Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Laser Vision-Based Plant Geometries
Computation in Greenhouses
Yu Zhang 1, 2
and Xiaochan Wang 1, 2*
1. College of Engineering, Nanjing Agricultural University, Nanjing, 210031, China;
2. Jiangsu Province Engineering Laboratory for Modern Facilities Agricultural Technology and Equipment, Nanjing,
210031, China
*Corresponding author, Email, [email protected], [email protected]
Abstract—Plant growth statuses are important parameters
in the greenhouse environment control system. It is time-
consumed and less accuracy that measuring the plant
geometries manually in greenhouses. To find a portable
method to measure the growth parameters of plants
portably and automatically, a laser vision-based
measurement system was developed in this paper, consisting
of a camera and a laser sheet that scanned the plant
vertically. All equipments were mounted on a metal shelf in
size of 30cm*40cm*100cm. The 3D point cloud was obtained
with the laser sheet scanning the plant vertically, while the
camera videoing the laser lines which projected on the plant.
The calibration was conducted by a two solid boards
standing together in an angle of 90. The camera’s internal
and external parameters were calibrated by Image toolbox
in MatLab®. It is useful to take a reference image without
laser light and to use difference images to obtain the laser
line. Laser line centers were extracted by improved centroid
method. Thus, we obtained the 3D point cloud structure of
the sample plant. For leaf length measurement, iteration
method for point clouds was used to extract the axis of the
leaf point cloud set. Start point was selected at the end of the
leaf point cloud set as the first point of the leaf axis. The
points in a radian of certain distance around the start point
were chosen as the subset. The centroid of the subset of
points was calculated and taken as the next axis point.
Iteration was continued until all points in the leaf point
cloud set were selected. Leaf length was calculated by curve
fitting on these axis points. In order to increase the accuracy
of curve fitting, bi-directional start point selection was
useful. For leaf area estimation, exponential regression
model was used to describe the grown leaves for sampled
plant (water spinach) in this paper. To evaluate the method
in a sample of 18 water spinaches, planted in the greenhouse
(length 16 meter and width 8 meter) on the roof of library
building in Nanjing Agricultural University, the lengths of
200 leaves and were measured manually and plotted versus
their automatically measured counterparts. The accuracy of
leaf lengths is 95.39% respectively. For the leaf length
measurement, the average error is less than 5mm which is in
the boundary of error. A few laser measurement results take
on larger errors, because that some leaves were shaded from
the others which make an effect on the curve fitting. In
additional, the results are affected by the over-wide laser
line. Manual measurement gave result information an
accuracy of millimeter level, while laser measurement
improved the measurement precision. For the leaf area
estimation, the accuracy of modeling trained by 200
sampled leaves, is 81.9% with 50 testing sampled leaves. For
the plant growth properties, plant height and canopy width
are obtained by point clouds reconstruction of plant. The
test experiment proved that laser vision-based method could
be used on plants geometry measurement and growth
monitoring in greenhouses. This method will improve
visualization and digitalization of plants in greenhouses, and
make a progress on greenhouse environment control system.
Index Terms—Laser Vision; 3D Point Cloud; Plants
Geometry; Leaf Length; Leaf Area; Plant Height; Canopy
Width
I. INTRODUCTION
A facility farming is growing fast in recent years.
Many scientists and experts are working hard on the
artificial control of greenhouse environment [11, 14, 25,
27, 30, 31]. But the target of greenhouse environment control is providing appropriate environment for the
crops in the greenhouse. Results of the control system
were displayed on the status of crop growth. So it is of
more importance to detect the crop response at different
growth status [25]. Remote sensing [17] is widely used in
field farm crops, while in greenhouses it is not adaptive
due to the block of glass. So many experts made endeavor
on the geometric crop properties detection in greenhouses
to obtain the crop growth response for the plant speaking-
based greenhouse environmental control system.
In conventional greenhouse, crop growing conditions are obtained via human observations or presetting
environmental parameters, instead of the plants’ specific
needs at a different status. Contact sensing is typically
used to determine a plant’s physical characteristics, a
process that is cumbersome, labor-intensive, and often
destructive. Computer vision is a good choice for
geometric crop properties detection in greenhouses,
because of non-contact sensing, non-destructive and time-
saving. David Story [4] used machine vision-guided plant
sensing and monitoring system to detect calcium
deficiency in lettuce crops grown in greenhouse
conditions using temporal, color and morphological changes of the plant. The machine vision system
extracted plant features to determine overall plant growth
and health status, including top projected canopy
area(TPCA)as a morphological feature; red–green–
blue(RGB) and hue–saturation–luminance(HSL)values as
color features; and entropy, energy, contrast, and
534 JOURNAL OF MULTIMEDIA, VOL. 9, NO. 4, APRIL 2014
© 2014 ACADEMY PUBLISHERdoi:10.4304/jmm.9.4.534-541
TABLE I. NOTATION
α Coefficient of leaf area regression model ID Inner diameter of donut point on calibration board (Fig. 3)
β Coefficient of leaf area regression model OD Outer diameter of donut point on calibration board (Fig. 3)
dcrit
Predefined distance in iteration for leaf axis points extraction Oc Center point of camera pinhole model (Fig. 2)
Leaf-i The target leaf points subset OcZc Camera optic axis (Fig. 2)
f Focal length (Fig. 2) L L LO X Y A planar coordinates on the laser light plain(Fig. 2)
l Leaf length, cm (Fig. 11,Tab3) 0P ,
1P , ∙∙∙ Basal point in iteration for leaf axis points extraction
L Laser line sheet (ui,vi) Image pixel coordinates (Fig. 2)
LA Estimated leaf area, cm2 (Fig. 11,Tab3) (xwi,ywi,zwi) World coordinates (Fig. 2)
homogeneity as textural features. The methodology
developed was capable of identifying calcium-deficient lettuce plants 1 day prior to visual stress detection by
human vision. Ling et al [12] used spectral and
morphological characteristics of lettuce leaves to detect
nutrient deficiency. This study showed that the
reflectance of wave bands between 415nm and 720 nm
could be used as a signal wave band for machine vision
implementation. A wave band closer to the visible was
recommended because it gives a better signal strength.
The study also suggested the possibility of using multiple
signals based on water and nutrient stress detection in
plants, using machine vision systems that could determine deviations from “well-grown” as an indicator
of deficiencies and plant status. Meyer et al [16] used a
machine vision system to detect single leaves and
poinsettia foliage, and reported that a normalized
difference index provided the best method of
discriminating nitrogen-deficient from healthy plants.
Low-nitrogen plants grown in greenhouses and growth
chambers showed similar increases in red reflectance, but
had different levels of near-infrared reflectance due to
differing amounts of plant canopy cover. E.V.Henten [18]
used an indirect non-destructive measurement by means
of image processing to detect crop growth of lettuce and build 3D models of the relationship between relative soil
coverage by the crop canopy and by dry weight.
Geometric crop properties measurements are a broad
research topic in agriculture. With the development of
computer vision, 3D measurement shows a great potential
in vegetation growth parameters measurement [1-3, 10].
Hans J. Andersen [7] investigated the potential of using
area-based binocular stereo vision for three-dimensional
analysis of single plants and estimation of geometric
attributes such as height and total leaf area. Dominik
Seidel [5] measured total above-ground biomass (stems, twigs, leaves), the biomass of axes (stems and twigs),of
leaves biomass and the leaf area of 63 experimental trees
by means of terrestrial 3D laser scanner. Hosoi and
Omasa [8] used a portable 3D laser scanner to calculate
canopy leaf area density profiles for deciduous trees. His
studies focused on measurements of the biomass, and
mostly concentrated on forest trees resting upon
relationships among parameters which were measured
with terrestrial laser scanners, e.g. diameter at breast
height or total tree height, and the trees biomass [28]. The
application of terrestrial laser scanning (TLS) and mobile or vehicle based laser scanning (MLS) enables fast 3D
data acquisition. They provide a direct capture of tree
geometry generating high resolution point measurements
representing the geometrical characteristics of target
objects in 3D space [13]. Generally, Terrestrial LiDAR (Light Detection and Ranging) Scanners (TLSs) have
recently turned into potential tools for 3D vegetation
structure measurement. It provides accurate measures of
distances to objects using a large number of sampling
laser beams within the instrument field of view, thus
creating a cloud of points positioned in 3D space. It is
widely used in field farms and forests. But in greenhouses,
TLSs do not show the superiorities because of the space
limitation.
In our studies, we are aiming to find a portable and low
cost vegetation geometric properties measurement system basing on laser vision. The main parameters of plant
geometric properties include the leaf length, leaf area,
plant height and canopy width of a single crop in
greenhouses. This laser vision-based measurement
system will have great prospects on both improving the
visualization and digitization for plants growth
monitoring and analysis and providing multi-aspect data
for the greenhouse environmental control system.
II. MATERIALS AND METHODS
A. Crops and Acquisition
A growth experiment with water spinaches in
greenhouses which was built on the roof of laboratory
building at College of Engineering, Nanjing Agricultural
University served as the study object to test the
applicability of 3D-laser scanning as a non-destructive
method for growth parameters detection in greenhouses,
which is a three ridge-Venlo greenhouse size 8m* 16m.
The measurement principles, as shown in Fig. 1, is
comprised of a CCD web-camera (HikVision® DS-
2CD3332D-I) that recorded the video of laser scanning
line, projected onto the crop stems by a laser sheet with a downward slope angle of 45 degree. The laser sheet was
created by mounting a horizontal raster on a 80 mW, 520
nm (green) laser diode. And it had an approximate
thickness of 2 mm and an aperture angle of 60 degree.
The laser sheet was mounted on a Pan-Tilt (PTZ®W3300)
with horizontal rotation angle of 355 degree and vertical
lift angle of 60 degree. All the components were mounted
on a mental shelf with 30cm by 40cm by 100cm. The
laser sheet scanned the crop up and down with a
horizontal laser line. A PC machine was used to control
the Pan-Tilt through 485-cable and the camera through a
network, respectively. The video resolution was 1080*1270 pixels with the frame of 30 fps, and the frame
image resolution was 640*480 pixels by programming in
JOURNAL OF MULTIMEDIA, VOL. 9, NO. 4, APRIL 2014 535
© 2014 ACADEMY PUBLISHER
MatLab®. The camera was calibrated using a standard
procedure contained in a MatLab® Toolbox based on
Tsai’s method, which enabled distortion correction of the
imagery.
Laser sheet
PTZ Pan-tilt
Camera
Metal shelf
Figure 1. 3D laser sheet measurement system
B. Calibration of 3D Measurement System
3D measurement system consists of camera perspective transformation model and light plain equation
model [32, 34] which is used for building the relation
between 2D coordinates and 3D coordinates of target
points, shown in Fig. 2. Oc is the center point of camera
pinhole model. OcZc is the camera optic axis, which is
perpendicular to CCD plain, with point O as the
intersection. The distance of O-Oc is the focal length, f.
X-axis and Y-axis in retinal coordinates is parallel to u-
axis and v-axis in image coordinates, respectively.
yL
xL
Laser
source
OL
X
Y
Zc
xc
yc
u
v
O
Oc
P(u,v)
O(u0,v0)
Zw
Xw
YwOw
Laser light plane
Object
CCD
Screen
Figure 2. Sketch of 3d measurement principle of line plane equation
In the process of measurement, laser line sheet (L)
projects a curve on the surface of the object that contains
3D spatial information of the object. In the world coordinates, laser light plain equation in the camera
coordinates can be described as:
acXc + bcYc + ccZc + dc = 0 (1)
On the laser light plain, planar coordinates L L LO X Y
was built. The relation between the laser light plain and
the retinal plain is
31 2
4 5 6
7 8 91
L
x
L
y
L
z
Xr tr rx
Yy r r r t
Zz r r r t
(2)
where 0Lz , * , *x y
X f Y fz z
were substituted
into Eq. (2).
1 2
4 5
7 8
1 2 3
4 5 6
7 8
1 1
*
1 1
* *
1
x L
y L
z
L
Z L
L
Z L
X fr fr fT X
Y fr fr fT Y
r r T
a a a X
T a a a Y
a a
X
T A Y
(3)
Penalty function was used to solve the optimal
equation, as followed,
1 2 3
7 8
4 5 6
7 8
*( 1)
*( 1)
xi Li Li
i Li Li
yi Li Li
i Li Li
f a X a Y a
X a X a Y
f a X a Y a
Y a X a Y
(4)
where i=1,2,3,…N, and the constrains of the penalty
function is
2 2 2
1 1 1 4 7
2 2 2
2 2 2 4 8
3 3 1 2 4 5 7 8
( 1)
( 1)
( )
p
p
p
f M a a a
f M a a a
f M a a a a a a
, corresponding
optimal target function 2 2 2 2 2
1 2 8 1 2 31( , ,..., ) ( ) min
n
xi yi p p piI a a a f f f f f
. Gaussian-Newton method was used to solve the parameters.
Integrating traditional camera calibration plane targets
and light plain calibration [27, 35] 3D targets, a vertical
target was used in this paper, consisted of two rectangular
boards standing at an angle of 90 degree. A certain
number of solid dots 10mm in diameter and donut dots
with OD 10mm and ID 6mm were spread on each of the
board target evenly. The distance between centers is
50mm. Two targets, in the size of 210mm width and
297mm height, were mounted at a corner at an angle of
90 degree. The intersection of the two boards was defined as Y-axis, defined the mid-point as the original point of
the world coordinates. OX and OY were defined on the
target board in the right-hand rule, respectively. An
image without laser line was taken. The image
coordinates of all the solid dots and donut dots and that in
the world coordinates were known. And considering the
radial constrains in the camera, we can obtain
1 2 3 4
5 6
w d w d w d w d
w d w d d y
x Y r y Y r z Y r x X r
y Y r z X r X t
(5)
Substitute the known point coordinates
, , , ,wi wi wi i ix y z u v into equation (5), we can obtain
,* dA B x (6)
where the row vector , , ,
w d w d w d d w d w d w dA x y y y z y y x x y x z x
536 JOURNAL OF MULTIMEDIA, VOL. 9, NO. 4, APRIL 2014
© 2014 ACADEMY PUBLISHER
is known, and the column vector
1 2 3 4 5 6
T
y y y x y y y yB r s t r s t r s t t s t r t r t r t
is unknown. All the parameters can be solved when the
known point coordinates were substituted into Eq. (6). An
image with laser line on it was taken. The points were
both on the laser line and the plain of XY and YZ.
Similarly, the known point coordinates were substituted
into Eq.(1), and the equation of laser light plain in the world coordinates was
1 1 1
2 2 2
0
0
0
w w w w w w w
w w w w w w w
w wn w wn w wn w
a x b y c z d
a x b y c z d
a x b y c z d
(7)
Least-square method was used to solve the equation set,
and the parameters of laser light plain equation in the
world coordinates were obtained. Since the vector
x y zt t t , 1 4 7 2 5 8r r r and r r r are
transformation vector and the unit direction vectors of
LX -axis and LY -axis in camera coordinates respectively,
we can obtain
4 8 5 7
2 7 1 8
1 5 2 4
c
c
c
c c x c y c z
a r r r r
b r r r r
c r r r r
d a t b t c t
(8)
Parameters 3r ,
6r , 9r can be calculated by the
transformation of camera coordinates and world
coordinates.
○ ○
○
○ ○○
●
●
●●
●
●
●
●
●●
●
●●
●●
●●
●
●●
●
●●
●●
●
●
●●
●●
●●
●●
●●
●●
●●
●●
●
○○
○
○
○
○
Y
XZ
○ ○
○
●
●
●
●
●
●
○●
●
●
●
● ●
●●●
●
●●
●
●
○●
●
○●●
●●●
●
● ●
●●
●
●
●●
●●
●●
○○
○ ○
○
○
XZ
Y
Figure 3. Sketch of calibration
C. Images Processing and Laser Line Centroid Method
1) Images Processing The first step was background removing to reduce the
noise disturbances after extracting the frame images.
Difference imaging was used to remove the background.
No laser line projected video was recorded at the beginning. Afterwards, laser sheet was opened to scan on
the crop. So images with laser line projected and without
laser line projected were extracted. The laser line stripe
was shown in Fig. 4.
Figure 4. Background removing result
2) Laser Line Centroid Extraction and Point Clouds
Reconstruction The second step was extracting the laser line centers.
The noise of the sunlight, reflection of the surface of the
target crop and other disturbances would make an
influence on the laser line centers extraction. So the noise
filtering was a necessary process before proceeding to the
extraction. Ideal distribution of laser light is Gauss
Distribution, so Gauss filter was chosen to remove the
noises.
Conventional center extraction methods were
extremum method and threshold method, while the two
methods were affected by the noises severely. In this
paper, centroid method [19, 20] based on the center of gravity computation of rigid-body, was chosen to obtain
the centroid of the pixels gray distribution on every cross-
section of the laser line stripe. Fig. 5 showed result of
laser line centers on one image. The pixel coordinate of
the center points were easily obtained by programming in
MatLab®. Thus, 3D point clouds construction [9, 17, 18,
21, 22, 26] of target plant was plotted in Fig. 6.
Figure 5. Laser line centers
D. Leaf Length Calculation
In this paper, we proposed an iteration which fit with
the features of plant point clouds. Axis points of a leaf
were extracted and fitted by curve fitness. The length of
axis curve was the leaf length. Details were as followed:
Firstly, rotate the main plot of the plant, select the target
leaf, and save the target leaf points as leaf subset, named
Leaf-i. A basal point 0P was chosen by mouse click,
JOURNAL OF MULTIMEDIA, VOL. 9, NO. 4, APRIL 2014 537
© 2014 ACADEMY PUBLISHER
which is usually at the end of the leaf point subset in 3D
plot in the first iteration step(i=1). This basal point 0P
became the first point in the axis point set. Secondly, a
known distance dcrit
was predefined. All points in Leaf-i
with a distance to 0P less than d
crit are selected by going
through all over the points in Leaf-i, namely subset i.
Calculation of the centroid of subset i is conducted,
marked as 1P ,which is the basal point in the next iteration
step. And all the points selected in the first iteration step
are deleted. The next iteration step stats from the reduced
leaf subset. Iteration continues until all the points in the
subset of Leaf-i are deleted. At last, all centroid points are
saved in point set named Ta. The length of the curve
made of these centroid points is the length of the leaf. Fig.
7 shows the result of the leaf axis extraction. Sometimes, bi-directional iteration is necessary to avoid the missing
point at the end of the point set. The red points and the
blue points are two axis extraction results basing on two
start points at two ends of the point set. All the two axis
points are used to fit the curve. This method shows robust
in case that there are some missing points in the leaf point
set because of laser sheet scanning angle.
Figure 6. Point clouds of crops
5060
7080
90100
40
50
60
70
80
50
55
60
65
70
75
80
85
90
Points Cloud
Figure 7. Leaf axis extraction
E. Leaf Area Estimation
Leaf area takes on a high correlation to leaf length,
which can be described by mathematical model. Different
plants show different correlation models on leaf length
and leaf area. In this paper, Pearson Correlation Analysis
was conducted at a sample of 200 water spinach leaves
obtaining regression model [33].
ln( ) *LA l (9)
III. RESULTS
A. Leaf Length Calculation
In the process of leaf length calculation, the number of
leaf axis points affected the result of leaf axis curve
fitting. Technically, the value of dcrit
should be larger
than half of the width of leaf point set. The value of dcrit
is in the inverse proportional relationship to the number
of axis points. If dcrit
was assigned a large value, the
number of axis points was too rare to finish the curve
fitting. Otherwise, if dcrit
was assigned a small value, the
number of axis points was too redundant to ensure the
accuracy of curve length calculation. In this paper, the
value of dcrit
ranged from 1 to 20, producing a set of axis
point number and fitted leaf length couples. In Fif.8,a
singular point valued 50 of the axis point number fit with
d crit valued 1. In this figure, the value of d crit can be
inversely computed from the average number of axis
point 7. Finally, the length of leaf can be obtained.
dcirt Number of axis point Leaf length
0
10
20
30
40
50
Figure 8. Iteration step result
La
ser
mea
sure
men
t/cm
Manual measurement/cm
Figure 9. Comparison of laser measurement and manual measurement
of leaf length
538 JOURNAL OF MULTIMEDIA, VOL. 9, NO. 4, APRIL 2014
© 2014 ACADEMY PUBLISHER
TABLE II. RESULTS OF ANOVA ANALYSIS OF LEAF AREA ESTIMATION MODEL
Sum of Squares df Mean Square F Sig.
Regression 17.383 1 17.383 267.97 .000
Residual 12.455 192 0.065
Total 29.838 193
The independent variable is length
TABLE III. COEFFICIENTS AND ANALYSIS OF LEAF AREA ESTIMATION MODEL
Unstandardized Coefficients Standardized Coefficients
t Sig. B Std.Error Beta
l (length) 0.164 0.010 0.763 16.370 .000
(Constant) 1.923 0.174 5.315 11.021 .000
The independent variable is l(length)
The dependent variable is ln(LA)
To evaluate the performance of the laser scanning
measurement system, a sample of 180 water spinach
leaves were measured using manual measurements and
laser scanning measurement system. Fig. 7. shows the result of leaf axis extraction. Fig. 9. shows the calculated
leaf length from laser scanning measurement system,
versus the manually measured leaf length. A straight line
was fitted, yielding a slope of 0.998, an intercept of 0,
and a standard error of 0.909 and a coefficient
determination of 0.915. For the leaf length measurement,
the average error is less than 5mm with accuracy of leaf
lengths is 95.39%, which is in the boundary of error. A
few laser measurement results take on larger errors,
because that some leaves were shaded from the others
which make an effect on the curve fitting.
B. Leaf Area Estimation
To establish the regression model of leaf length and
leaf area, a destructive experiment of leaf sampling was
conducted. All the sampled leaves were listed on a piece
of A4 paper, which is marked with a calibration square
with area of 1*1cm². Real areas of leaves were computed by pixel counting, comparing with the calibrated square.
The regression model was established with 200 training
leaves and 50 testing leaves. Fig. 10 shows the original
image and segmented image of sampled leaves.
Figure 10. RGB image and segmentation image of leaves
Exponential regression model was used to evaluate the
leaf area, and the result was derived, with R 2
adj of 0.9864
and confidence interval of 95%. The result of the leaf
area model is shown in Fig. 11.
ln( ) 0.164 1.923LA l (10)
Results of ANOVA analysis of leaf area estimation
model are shown in Table 2 and Table 3 with the model
significance of 0 and F statistical testing of 267.97. The
coefficients of the leaf area model are shown in Table 3
with standard error of 0.763 and 5.315. The accuracy of
the model is 81.9%.
To test the leaf area model, the area of 50 sampled
leaves were measured by pixels calibration and estimated by the model. Relationship of leaf area estimation and
pixels calibration measured was shown in Fig. 12. A
straight line was fitted, yielding a slope of 0.947, an
intercept of 0.434.
Leaf length/cm
Lea
f are
a/c
m²
Figure 11. Regression model of leaf length and leaf area
Manual measurement/cm²
La
ser
mea
sure
men
t/cm
²
Figure 12. Relationship of leaf area estimation and observed
measurement
C. Growth Trend
Laser scanning the whole plant profile can be used to
record and monitor plant growth with leaf area, plant
height and canopy width. A sampled water spinach
growth conditions (plant height, canopy width) was
monitored within 10 days after field planting. The height
JOURNAL OF MULTIMEDIA, VOL. 9, NO. 4, APRIL 2014 539
© 2014 ACADEMY PUBLISHER
growth of sampled plant was shown in Fig. 13(a), and the
canopy width growth of sampled plant was shown in Fig.
13(b). Fig. 14 shows the leaf area growth of six sampled
plants during the period of June twenty-seventh and July
twentieth in 2013.
(a)
(b)
Figure 13. Growth of heights and canopy area(Field planting Day 1,
Field planting Day 4, Field planting Day 7, Field planting Day 10)
Figure 14. Total leaf area growth trend
IV. DISCUSSIONS
Plant geometries measurement system based on laser vision performed well in measuring leaf length, plant
height and canopy width. Our leaf axis point extraction
has been based mainly on iteration fitting to the
properties of plant 3D point clouds. This method
addressed the situation of missing points in the leaf point
set. And bi-directional iteration increased the accuracy of
axis point extraction (shown in Fig. 7). Exponential
modeling was applied to estimate the leaf area using leaf
length avoiding destructive sampling in leaf area
measurement. To record and monitor the whole process
of plant growth, leaf areas models of different growth
periods are necessary. In this paper, we modeled the leaf
area with leaf length to estimate the leaf area, but in
additional, comparison a mound the leaf area estimation
using TLS [15] and the traditional LAI estimation is of
value.
For the leaf length measurement, the average error is
less than 5mm with accuracy of leaf lengths is 95.39%,
which is in the boundary of error. A few laser
measurement results take on larger errors, because that some leaves were shaded from the others which make an
effect on the curve fitting. In additional, the results are
affected by the over-wide laser line. Manual
measurement gave result information an accuracy of
millimeter level, while laser vision-based measurement
system improved the measurement precision. For the leaf
area estimation, the accuracy of modeling trained by 200
sampled leaves is 81.9% with 50 testing sampled leaves.
For the plant growth properties, plant height and canopy
width are obtained by point clouds reconstruction of plant.
The test experiment proved that laser vision-based method could be used on plants geometry measurement
and growth monitoring in greenhouses. This method will
improve visualization and digitalization of plants in
greenhouses, and make a progress on greenhouse
environment control system.
Although we have concentrated on laser vision-based
plant geometries measuring, we would also expect, more
generally, that 3D point clouds information would be
capable of measuring more phenotypes [3] of plants in
greenhouses. The laser vision sensors could also be easily
combined with additional sensors to monitor automatically for stress, deficiency [4], spectral
information et al at the same time. An overall phenotypes
database of plant would be established to provide
information for greenhouse environment control system.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the funding of
their projects of ‘Research of Multi-scale Greenhouse
Environmental Control System Based on Crop
Information Fusion’ supported by National Natural
Science Foundation of China.
REFERENCES
[1] Alexander M S, Keith L E, Randy W, etc. “Measuring Canopy Coverage with Digital Imaging,” Communications in Soil Science and Plant Analysis, 2007, 38 pp. 7-8, 895-902.
[2] Carlos C, Garcia MI, Daza C, etc. “Study of a Non-destructive Method for Estimating the Leaf Area Index in Vegetable Crops Using Digital Images,” HORTSCIENCE, 2010, 45(10) pp. 1459–1463.
[3] Christopher J. K, John M. N, Stith T. G. “Measurements of leaf orientation, light distribution and sunlit leaf area in a boreal aspen forest,” Agricultural and Forest Meteorology, 1998, 91 pp. 127-148.
[4] David S, Murat K, Chieri K, etc. “Lettuce calcium deficiency detection with machine vision computed plant
540 JOURNAL OF MULTIMEDIA, VOL. 9, NO. 4, APRIL 2014
© 2014 ACADEMY PUBLISHER
features in controlled environments,” Computers and Electronics in Agriculture, 2010, 74 pp. 238–243.
[5] Dominik S, Friderike B, Dietrich H, etc. “3D-laser scanning: A non-destructive method for studying above-ground biomass and growth of juvenile trees,” Agricultural and Forest Meteorology, 2011, 151 pp. 1305–1311.
[6] Fontana M, Rizzi C, Cugini U. “3D virtual apparel design for industrial applications,” ComputerAided Design, 2005, 37(6) pp. 609622.
[7] Hans J A, Lars R, Kristian K. “Geometric plant properties
by relaxed stereo vision using simulated annealing,” Computers and Electronics in Agriculture, 2005, 49 pp. 219–232.
[8] Hosoi, F., Omasa, K., “Factors contributing to accuracy in the estimation of the woody canopy leaf area density profile using 3D portable lidar imaging,” J. Exp. Bot., 2007, 58, 3463–3473.
[9] Huang J, Menq C H. “Combinatorial manifold mesh
reconstruction and optimization from unorganized points with arbitrary topology,” Comput. Aided Des, 2002, 34, 149–165.
[10] Jiangui Liu, Elizabeth P, Stuart A. “Assessment of in situ crop LAI measurement using unidirectional view digital Photography,” Agricultural and Forest Meteorology, 2013, 169 pp. 25–34.
[11] JoséBoaventura C, Moura Oliveira de J P. “Optimal
Management of Greenhouse Environments,” EFITA 2003 Conference, 2003(7), 5-9.
[12] Ling P P, Giacomelli G A, “Russell T P. Monitoring of plant development in controlled environment with machine vision,” Advances in Space Research, 1996, 18(4–5) pp. 101–112.
[13] Livny, Y., Feilong, Y., Olson, M., Chen, B., Zhang, H., El-Sana, J.. “Automatic reconstruction of tree skeletal
structures from point clouds,” ACM Transactions on Graphics, 2010, 29(6), 1511–1518.
[14] Mario T P, Gilberto H R, Jose R M, etc. “Greenhouse Energy Consumption Prediction using Neural Network Models,” International Journal of Agriculture and Biology, 2009, 11(1) pp. 1-6.
[15] Martin B, Jean-Luc W, Richard A F, etc. “Estimating leaf area distribution in savanna trees from terrestrial LiDAR Measurements,” Agricultural and Forest Meteorology,
2011, 151 pp. 1252–1266. [16] Meyer G E., Troyer W W, “Fitzgerald J B, etc. Leaf
nitrogen analysis of poinsettia (Euphorbia Pulcherrima Will D.) using spectral properties in natural and controlled lighting,” Applied Engineering in Agriculture, 1992, 8(5) pp. 715–722.
[17] Pachidis T P, Lygouras J N. Pseudo. “Stereo-vision system: a monocular stereo-vision system as a sensor for real-time
robot applications,” IEEE Transactions on Instrumentation and Measurement, 2007, 56(6), 2547-2560.
[18] Pagàs J, Salvi J, Garcäa R, Matabosch C. “Over view of coded light projection techniques for automatic 3D profiling,” IEEE International Conference on Robotics and Automation, ICRA 2003, pp. 133-138, Taipei, Taiwan, 14-19th September 2003.
[19] Parsons N R, Edmondson R N, Song Y. “Image analysis
and statistical modelling for measurement and quality assessment of ornamental horticulture crops in glasshouses,” Biosystems engineering, 2009, 104 pp. 161–168.
[20] Remondino F. “3D reconstruction of static human body shape from image sequence,” Computer Vision and Image Understanding, 2004, 93(1) pp. 6585.
[21] Saxena A, Chung S H, Ng A Y. “3-D depth reconstruction from a single still image,” International Journal of
Computer Vision, 2007, 76(1) pp. 53-69. [22] Surmann H, Nuchter A, Hertzberg J. “An autonomous
mobile robot with a 3D laser range finder for 3D exploration and digitalization of indoor environments,” International Journal of Robotics and Autonomous Systems, 2003, 45(2) pp. 181-198.
[23] Tao Li, Sun Changku, He Li, Ye Shenghua. “Research on the color 3D profilometry and parameter calibration
technology,” Proceedings of SPIE-The International Society for Optical Engineering, v5638, Optical Design and Testing, 2005, II pp. 464-472.
[24] Tetsuo M, Yasushi H. “An intelligent control for greenhouse automation, oriented by the concepts of SPA and SFA—an application to a post-harvest process,” Computers and Electronics in Agriculture, 2009, 29 pp. 3-20.
[25] Van Henten E J, Bontsema J. “Non-destructive Crop Measurements by Image Processing for Crop Growth,” J. agric. Engng Res, 1995, 61 pp. 97-105.
[26] Wang Dingcheng, Qiao Xiaojun, Wang Chunxiu, et al. “Simulation of greenhouse environment SVMR control based on plant response,” Transactions of the CSAE, 2010, 26(Supp. 2) pp. 290-293. (in Chinese with English abstract).
[27] Watanabe T, Hanan J S, Room P M, etc. “Rice morphogenesis and plant architecture: measurement, specification and the reconstruction of structural development by 3D architectural modeling,” Ann. Bot, 2005, 95, 1131–1143.
[28] Watt, P. J., Donoghue, D. N. M., Dunford, R. W., 2003. “Forest parameter extraction using terrestrial laser scanning. In: Proceeding of the ScandLaser Scientific
Workshop on Airborne Laser Scanning of Forests,” in Umea, Sweden, 2–4 September 2003, pp. 237–244.
[29] Weixing Zhu, Hanping Mao, Pingping Li, etc. “A Fuzzy Controller Based on Genetic Optimization and Its Application to a Distributed Control System of Greenhouse,” Transactions of the Chinese Society for Agricultural Machinery, 2002, 33(3) pp. 76-78.
[30] Wilczkowiak M, Boyer E, Sturm P. “Camera calibration and 3D reconstruction from single images using
parallelepipeds,” IEEE International Conference on Computer Vision, 2001(1) pp. 142-148.
[31] Xiaochan Wang, Weimin Ding, Weihong Luo. “Design Experiment to Measure Microclimate and Study Transpiration in Summer in Greenhouse,” Transactions of the Chinese Society for Agricultural Machinery, 2003, 34(2) pp. 86-89.
[32] Xiaochan Wang, Weimin Ding. “Simulation and analysis
of micro-climate of gutter connected Venlo greenhouse during rainy season in Jianghuai region of China using BP neural network,” Transactions of the CSAE, 2004, 20(2) pp. 235-238.
[33] Xiaojin Gong, Ying Lin, Jilin Liu. “Extrinsic calibration of a 3D LIDAR and a camera using a trihedron,” Optics and Lasers in Engineering 51(2013) pp. 394–401.
[34] Yin Xinyou, van L, H. H.. “Crop Systems Dynamics: An
eco-physiological Simulation Model for Genotype-by-Environment Interactions,” Wageningen Academic Publishers, 2005.
[35] Zhiqin Xu, Li Tao, Changku Sun, YiZhong Zheng. “A Rapid Calibration Method for the Laser Scanning 3D Measurement System,” Proceeding of ISMTI’2003. Hong Kong, China, 2003, 151 Key Engineering Materials, 2003, 295-296 pp. 735-740.
JOURNAL OF MULTIMEDIA, VOL. 9, NO. 4, APRIL 2014 541
© 2014 ACADEMY PUBLISHER