line-of-sight path between the tags and the sensors, and the average ac-curacy is 17 cm, whereas the tethered system gives 10 cm accuracyin open space. They have concluded that the accuracy seems sen-sitive mainly to the location and facing angle of sensors, whichaffect the chance of having a line-of-sight transmission path frommobile tags.

However, previous research did not investigate the requirementsof using UWB RTLSs for estimating the pose of cranes or otherconstruction equipment, which is necessary for accurate collisionavoidance. For example, tracking only the hook of a crane is notenough for collision avoidance. Raw location data of individualtags should be processed to produce the pose of a crane. Further-more, the setting of the sensor’s location and orientation and thenumber and location of tags should be investigated in detail toget more visibility of tags and less noisy object locations.

Methodology of Using UWB RTLS for Crane PoseEstimation

The present paper focuses on UWB data collection and processingto detect the poses of the crane boom in near real time. First, therequirements of using UWB technology in construction sites totrack crane movements are defined. Then, the details of the methodfor the UWB system setting are investigated to decide the locationsof sensors and the number and locations of tags attached to differentcomponents of a crane. Finally, a method for processing UWB datafor pose estimation is proposed.

UWB RTLS Requirements

On the basis of the literature review and experimentation with oneUWB system, the following requirements are identified to realizethe proposed approach: accuracy, visibility, scalability, real time,tag form factor, power, and networking requirements.1. Accuracy requirement. Accuracy is the most important re-

quirement to guarantee that valuable data are collected. Angleof arrival (AOA) and time difference of arrival (TDOA) can beused in UWB RTLSs to locate tags based on trilateration. Togain accurate location data, calibration of the sensors is essen-tial. Data filtering should be applied to reduce errors in nearreal time and to improve the accuracy.

2. Visibility requirement. The sensors should be set in a way toutilize their field of view (FOV) in both the azimuth and the

626 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

elevation. A reasonable monitoring area should be definedwithin these ranges considering the coverage of the sensors.If the area to cover is big, more sensors should be installedto form multiple cells. In addition, multiple tags attached tothe same object should be considered as a way to improvethe visibility of that object by increasing the probability ofdetecting these tags.

3. Scalability and real-time requirements. Because in some com-mercial UWB systems there is only a single UWB channelused in time division mode, only one tag can be located ata time in each sensor cell. For the system used in the presentresearch (Ubisense 2009), which has a nominal cell frequencyof R ¼ 160Hz, 1 s is divided into 153 time slots. Each slothas a duration of 6.5 ms. Different slot intervals can be se-lected in the system to determine how often the tags’ locationsare updated, i.e., how often the system listens for data andschedules messages from the master sensor. The shortest slotinterval can be set to four slots, which means the update inter-val is 26 ms, corresponding to a maximum update rate pertag of approximately 38 Hz with four or less tags in the cell(Ubisense 2009). With a large number of tags in a sensor cell,the update rate of tags will decrease to allow the system tocover all tags with the fixed total number of time slots. Forexample, if the time slot is set to four and only four tags arein the cell, the four tags are updated every 26 ms (38 Hz).When more tags are detected in the cell, e.g., eight tags,the update rate is decreased to 19 Hz. The more tags in thesystem, the bigger the required slot interval and the lowerthe update rate. Fig. 1 shows how the system assigns updatesfor four tags with a slot interval of four time slots. One con-sideration when setting the update rate is the moving velocityof the object. Objects with high velocity need more frequentupdates to accurately track their traces. Therefore, it is essen-tial to select a suitable number of tags with an appropriateupdate rate based on their velocity in order to achieve thebalance between the conflicting requirements of visibilityand accuracy in near real time as discussed in the subsectionNumber of Tags and Update Rate.

4. Tag form factor. Even if the basic functionality of the tags isthe same, tags come with different form factors. Some tags arespecifically designed to be worn by a person as a badge; othersare ruggedly designed to be attached to objects in a harshenvironment.

5. Power requirement. The sensors must be connected to a stablepower source for precision measurements. Tags require a bat-tery, the life of which depends on the update rate establishedfor the system. The tag’s update rate can be dynamically andautomatically varied depending on the activity of the tag. If thetag moves quickly, a high update rate can be assigned for besttracking; if it moves slowly, the update can be reduced for bestbattery lifetime. When stationary, a tag goes into sleep mode to

conserve power, and an in-built motion detector ensures thatthe tag transmits again as soon as it is moved.

6. Networking requirement. The sensors can be connected bycables or wirelessly to the location server. Both data cables andtiming cables are needed for a wired system. The length of thecables should not exceed the maximum length recommendedby the manufacturer to avoid noise problems (Ubisense 2009).As explained earlier, AOA and/or TDOA can be used in UWBRTLSs to locate tags. In some commercial UWB systems, onlyTDOA is available, which requires connecting the sensorswith timing cables to accurately measure the TDOA betweensensors. The AOA technique does not require cabling, and thedata communication between sensors can be done wirelessly.Therefore, the wireless UWB system depends only on AOAcalculations because wireless communication is not fast en-ough to support TDOA calculations. The choice of the type ofthe network (wired versus wireless) has a direct impact on ac-curacy (Cho et al. 2010). In some construction sites wherecables are not desirable (e.g., narrow work zones along high-ways), the wireless solution is preferable. However, the choiceof the type of the network (wired versus wireless) has a directimpact on accuracy (Cho et al. 2010).

There are only a few commercially available UWB systems, andeach of them has certain advantages and limitations. For example,the system used in the current research supports both TDOA andAOA but suffers from limited update rate that will affect the scal-ability of the system. Other systems have higher update rate butthey support TDOA only (Saidi et al. 2011), which may limit theirapplicability in tight construction sites where cables between sen-sors may disturb the construction work. The full comparison ofthese different systems is beyond the scope of the current paper.

UWB System Settings for Satisfying Requirements

To satisfy the previously discussed requirements, the UWB systemsetting method has been investigated in the present section to trackthe movement of a hydraulic crane boom on construction sites.

Sensor CoverageIn a typical setting, the four sensors of a cell are usually located atthe corners of a rectangular monitoring area at a high position fac-ing down toward the center of the area. In the case of monitoringthe movement of a large hydraulic crane, the sensors should befixed at a high position using tripods, and their pitch angle shouldbe adjusted to cover all the tags attached to the crane, as shown inthe upper set of sensors in Fig. 2. In this case, a second cell could benecessary to monitor workers working on the ground because tagsattached to them may not be detected by the upper cell because ofobstruction by the crane or the limited FOV of the sensors. Thistwo-cell setting to monitor the same area at two elevations is neededin sites where a large vertical coverage is needed. However, in other

Fig. 1. Tag updates for a 160-Hz system with slot interval four time slots

JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012 / 627

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

cases, using only one cell at the ground level is suitable. In thiscase, the pitch angle of the sensors should be set to cover allthe tags on site.

To specify rules for assuring sensor coverage for monitoring hy-draulic cranes, two extreme scenarios of lifting can be considered:(1) an extended long boom with a large angle to the ground,e.g., lifting objects from the ground to a destination with largeheight; and (2) an extended boom with a small angle to the ground,e.g., lifting with vertical spatial constraints. The latter scenario canbe illustrated in the case of truss bridge deck replacement projects,where prefabricated deck panels should be lifted from a trailer withspatial constraints from the truss structure (Hammad et al. 2007).To cover all the possibilities of the lifting tasks and onsite con-straints, this study consider a general case where the maximumlength of the boom and the maximum height of its tip can beadjusted as necessary to fit the actual case. Fig. 3 shows a two-dimensional (2D) projection of a sensor facing a crane. The sensorhaving a vertical FOVof �β is mounted on a tripod at a height Hswith a pitch angle of θ. The areas out of the FOVare not covered bythis sensor but can be covered by other sensors in the same cell. Theworking range of the crane should be considered to decide the ap-propriate position and the orientation of the sensors based on themaximum boom length and maximum tip height.

The horizontal distance between the sensor and the base of theboom L should satisfy the following condition:

L ≥ Hs · tan ð90° − β þ θÞ þ Lb (1)

where Lb can be simplified as the maximum boom length Lmaxwhen the boom is almost horizontal. The height of the sensorHs should meet the requirement of covering the height of the boomtip Hb:

Hs þ ðL − LbÞ= tan ð90° − β − θÞ ≥ Hb (2)

Hb is based on the maximum angle to the ground of the boom ac-cording to the working range of the crane. In an extreme scenario,Hb can be replaced by the maximum height of the boom Hmax. Onthe basis of these conditions, a set of L, Hs, and θ can be defined toimprove the coverage of the sensors. In practice, the system settingwill start by assuming the initial values for Hs and θ and the valueof Lwill be the larger of the two values calculated from inequalities(1) and (2). Furthermore, the size of a cell should satisfy the con-ditions of the maximum range of the UWB system and the length ofthe data and timing cables used for networking between the sensorsas explained in the networking requirement.

Simulation software (Autodesk Softimage 2010) is used to vis-ually check the sensor coverage with respect to tags attached to thecrane before the actual setting on site of the UWB system.

Sensor-1'

Sensor-2' Sensor-3'

Sensor-4'

Sensor-1

Sensor-2 Sensor-3

Sensor-4

Fig. 2. Vertical coverage of sensors at two elevations

90°-β

90°-β

β

β

Hs

L

Hb

Lb

θ

Ls

Field of view

Fig. 3. Parameters for defining sensor position and orientation

628 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

Number of Tags and Update RateTo meet the scalability and real-time requirements and to maximizethe update rate for the UWB system, one heuristic rule can be de-fined as follows based on the scheduling patterns of the system(Ubisense 2009):

m ≤ 4 × 2n (3)

where m = number of tags in the cell; n = minimum value thatmeets the inequality; and 4 × 2n = time slot interval that shouldbe set. For example, if there are 10 tags in the cell, the minimumvalue of n is 2; therefore, the time slot interval should be set to 16(i.e., the maximum update rate for each tag is 10 Hz).

If a specific update rate is required, another heuristic rule is de-fined as follows:

R=r ≥ 4 × 2n (4)

where R and r = update rates of the cell and the tags, respectively.For example, if an update rate of r ¼ 8 Hz is required for the tagsin a 160-Hz system, the maximum value of n is 2, and the time slotinterval can be set to 16. According to inequality (3), a maximum of16 tags can be used in the system to obtain this update rate. Similarinequalities can be derived for other UWB systems.

As previously described, r should be set according to the veloc-ity of the objects. For example, in the case of tracking a craneboom, if the velocity of the tip of the boom is 0.6 m=s, with aUWB system that has an accuracy of 15 cm, at least 4 Hz is neededto update the location of the boom’s tip to avoid potential collisions.

Location of TagsIn the case of monitoring the movement of a hydraulic crane, multi-ple tags should be attached to its different components to identifyits poses. Tags can be attached to the base of the first part of theboom and its tip for easy installation and to avoid damaging the

tags. Fig. 4 shows a schematic boom with three sets of tags(S1,S2,S3) attached to it. Each set Si includes four tags (Tagi1,Tagi2, Tag

i3, and Tagi4) fixed on each side of the boom. This re-

dundancy improves the visibility of the tags attached to the boomby the sensors when the boom rotates. The approximate location ofthe center point of a cross section Pi 0 can be calculated by averag-ing the locations of all or some of the four tags of set Si. The ori-entation and the length of the boom can be obtained by connectingthe two axis points P1 0

and P3 0. The purpose of having an addi-

tional set of tags S2 is to get a third point P2 0 on the axis of theboom so as to increase the accuracy by having more points alongthe axis, thereby allowing for the interpolation of the line represent-ing the axis. Fig. 4 also shows the pictures from the outdoor test,which is introduced in the subsection Removing Noisy Data andFilling in Missing Data.

Other SettingsThe power of the sensors is supplied by a power over Ethernet(POE) switch. Timing cables are used to connect the sensors tosynchronize the signals from a tag to different sensors. The sensorsare calibrated using a tag as a reference point with a known posi-tion. Compact tags are selected in the present research because oftheir omni-directional antennas and rugged design, which makethem more suitable for the tests with cranes. Compact tags are spe-cially designed for use in harsh industrial environments and includeseveral advanced features: an LED for easy identification and a mo-tion detector to instantly activate a stationary tag. The sleepingmode of the tags is disabled to ensure continuous monitoring.

UWB Data Processing Method for Pose Estimation

To improve data quality and compute the pose of an object in nearreal time, the following steps have been applied for data processing,as shown in Fig. 5:

S3

Axis of boom

S1

S2

Tagi1

Tagi2

Tagi3

Tagi4

Pi

P1

P2

P3 P3’

P2’

P1’

Tag22

Tag21

Tag24

(a) (b)

Fig. 4. Locations of tags on the boom and the cross section of the boom: (a) schematic representation of tags on a boom; (b) boom cross section withset of tags (Si)

JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012 / 629

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

1. The tags are identified by IDs and grouped according to theirgeometric relationship with respect to the objects they areattached to (e.g., tags attached to three sections of the boom).To approximately synchronize the locations of different tags, asmall time period T is defined according to the actual updaterate r of a tag. T should be big enough to capture at least onereading of each tag in the UWB cell and small enough for nearreal-time applications.

2. Readings of each tag are filtered within time tj according tosome heuristics, e.g., checking if the location of a tag attachedto an object is outside the expected area or outside the ex-pected height range. Another heuristic rule is checking the pat-terns of movement of the object. For example, assume that themaximum expected velocity vmax of an object is known andthat the measured velocity based on the distance between thepast captured location p

tj−1i of Tagi at time tj−1 and the new

one ptji is out of range. This condition indicates that p

tji has

an accuracy error and should be eliminated. Other heuristicrules can be applied based on the specific constraints of themovement of tags, such as the acceleration of movement.

3. Missing data for each tag are calculated using extrapolationaccording to the tag’s previous locations, assuming that theobject is moving with a known velocity.

4. If more than one location is captured for the same tag withinTk, these locations are averaged to obtain a single readingfor that tag. By doing this, data from different tags aresynchronized.

5. After synchronization, another filtering is applied according togeometric constraints of multiple tags attached to the same ob-ject. To illustrate the concept, Fig. 6 shows a simplified twodimensional example of the paths of two tags (Tagi and Tagi 0)attached to the same object. These paths are parallel with afixed distance Dii0 . The figure also shows the traces basedon the locations of tags at time Tk after averaging. It is noticedthat all points PTk

i and PTki0 have a certain number of accuracy

errors. However, the distances between the traces dTkii0 are

expected to be within the range of [Dii0 − 2ϵ,Dii0 þ 2 ϵ], whereϵ is the nominal accuracy of the UWB system (e.g., 30 cm). IfdTkii0 is outside this range, then PTk

i and/or PTki0 should be

checked for possible elimination. For example, in Fig. 6, ifdT5

ii0 is out of range compared with Dii0 , and PT5

i0 has beencalculated based on an extrapolated point, there is a higherprobability that PT5

i0 should be eliminated.6. Tag21After filtering, missing data can be calculated based on

extrapolation/interpolation of the data of other tags either inthe same group or in different groups. For example, inFig. 4(a), if the locations of the tags and Tag31 at the upperside of cross sections S2 and S3 are known at time tj, the loca-tion of Tag11 in cross section S1 can be calculated by extra-polation provided that the length of the boom does not change.

7. Locations of multiple tags in the same group are averaged(e.g., averaging the locations of the tags shown in Fig. 4(b)to get the center point of the cross section).

8. The pose of the object is calculated according to the positionsof the tags to which it is attached. For example, the pose of theboom can be found according to the calculated center pointson the axis of the boom. This pose is used for near real-timemotion replanning.

Case Study

Several lab, indoor, and outdoor tests have been applied using theproposed approach. One of the outdoor tests has been selected todemonstrate the applicability of the proposed approach. The othertests showed similar results. The test was done on the yard of acrane company on December 4, 2009, using a TMS300 crane(GUAY 2010). The UWB system has a norminal update rate of160 Hz. The test was designed in detail, including the sensors’ lo-cations, tags’ locations, cables’ connections, system calibration,data filtering, and task description. An information filter is used inthe system; details can be found in Ubisense (2009) manual. Fur-thermore, several indoor tests were applied in advance to test thestability of the UWB system, the influence of the magnetic mountsof the tags, etc. The following sections describe the settings of theoutdoor test following the discussion in the subsection UWB Sys-tem Settings for Satisfying Requirements.

Sensor Coverage

In this outdoor test, where the focus is on the crane, only four sen-sors at the ground elevation are deployed by adjusting the pitchangle to capture the boom movement, while satisfying the inequal-ities described in the following subsections. The FOVof the sensoris α ¼ � 90° in the azimuth and β ¼ � 50° in the elevation. Theyaw angles of the sensors were adjusted to face the center of thearea. The pitch angle and height of the sensors were approximatelyset to θ ¼ 20° and Hs ¼ 1.5 m, respectively. The maximum

(4) Synchronize data byaveraging data within Tk

(7) Average locationsof multiple tags in the

same group

(1) Identify tag IDs ondifferent components

(2) Filter readings foreach tag based on

heuristics

(8) Compute pose ofobject

(3) Calculate missingdata for each tag

using extrapolation

(5) Filter errors basedon geometric constraints

of multiple tags

(6) Calculate missingdata based on

geometric constraints

Fig. 5. Steps of data processing

Actual path of Tagi

Trace of Tagi

Trace of Tagi'

Actual path of Tagi'

Fig. 6. Detecting errors based on the distance between two tags in thesame group

630 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

boom length of the crane is 110 ft (33.5 m), and the minimum andmaximum angles to the ground of the boom are 10° and 80°,respectively. On the basis of the working range of the crane,when the boom is fully extended and reaches the highest point,the corresponding Hmax ¼ 120 ft (36.58 m), and Lb ¼20 ft (6.10 m). According to inequality (1), and taking Lb ¼110 ft (33.5 m), L should be greater than 36.1 m. According toinequality (2), to cover the maximum height of the boom tip,L should be greater than 18.86 m. By using the virtual crane modelin Softimage, it was possible to verify that the tag at the tip of theboom was within the FOV of the sensors.

At the site of the test, the actual pitch angles of the sensors werein the range of 20° to 26°, and the heights of the sensors were in therange of 1.45 to 1.67 m. Because of the limitations of the yarddimensions, which are approximately 18 by 22.5 m, the crane ispositioned in a way to make L equal approximately 21 m. However,this does not satisfy inequality (1). Therefore, it can be found usingthe same inequalities that the maxim length of the boom and theheight of its tip should be approximately 18 and 10 m, respectively.On that basis, the operator of the crane was asked to limit the ex-tension of the boom and the angle to the ground to stay within thoselimits. Fig. 7 shows the setting of the sensor cell for this test withthe timing and data cables and the locations and yaw angles of thefour sensors. The cables between the master and slave sensors areoutdoor industrial cables that are set to form a U shape to leave anopening for equipment to enter the monitored area and to reduce thedisruption of the work. It should be noted that in actual constructionsites, the cables should be laid near the peripherals of the sitefollowing safety regulations. A car was positioned as an obstacleon the moving path of the lift.

Tag Settings

Twenty-two tags were attached to the crane body, with three sets oftags (12 tags) attached to the boom, as proposed in Fig. 4. Othertags were attached to the outriggers, operator cab, hook, and liftedobject. Moreover, four tags were attached to the hard hats of twoworkers (two tags on each hard hat) to track their movements onsite. Fig. 8 shows the tags attached to different objects.

To test the scalability of the UWB system, which has a high cellupdate rate of 160 Hz, the writers kept 48 additional tags in thesame area so that the total number of tags in the cell was 74.According to the inequality (3) introduced in the subsection Num-ber of Tags and Update Rate, the time slot interval should be set to128, where the update rate is 1.2 Hz for each tag according to in-equality (4) By observing the collected data, it was found that theactual update rate was approximately 2 Hz. Therefore, in this test,the syncronization of multiple tags was based on T ¼ 500 ms. Aninformation data filter provided by the UWB system was usedto improve the accuracy with a motion model of position andGaussian noise on position (Ubisense 2009).

The total duration for the outdoor test was approximately 2hours, including the system configuration, measurement, calibra-tion, moving the crane into the monitored area, and collecting dataduring the crane operation. The task given to the crane operator wasto lift an object from one place to another by swinging and raisingup the boom while avoiding the collision between the object andthe car (Fig. 9).

During the lifting, the length of the boom and the length of thecable were fixed. A part of the raw data collected in the test isshown as traces in Fig. 9(b). The tags shown in three cross sectionsare Tag14, Tag

21, Tag

23, Tag

31, and Tag34. The data analysis of this

test is discussed in the next section.

Data Analysis

The tags attached to the upper side of the boom had very goodvisibility and less noisy data compared with those attached tothe bottom and the sides of the boom, which had a large numberof missing points and noisy data. The raw UWB data were proc-essed following the steps explained in Fig. 5 to get the poses of theboom. However, because of the low update rate (2 Hz) and the largeamount of missing data, some steps were not always applicable(e.g., averaging or extrapolation at a certain time period). Never-theless, the redundancy provided by having multiple tags on theboom made it possible to calculate the poses of the boom basedon the traces as shown later in this paper.

Timing Cable

(10.16, 12.1, 1.57)

Sensor-4 Sensor -1 (Master)

Sensor-2 Sensor-3

Switch

Reference point 1

Referencepoint 2

X

Y

Data CableSensor with orientation (yaw)

(-8.52, -10.54, 1.61) (9.07, -10.31, 1.45)

(-9.36, 12.47, 1.67)

Fig. 7. Site layout

(a)

(c) (d)

(b)

Fig. 8. Tags attached to different objects

JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012 / 631

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

Visibility Analysis

As mentioned before, 22 tags were attached to the crane and fourtags were attached to the hard hats of workers. Within the recordingtime of 36 s, which was the duration of the lifting task, location dataof all the tags were captured with different update rates, as shown inTable 1. This indicates that the coverage of the sensors is good, withthe settings following the equations proposed in the subsectionSensor Coverage. The measured update rate r0 is calculated by

dividing the number of updates of each tag by 36 s. Because ofmissing data, some tags have lower update rates than other tags.From this table, it can be seen that tags on the upper and bottomsides of the boom had better visibility than the ones attached to theside surfaces. Tags attached to the side with a truss structure(i.e., Tag14, Tag

24, and Tag

34) received fewer updates compared with

other tags in the same cross section. This could be explained byradio signal reflections on the truss, as shown in Fig. 4. Tags at-tached to the cab also showed bad visibility because the rotationof the cab prevented the direct line-of-sight from two sensors.All the four tags attached to the hook had excellent visibility. Tagsattached to the lift object had bad visibility that may be explainedby the lack of direct line-of-sight. One tag attached to the leftoutrigger had good visibility.

Accuracy Analysis

The accuracy used in this study is defined as the difference in dis-tance between a known position and position captured by the UWBsystem. The location of two static tags on the two outriggers wereanalyzed to reveal the accuracy of the system based on the mea-sured locations of these tags. These locations were measured witha tape using five measurements with averaging. The differencesbetween the average and the individual measurements were lessthan 5 cm. Table 2 shows the mean difference and the standarddeviations in three directions of these data. The accuracy of the datacollected for these two tags is approximately 25 cm. The tag on theleft outrigger has more readings than the tag on the right outrigger(74 versus 34 readings, as shown in Table 1), thereby contributingto the more accurate results of the left tag.

Removing Noisy Data and Filling in Missing Data

Example of Filtering Readings of Tags Based on HeuristicsOn the basis of the steps defined in Fig. 5, errors have been iden-tified and eliminated in near real time. After identifying tag IDs ondifferent crane components, the heuristic of the maximum expected

Traces of tags on hook

Right outrigger

First boom section

Boom tip

Traces of tags of S2

Traces of tags of S3

Left outrigger

Traces of tags of S1

(a) (b) (c)

Fig. 9. Part of the raw data collected: (a) initial crane pose; (b) crane poses based on raw data; (c) final crane pose

Table 2. Mean Difference and Standard Deviation in Three Directions for Static Tags

Tag nameMean difference

in X ðmÞMean difference

in Y ðmÞMean difference

in Z ðmÞStandard deviation

in X ðmÞStandard deviation

in Y ðmÞStandard deviation

in Z ðmÞOl −0.129 −0.089 −0.2 0.2084 0.1976 0.2787Or −0.212 −0.085 0.248 0.1917 0.1807 0.206

Table 1. Tag Updates

Tag locationTagname

Number of updatesin 36 s

Measured updaterate r0 (Hz)

Boom S1 Tag11 61 1.7Tag12 37 1Tag13 73 2Tag14 24 0.7

S2 Tag21 74 2.1Tag22 50 1.4Tag23 70 1.9Tag24 24 0.7

S3 Tag31 74 2.1Tag32 18 0.5Tag33 42 1.2Tag34 20 0.6

Cab C1 12 0.3C2 39 1.1

Hook H1 74 2.1H2 74 2.1H3 73 2H4 73 2

Lift L1 27 0.8L2 20 0.6

Outrigger Right Or 34 0.9Left Ol 74 2.1

Hard hat-1 H1r 50 1.4

H1l 24 0.7

Hard hat-2 H2r 72 2

H2l 61 1.7

632 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

velocity vmax can be set for specific tags. On the basis of the ourobservation, the average velocity of tags in cross section S2 of theboom is approximately 0.5 m=s. By adding the UWB system error,which is approximately �30 cm in all readings, vmax used to filterthe UWB readings in near real time for tags in S2 is set to 1.5 m=s.Taking Tag23 as an example, there is a sudden movement in the Zdirection at t24, as shown in Table 3 and Fig. 10, and the velocity ofTag23 is calculated as 4.53 m=s, which by far exceeds vmax . There-fore, the reading at t24 is rejected and replaced by a location calcu-lated based on extrapolation according to the Δ value in eachdimension (X, Y, and Z). The purpose of calculating the differencein each dimension individually is that the accuracies in these threedimensions are different, and based on these observations and thoseof Muthukrishnan and Hazas (2009), the accuracy in the Z dimen-sion is lower than those in the X and Y dimensions. The averageΔvalue (μΔ) and the standard deviation (σΔ) are calculated accord-ing to previous data history during the last 5 s. Only points with aΔin any of the X, Y, Z dimensions that are out of the range of[μΔ − 2σΔ; μΔþ 2σΔ] are corrected in those specific dimen-sions using extrapolation from two previous points. This range con-tains 95.44% of the data assuming that the differences follow anormal distribution (Allen 2006). As shown in Table 3, the Δ val-ues in the Y and Z dimensions are out of range at t24, whereΔy ¼ 0.21 m, Δz ¼ 2.13 m, and out of the ranges of [−0.073 m,0.059 m] and [−0.184 m, 0.220 m], respectively, where μΔy ¼−0.007 m, μΔz ¼ 0.018 m, σΔy ¼ 0.033 m, σΔz ¼ 0.101 m(these values are at t23). Extrapolation is done based on the locationdata at t22 and t23 for those two dimensions (Y and Z). It should benoted that the information filter used for all the tags in the Ubisensesystem always predicts location data based on previous readings;therefore, the data collected for the next time periods (from t25 tot31) are all affected by the prediction based on errors, and they haveto be recalculated by extrapolation similar to the point at t24 toavoid exceeding vmax . This extrapolation results in creating newdata as shown in Table 3. The results are shown in Fig. 10, wherethe raw data and the processed data are plotted. The big jump in theZ dimension is eliminated. It should be clarified that by chance themovement of Tag23 during the period between t24 and t32 is almostparallel to the X axis and to the X − Y plane, as can be seen inFig. 15; therefore, after correction, the Δy and Δz values are closeto 0. A flowchart is shown in Fig. 11 to summarize the near real-time data processing for single tags.

Example of Calculating Missing Data Based on GeometricConstraintsThe same procedure is applied to Tag21 as shown in Fig. 12. How-ever, in some cases, missing data occur more than two consecutivetimes because of radio interference, for example, between t41 andt57, as shown in Fig. 12. In these cases, repeating extrapolation ac-cording to the history of the tag itself may increase the error, whichcould be detected by checking geometric constraints. As describedin Step 5 in the subsection UWB Data Processing Method for PoseEstimation, multiple tags are used to filter errors and fill in the miss-ing data based on geometric constraints of the object. The distancebetween Tag21 and Tag

23 in each time period t is calculated to check

if it is within the range of [Dii0 − 2ϵ, Dii0 þ 2 ϵ], where Dii0 is 1.6 mand ϵ is 30 cm, resulting in a range of [1.0 m, 2.2 m]. This step hasbeen applied starting from t42, where extrapolation is appliedfour times in a row to fill in the missing data of Tag21. However,at t46, the distance between Tag21 and Tag23 is 2.44 m, which is outof range. Therefore, the location of Tag21 calculated according toextrapolation is not acceptable. In this case, according to Step 6, thedata of Tag31 and Tag11 are used to calculate the missing data ofTag21 between t46 and t57 based on the known distances from Tag11T

able

3.DataProcessedforTag

32in

RealTim

e

Raw

data

ofTag

32

Processeddata

ofTag

32

Tim

ex(m

)y (m)

z(m

)v

(m=s)

ΔAverage

Δðμ

ΔÞ

Std.

deviation(σ

Δ)

Tim

ex_n

(m)

y_n

(m)

z_n

(m)

v(m

/s)

ΔAverage

Δðμ

ΔÞ

Std.

deviation(σ

Δ)

ΔX

ΔY

ΔZ

ΔX

ΔY

ΔZ

ΔX

ΔY

ΔZ

ΔX

ΔY

ΔZ

ΔX

ΔY

ΔZ

ΔX

ΔY

ΔZ

t 23

0.400

5.4

3.750

0.047

0.02

0.000

0.010

0.082

−0.01

0.018

0.170

0.033

0.101

t 23

0.400

5.430

3.750

0.047

0.020

0.000

0.010

0.153

−0.05

0.063

0.250

0.033

0.101

t 24

0.700

5.6

5.880

4.531

0.3

0.210

2.130

0.085

−0.02

0.035

0.180

0.080

0.669

t 24

0.700

5.430

3.760

0.629

0.300

0.000

0.010

0.153

−0.02

0.035

0.180

0.033

0.100

t 25

0.710

5.6

5.870

0.030

0.01

0.000

−0.010

0.115

0.002

0.248

0.183

0.079

0.673

t 25

0.710

5.430

3.770

0.030

0.010

0.000

0.010

0.115

−0.02

0.036

0.183

0.034

0.010

t 26

1.040

5.2

5.630

1.228

0.330−0

.420

−0.240

0.104

0.005

0.215

0.193

0.156

0.685

t 26

1.040

5.430

3.780

0.694

0.330

0.000

0.010

0.104

−0.02

0.005

0.193

0.034

0.008

t 27

1.060

5.2

5.600

0.078

0.020

0.010

−0.030

0.135

−0.04

0.192

0.193

0.157

0.686

t 27

1.060

5.430

3.790

0.047

0.020

0.000

0.010

0.135

−0.02

0.007

0.193

0.034

0.006

t 28

1.280

5.4

5.150

1.096

0.220

0.150

−0.450

0.135

−0.03

0.19

0.192

0.167

0.714

t 28

1.280

5.430

3.800

0.462

0.220

0.000

0.010

0.135

−0.01

0.009

0.192

0.034

0.006

t 29

1.290

5.4

5.140

0.030

0.010

0.000

−0.010

0.153

−0.02

0.144

0.194

0.167

0.714

t 29

1.290

5.380

3.810

0.109

0.010−0

.05

0.010

0.153

−0.01

0.009

0.194

0.036

0.005

t 30

1.320

5.4

5.100

0.107

0.030−0

.01

−0.040

0.149

−0.02

0.143

0.132

0.164

0.716

t 30

1.320

5.370

3.820

0.070

0.030−0

.01

0.010

0.149

−0.02

0.010

0.132

0.016

0.003

t 31

1.690

5.2

4.460

1.595

0.370−0

.18

−0.640

0.096

−0.01

0.137

0.153

0.173

0.757

t 31

1.690

5.360

3.830

0.776

0.370−0

.01

0.010

0.096

−0.01

0.009

0.153

0.016

0.000

t 32

1.690

5.2

4.450

0.021

0.000

0.000

−0.010

0.133

−0.02

0.073

0.154

0.173

0.757

t 32

1.690

5.190

4.450

1.348

0.000−0

.17

0.620

0.133

−0.01

0.010

0.154

0.054

0.193

t 33

1.710

5.2

4.430

0.063

0.020

−0.01

−0.020

0.131

−0.02

0.071

0.154

0.173

0.757

t 33

1.710

5.180

4.430

0.063

0.020−0

.01−0

.02

0.131

−0.02

0.071

0.154

0.053

0.194

JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012 / 633

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

to Tag21 and from Tag21 to Tag31 (3.9 m and 8.4 m, respectively), asshown in Fig. 15. Fig. 12 shows the extrapolation based on the his-tory of Tag21 from t42 to t44 and the interpolation based on geometry

according to the other two tags (Tag31 and Tag11) from t46 to t56.From t57 the system is able to capture the data for Tag21 again.

The data of all tags are assumed to be almost synchronized (Step4 in Fig. 5). However, it should be noticed that in extrapolationbased on geometry using two tags on the boom, the small time gapsbetween different tags can cause problems when the update rate oftags is not high enough. For example, in this test, for tags attachedto the upper side of the boom, based on the automatic scheduling ofthe UWB system in each time period t, the data were captured in theorder of Tag31, Tag

11, Tag

21,with fixed time differences of 119 and

74 ms, respectively.As shown in Fig. 13(a), a point with a large error was captured for

Tag11 at time tþ 119 ms; therefore, extrapolation based onTag21 andTag31 is applied to calculate thepositionofTag

11. Theblack circles are

the location data captured by the system, whereas the solid whitecircles are the real locations of the tags at specific times, and the dot-ted circles are the ones calculated according to extrapolation. Thisfigure also shows the traces of Tag11, Tag

21, and Tag

31 and the boom

poses basedonextrapolation as explained above.Notice that thewrit-ers are ignoring the accuracy errors for Tag31, and Tag

21 in Fig. 13(a),

and only for Tag31 in Fig. 13(b). However, because of the smalltime gap and the relatively big distance between these three tags (ap-proximately 12.3 m between Tag11 and Tag31, as shown in Fig. 15)during the lifting task, a big offset of the location of Tag11 is expectedwhen applying extrapolation. Moreover, because of the static infor-mation filter of the Ubisense system used in this test (with Gaussiannoise on position), small movements of a tag are ignored when pre-dicting the next location of the tag. This filtering results in a cluster ofalmost overlapping points. The use of these data for extrapolationmay cause a backward movement of Tag11, as shown in Fig. 13(b).

As an illustration of this problem, the trace for Tag11 is shown inFig. 14, which gives the data processed in near real time. Fig. 14(b)focuses on the zigzag shape of the trace and the crossing of theboom poses at times t5 and t10 and at times t20 and t25. On the basisof this observation, the continuous extrapolation for Tag11 based onthe other two tags may increase errors.

Calculating the Poses of the Boom

As described in Step 7 in the subsection UWB Data ProcessingMethod for Pose Estimation, averaging the data of multiple tagsin the same cross section should be applied to get the center pointsof these sections, thereby defining the axis of the boom. A bound-ing shape (e.g., a cylinder) to cover these three points at each timeperiod can be created with a suitable buffer according to the crosssection dimensions of the boom.

Eliminated errors

t23

t24

t32

t74t2

Fig. 10. Comparison of traces of Tag23 in X − Z plane before and after correction

Read location data for tj

Velocity withinrange?

Yes

Calculate x, y, and z

No

within range?

Extrapolate and replacelocation data for thatspecific dimension

No

Yes

Start

Test finished?

No

Finish

Yes

j = j + 1

j = 1

Fig. 11. Flowchart of near real-time data processing for single tags

t41

t45

t46

t57t74

t2

Fig. 12. Trace of Tag21 based on extrapolation of its history andinterpolation of other two tags

634 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

This method assumes that the quality of the data of each tag isequal; however, based on the actual collected data, the method ofcalculating the poses of the boom should be adapted so as to pre-serve the data of high quality. On the basis of these observation,tags on the top side of the boom have better quality; therefore,the traces of these tags (Tagi1) are used to create the poses of theboom. As shown in Fig. 15, the three traces show the poses ofthe boom at different times.

Discussion and Recommendations

The proposed approach and the results of the case study have beenpresented to engineers and safety experts from two crane compa-nies and the Commission of Work Health and Safety of Quebec(CSST). These engineers and experts provided a positive evaluationof the applicability of the proposed methods in practice.

On the basis of these observations of the UWB tests, the writerssuggest the following recommendations for future research: (1) thenumber of tags in the monitored area should be kept smaller than

the maximum number calculated using the two heuristic rules[inequalities (3) and (4) in the subsection of Number of Tagsand Update Rate]; otherwise, the update rate of each tag will de-crease. To meet the update rate requirement, more cells should beused by dividing the monitored area into smaller areas sensed bydifferent groups of sensors. In this way, the UWB system is capableof reading the tags with the required update rates; (2) regarding thevisibility requirement, tags should be attached to the upper and bot-tom sides of the boom to obtain a better visibility and better dataquality. Attaching one tag to the hook is enough. More tags shouldbe attached to the lift object to avoid obstruction of radio signals,and it is better to attach the tags to the top surface of the lift object.Tags should be attached to the top of the operator’s cab to achievebetter visibility; and (3) The proposed approach of estimating craneboom poses can be used for collision avoidance. Potential colli-sions can be avoided using the pose information, and the path ofthe crane can be replanned in a safe way (AlBahnassi and Hammad2012). Buffers should be added to the potential obstacles for col-lision detection. The size of the buffer can be adjusted according to

OffsetLarge error of

At time t +119 ms+ 74 ms

(a)

At time t

At time (t+1)+119 ms

(b)

Large error of

At time (t+1)

At time (t+1)+119 ms + 74 ms

Offset

At time t +119

Captured pointReal pointExtrapolated pointReal poseExtrapolated poseTags’ traces

Fig. 13. Conceptual figure of extrapolation errors: (a) data at time period t; (b) data at time period tþ 1

t5

t15

t10

t20

t25

(a) (b)

Fig. 14. Data processed in real time showing the traces of three tags at different times

JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012 / 635

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

the expected error and the update frequency of the UWB systemand to the moving velocity of the crane. Less accurate data, lowerupdate frequency, and higher velocity require selecting a biggerbuffer around obstacles.

Conclusions and Future Work

This paper discussed using UWB RTLS in construction sites formonitoring and estimating the poses of cranes. The requirementsand system settings for the UWB RTLS are discussed in detail.An outdoor test was applied on a cranewith tags attached to differentcomponents. The results of the test showed a good potential to usetheUWBsystem on construction sites for crane pose estimation. Theconclusions of the present paper are: (1) the requirements of apply-ing UWB RTLS on construction sites have been defined; (2) amethod for the UWB system setting to has been investigated to meetthese requirements for estimating cane poses. Heuristic rules havebeen defined to maximize the sensor coverage and to calculate thenumber of tags used in a sensor cell with the required update rate;(3) a location data processing method has been proposed to improvedata quality by filtering noisy data and filling in missing data in nearreal time; and (4) testing of the UWB system using the proposedmethods has been carried out, which has demonstrated that the poseof the crane boom can be estimated in near real time.

The limitation of the proposed data processing method is thatlinear interpolation and extrapolation are used based on two pointsonly. Further improvement of data processing could be carried outby using curve fitting or other methods based on several previouspoints. In addition, Kalman filtering combined with geometricconstraints (Arras et al. 2003) could be investigated in the futureto improve the accuracy of UWB data. Furthermore, research isneeded to investigate the cost-benefit effectiveness and the practicalissues for deploying UWB technology in construction projects.

Acknowledgments

This research project is supported by a grant from the Institut derecherche Robert-Sauvé en santé et en sécurité du travail (IRSST).

We would like to thank the members of the follow-up committeeof the project, Mr. Jean-Louis Lapointe and Mr. Rafael Palomar,from GUAY Crane Company for their help in providing the craneand operator to achieve the test. The help of Junyu Jan, Yu Sato,Homam Al-Bahnasi, and Ali Motamedi in realizing the test isappreciated.

References

AlBahnassi, H., and Hammad, A. (2012). “Near real-time motion planningand simulation of cranes in construction: Framework and system archi-tecture.” J. Comput. Civ. Eng., 26(1), 54–63.

Allen, T. T. (2006). Introduction to engineering statistics and six sigma:Statistical quality control and design of experiments and systems,Springer, London.

Arras, K. O., Castellanos, J. A., Schilt, M., and Siegwart, R. (2003).“Feature-based multi-hypothesis localization and tracking usinggeometric constraints.” Rob. Auton. Syst., 44(1), 41–53.

Autodesk Softimage. (2010). ⟨http://usa.autodesk.com/adsk/servlet/pc/index?id=13571168&siteID=123112⟩ (Jun. 2010).

Beavers, J. E., Moore, J. R., Rinehart, R., and Schriver, W. R. (2006).“Crane-related fatalities in the construction industry.” J. Constr. Eng.Manage., 132(9), 901–910.

Bureau of Labor Statistics. (2008). “Crane-related occupational fatalities.”⟨www.bls.gov/iif/oshwc/osh/os/osh_crane_2006.pdf⟩ (Jun. 25, 2010).

Carbonari, A., Naticchia, B., Giretti, A., and De Grassi, M. (2009). “A pro-active system for real-time safety management in construction sites.”Proc., 26th Int. Symp. on Automation and Robotics in Construction,Austin, TX.

Chae, S., and Yoshida, T. (2008). “A study of safety management usingworking area information on construction site.” Proc., 25th Int.Symp. on Automation and Robotics in Construction, VilniusGediminas Technical University Publishing House, Vilnius, Lithuania,292–299.

Cho, Y. K., Youn, J. H., and Martinez, D. (2010). “Error modeling for anuntethered ultra-wideband system for construction indoor assettracking.” Autom. Constr., 19(1), 43–54.

Commission de la santé et de la sécurité du travail du Québec (CSST).(2010). “Commission de la santé et de la sécurité du travail duQuébec.” ⟨www.csst.qc.ca/portail/fr/⟩ (Jun. 25, 2010).

Fig. 15. Boom poses at different time periods

636 / JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.

Ghavami, M., Michael, L. B., and Kohno, R. (2004). “Ultra widebandsignals and systems in communication engineering.”Wiley, Chichester,UK.

Giretti, A., Carbonari, A., Naticchia, B., and De Grassi, M. (2009).“Design and first development of an automated real-time safetymanagement system for construction sites.” J. Civil Eng. Manage.,15(4), 325–336.

GUAY. (2010). ⟨www.gruesguay.com/en/index.html⟩ (Jan. 2010).Hammad, A., Zhang, C., Al-Hussein, M., and Cardinal, G. (2007). “Equip-

ment workspace analysis in infrastructure projects.” Can. J. Civ. Eng.,34(10), 1247–1256.

Muthukrishnan, K., and Hazas, M. (2009). “Position estimation from UWBpseudorange and angle-of-arrival: A comparison of non-linear regres-sion and Kalman filtering.” Proc., 4th Int. Symp. on Location andContext Awareness, Springer, 222–239.

National Institute for Occupational Safety and Health (NIOSH). (2007).“Goal 1: Reduce the major risks associated with traumatic injuriesand fatalities in construction.”⟨www.cdc.gov/niosh/nas/construction⟩(Dec. 2009).

Navon, R., Goldschmidt, E., and Shpatnisky, Y. (2004). “A concept provingprototype of automated earthmoving control.” Autom. Constr., 13(2),225–239.

Orbit Communications. (2008). “Vehicle proximity alert and collisionavoidance system from orbit communications.” ⟨www.ferret.com.au/c/Orbit-Communications/Vehicle-Proximity-Alert-and-Collision-Avoidance-System-from-Orbit-Communications-n778105⟩ (Jun. 25, 2010).

Saidi, K., and Lytle, A. (2008). “NIST research in crane automation: 2007overview.” 87th Annual Meeting, Transportation Research Board.

Saidi, K. S., Teizer, J., Franaszek, M., and Lytle, A. M. (2011). “Static anddynamic performance evaluation of a commercially-available ultrawideband tracking system.” Autom. Constr., 20(5), 519–530.

Task Committee on Crane Safety on Construction Sites. (1998). “Cranesafety on construction sites.”Manual of practice No. 93, ASCE, Reston,VA.

Teizer, J., Allread, B. S., Fullerton, C. E., and Hinze, J. (2010).“Autonomous pro-active real-time construction worker and equipmentoperator proximity safety alert system.” Autom. Constr., 19(5),630–640.

Teizer, J., Venugopal, M., and Walia, A. (2008). “Ultrawideband for auto-mated real-time three-dimensional location sensing for workforce,equipment, and material positioning and tracking.” Transportation Re-search Record, Vol. 2081, Transportation Research Board, Washington,DC, 56–64.

Trimble GCS900. (2010). “Trimble GCS900 grade control system.”⟨www.trimble.com/GCS900.html⟩ (Jun. 25, 2010).

Ubisense. (2009). Ubisense location engine configuration user manual,Ubisense, Cambridge, UK.

Ward, A. M. R. (2007). In-building location systems, The Institution ofEngineering and Technology Seminar on Location Technologies,London, UK, 1–18.

WorkSafeBC. (2010). ⟨http://worksafebc.com⟩ (Jun. 25, 2010).Zhang, C., AlBahnassi, H., and Hammad, A. (2010). “Improving construc-

tion safety through real-time motion planning of cranes.”Int. Conf. onComputing in Civil and Building Engineering, Univ. of Nottingham,Nottingham, UK.

Zhang, C., Hammad, A., and AlBahnassi, H. (2009a). “Collaborative multi-agent systems for construction equipment based on real-time field datacapturing.” ITcon, 14, 204–228.

Zhang, C., Hammad, A., and AlBahnassi, H. (2009b). “Path re-planning ofcranes using real-time location system.” Int. Symp. on Automation &Robotics in Construction, ISARC, Austin, TX.

JOURNAL OF COMPUTING IN CIVIL ENGINEERING © ASCE / SEPTEMBER/OCTOBER 2012 / 637

J. Comput. Civ. Eng. 2012.26:625-637.

Dow

nloa

ded

from

asc

elib

rary

.org

by

Con

cord

ia U

nive

rsity

Lib

rari

es o

n 11

/01/

12. C

opyr

ight

ASC

E. F

or p

erso

nal u

se o

nly;

all

righ

ts r

eser

ved.