1
THREE DIMENSIONAL URBAN BUILDING DETECTION USING LiDAR DATA Indu Indira Bai, Dr. Rama Rao Nidamanuri Cochin University of Science and Technology, Indian Institute of Space Science & Technology BACKGROUND The 3D city modelling is very useful and applicable in contemporary urban planning. Airborne laser scanners could not be able to capture side view of the buildings. Terrestrial laser scanners can precisely give the side view of the buildings. There is a need to have an open source software approach for processing and 3D visualisation of Terrestrial Data. OBJECTIVES Proposes a methodology for simple 3D city modelling using TLS data and an open source platform(PCL) Implementation of Colour Based Region Growing Algorithm Object or feature detection of the point cloud data. The combination of geometrical and colour information leads to meaningful and effective segmentation results, Work Flow For The Data Set Point cloud from terrestrial laser scanner Clustering based on color based region growing segmentation algorithm Visualization and building object detection Processing of point cloud : segmentation Figure 1: Representation of the Objective Figure 2: 3D View of the Original Data Figure 3: Figure 4: pcd files in the cloud compare viewer METHODS RESULTS SUMMARY ACKNOWLEDGEMENT CONTACT INFORMATION BIBLIOGRAPHY The data set is in las format. The sensor used for obtaining this data was Terrestrial laser scanner RIEGL VZ-400.The method of segmentation was done using the point cloud library (PCL), an open source software library licensed under Berkeley Software Distribution (BSD). Convert LAS to txt format by LAS tools Note the range of R,G,B in the txt and the cloud width Clustering of the point cloud to form .pcd files Visualization of .txt files Implement color based region growing segmentation algorithm Segmentation based on both spatial and spectral homogeniety XYZRGB(txt) format is converted to .pcd files. Clusters are formed Validation Clusters are converted to txt format by C++ code Color Based Region Growing Segmentation Algorithm Color is taken as the discriminator and Kd tree Search is used here. The merging of the algorithm prevents over and under segmentation. The algorithm attempts to merge clusters with close colours. Two neighbouring clusters with a small difference between average colours are merged together. After that the next merging step takes place.Every single cluster is verified by the number of points that it contains. If this number is less than the user-defined value then the current cluster is merged with the closest neighbouring cluster. The procedure iterates and is terminated when all the points are labelled. Here segmentation is done based on both geometrical and color information. GSN Perera,N Hetti Arachchige “Efficient 3D City Modelling from Airborne Laser Scanning point clouds”, Proceedings of 8th International Research Conference, KDU, Published November 2015 Rusu, R.B, Cousins, S; "3D is here: Point Cloud Library (PCL)," Robotics and Automation (ICRA), 2011 IEEE International Conference on , vol., no., pp.1,4, 9-13 May 2011 Shaohui Sun and Carl Salvaggio; “Aerial 3D Building Detection and Modelling From Airborne LiDAR Point Clouds” IEEE Journal Of Selected Topics In Applied Earth Observations And Remote Sensing, Vol. 6, No. 3,June 2013,Pp1440- 1449 Douillard, B., Underwood, J, Kuntz, N., Vlaskine, V, Quadros, A, Morton, P, Frenkel, A; "On the segmentation of 3D LIDAR point clouds," Robotics and Automation (ICRA), 2011 IEEE International Conference on , vol., no., pp.2798-2805, May 2011; Ying LIU, Ruofei Zhong; “Buildings and Terrain of Urban Area Point Cloud Segmentation based on PCL”, 35th International Symposium on Remote Sensing of Environment (ISRSE35) IOP Publishing, IOP Conf. Series: Earth and Environmental Science 17 (2014) Parameters changed by iterative method are the following Distance threshold Point colour spacing Region colour threshold Minimum cluster size In order to obtain effective segmentation and better results some of the parameters in the C++ code were changed by iterative method Figure 5: Segmented result for the minimum cluster size 500 Figure 6: Segmented result for the minimum cluster size 600 Figure 8: Segmented result for the minimum cluster size 100 Figure 7: Figure 9: Segmented result for the minimum cluster size 150 Segmented result for the minimum cluster size 75 Table 1: Iterative results showing the variation of numbers of clusters with minimum cluster size Table 2: Variation of building features according to the cluster size Figure 10: Segmented result for the minimum cluster size 50 Threshold parameters are adjusted. From the processing, better results were obtained by keeping the minimum cluster size between 100 and 50. Other three parameters are kept in fixed values. Features become completely visible for the minimum cluster size values 50 and 75.The color based segmentation become more effective for the threshold values 50 and 75. Reducing the minimum cluster size, features become completely visible. Main parameter here is colour, affecting the segmentation. The integration of geometrical and colour information lead to more meaningful and useful information. There is a balance between over and under segmentation. The iterative method made it possible to find the key parameters. These tasks are extremely laborious when carried out manually. Helps in detecting objects or features of interest. We have to test the key parameters each time. There are no automatic methods for finding the suitable values of key parameter for effective segmentation. Key parameters are to be obtained by iteration method. Further research can be done on the inter dependance of the parameters. Terrestrial laser scanner and the colour based region growing segmentation method enable the building detection and urban applications more effective. Minimum cluster size Observed classes 600 500 150 100 75 50 Only very few walls, one pole, roofs More number of walls, roofs, 2 boards, 2 poles ,tree Walls, roofs, windows , tree, poles, boards Almost all walls and roofs are separately visualized, tree, poles and the boards Almost every parts including the air conditioner are visible Clear visualization of every part including walls, roofs, windows, boards, poles, air conditioner. Sincere thanks to Kerala State Council for Science and Technology, Cochin University of Science & Technology (CUSAT), Dr M.V. Harindranathan Nair, Associate Professor, School of Environmental Studies , CUSAT and Dr.Jaishanker.R.Nair, Associate Professor, Indian Institute of Information Technology & Management Kerala 1. Mrs Indu Indira Bai, Research Scholar, School of Environmental Studies, Cochin University of Science and Technology, Kochi, Kerala, India. Mob-+91 9895160999, Email- [email protected] 2. Dr Rama Rao Nidamanuri, Associate Professor, Department of Earth and Space Sciences, Indian Institute of Space Science and Technology, Valiamala, Trivandrum, Kerala, India Mob- +91 8129285705, Email – [email protected] Segmented result by changing the threshold values ASPRS - IGTF 2016

Indu Indira Bai, Dr. Rama Rao Nidamanuri...1. Mrs Indu Indira Bai, Research Scholar, School of Environmental Studies, Cochin University of Science and Technology, Kochi, Kerala, India

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Indu Indira Bai, Dr. Rama Rao Nidamanuri...1. Mrs Indu Indira Bai, Research Scholar, School of Environmental Studies, Cochin University of Science and Technology, Kochi, Kerala, India

THREE DIMENSIONAL URBAN BUILDING DETECTION USING LiDAR DATA

Indu Indira Bai, Dr. Rama Rao NidamanuriCochin University of Science and Technology, Indian Institute of Space Science & Technology

BACKGROUND

The 3D city modelling is very useful and applicable in contemporary urban planning. Airborne laser scanners could not be able to capture side view of the buildings. Terrestrial laser scanners can precisely give the side view of the buildings. There is a need to have an open source software approach for processing and 3D visualisation of Terrestrial Data.

OBJECTIVES

Proposes a methodology for simple 3D city modelling using TLS data and an open source platform(PCL)

Implementation of Colour Based Region Growing Algorithm

Object or feature detection of the point cloud data.

The combination of geometrical and colour information leads to meaningful and effective segmentation results,

Work Flow For The Data Set

Point cloud from terrestrial laser

scanner

Clustering based on

color based region growing segmentation

algorithm

Visualization and building

object detection

Processing of point cloud :segmentation

Figure 1: Representation of the Objective

Figure 2: 3D View of the Original Data

Figure 3: Figure 4:

pcd files in the cloud compare viewer

METHODS

RESULTSSUMMARY

ACKNOWLEDGEMENT

CONTACT INFORMATION

BIBLIOGRAPHY

The data set is in las format. The sensor used for obtaining

this data was Terrestrial laser scanner RIEGL VZ-400.The

method of segmentation was done using the point cloud

library (PCL), an open source software library licensed

under Berkeley Software Distribution (BSD).

Convert LAS to txt format by LAS tools

Note the range of R,G,B

in the txt and the cloud width

Clustering of the point cloud

to form .pcd files

Visualization of

.txt files

Implement color based

region growing segmentation

algorithm

Segmentation based on both spatial and

spectral homogeniety XYZRGB(txt) format is converted to .pcd files.

Clusters are formed

Validation

Clusters are converted to txt format by C++ code

Color Based Region Growing Segmentation

Algorithm

Color is taken as the discriminator and Kd tree Search is

used here. The merging of the algorithm prevents over and

under segmentation. The algorithm attempts to merge

clusters with close colours. Two neighbouring clusters with a

small difference between average colours are merged

together. After that the next merging step takes place.Every

single cluster is verified by the number of points that it

contains. If this number is less than the user-defined value

then the current cluster is merged with the closest

neighbouring cluster. The procedure iterates and is

terminated when all the points are labelled. Here

segmentation is done based on both geometrical and color

information.

GSN Perera,N Hetti Arachchige “Efficient 3D City Modelling from Airborne Laser Scanning point clouds”, Proceedings of 8th International Research Conference, KDU, Published November 2015

Rusu, R.B, Cousins, S; "3D is here: Point Cloud Library (PCL)," Robotics and Automation (ICRA), 2011 IEEE International Conference on , vol., no., pp.1,4, 9-13 May 2011

Shaohui Sun and Carl Salvaggio; “Aerial 3D Building Detection and Modelling From Airborne LiDAR Point Clouds” IEEE Journal Of Selected Topics In Applied Earth Observations And Remote Sensing, Vol. 6, No. 3,June 2013,Pp1440- 1449

Douillard, B., Underwood, J, Kuntz, N., Vlaskine, V, Quadros, A, Morton, P, Frenkel, A; "On the segmentation of 3D LIDAR point clouds," Robotics and Automation (ICRA), 2011 IEEE International Conference on , vol., no., pp.2798-2805, May 2011;

Ying LIU, Ruofei Zhong; “Buildings and Terrain of Urban Area Point Cloud Segmentation based on PCL”, 35th Internat ional Symposium on Remote Sensing of Environment (ISRSE35) IOP Publishing, IOP Conf. Series: Earth and Environmental Science 17 (2014)

Parameters changed by iterative method are the following

Distance threshold

Point colour spacing

Region colour threshold

Minimum cluster size

In order to obtain effective segmentation and better results

some of the parameters in the C++ code were changed by

iterative method

Figure 5:Segmented result for the minimum cluster size 500

Figure 6:Segmented result for the minimum cluster size 600

Figure 8:Segmented result for the minimum cluster size 100

Figure 7:

Figure 9:

Segmented result for the minimum cluster size 150

Segmented result for the minimum cluster size 75

Table 1: Iterative results showing the variation of numbers of clusters with minimum cluster size

Table 2: Variation of building features according to the cluster size

Figure 10:Segmented result for the minimum cluster size 50

Threshold parameters are adjusted. From the processing,

better results were obtained by keeping the minimum cluster

size between 100 and 50. Other three parameters are kept in

fixed values. Features become completely visible for the

minimum cluster size values 50 and 75.The color based

segmentation become more effective for the threshold values

50 and 75.

Reducing the minimum cluster size, features become completely visible.

Main parameter here is colour, affecting the segmentation.

The integration of geometrical and colour information lead to more meaningful and useful information.

There is a balance between over and under segmentation.

The iterative method made it possible to find the key parameters.

These tasks are extremely laborious when carried out manually.

Helps in detecting objects or features of interest.

We have to test the key parameters each time.

There are no automatic methods for finding the suitable values of key parameter for effective segmentation.

Key parameters are to be obtained by iteration method.

Further research can be done on the inter dependance of the parameters.

Terrestrial laser scanner and the colour based region growing segmentation method enable the building detection and urban applications more effective.

Minimum cluster size

Ob

serv

ed

cla

sse

s 600 500 150 100 75 50

Only

very

few walls,

one pole,

roofs

More number of

walls, roofs,

2 boards,

2 poles ,tree

Walls, roofs,

windows ,

tree, poles,

boards

Almost all

walls and roofs

are separately

visualized, tree,

poles and the

boards

Almost every

parts including

the air conditioner

are visible

Clear

visualization of

every part

including walls,

roofs, windows,

boards, poles,

air conditioner.

Sincere thanks to Kerala State Council for Science and

Technology, Cochin University of Science & Technology

(CUSAT), Dr M.V. Harindranathan Nair, Associate Professor,

School o f Env i ronmenta l Stud ies , CUSAT and

Dr.Jaishanker.R.Nair, Associate Professor, Indian Institute of

Information Technology & Management Kerala

1. Mrs Indu Indira Bai,

Research Scholar, School of Environmental Studies, Cochin

University of Science and Technology, Kochi, Kerala, India.

Mob-+91 9895160999, Email- [email protected]

2. Dr Rama Rao Nidamanuri,

Associate Professor, Department of Earth and Space Sciences,

Indian Institute of Space Science and Technology, Valiamala,

Trivandrum, Kerala, India

Mob- +91 8129285705, Email – [email protected] result by changing the

threshold values

ASPRS - IGTF 2016