Application of LiDAR to determine vine growth in the vineyard · Application of LiDAR to determine...
1
FOR FURTHER INFORMATION Figure 1: Above: The GRover frame in a vineyard. Numbers indicate: 1) The Sick LMS-400 LiDAR and GPS- IMU unit. 2) Hinge point for the collapsible mast. 3) Junction boxes routing power to the instruments. 4) Wheels driven by in-hub electric motors. 5) Wheel encoder that tracks linear distance 6) 48v batteries powering the wheels and a 24v instrument battery 7) Laptop for data collection. Left: GRover in a trailer with the mast retracted, contained in the boom-rest, ready for transportation to a vineyard. REFERENCES ACKNOWLEDGEMENTS Application of LiDAR to determine vine growth in the vineyard The Grapevine Rover (GRover), how LiDAR is deployed in the vineyard LiDAR operates in a similar way as sonar. Instead of using sound, though, it uses light. The LiDAR used on GRover is the Sick LMS-400-200 (Sick, Germany; Fig. 1,1). An eye-safe line-scanner that emits laser pulses 300 times a second in a 70 degree field of view (Fig. 1,1). Those pulses reflect off of a target and are returned to the LiDAR sensor, providing information about the position of the return in 3D space and the reflectivity of the target’s surface. Scans of vine growth are visualized and analysed using the open source 3D point cloud and mesh processing software Cloud Compare (http://www.danielgm.net/cc/). Figure 2a shows an example canopy and 2b an example of what a 3D point cloud of a vine looks like in Cloud Compare. The colours in 2b illustrate the differences in reflectivity of the different surfaces in the scan: leaf, wood, grass and even the plastic tag. Developing relationships between LiDAR scans and vine growth. Field trials and computational analysis After producing 3D point clouds, the challenge is to correlate scans with meaningful biological parameters. To that end, field work was conducted at the South Australian Research and Development Institute (Nuriootpa, SA), scanning vines before and after destructively quantifying the leaf area, cane weight, bunch number and cordon volume. The example canopy (Figure 2a) shown on the left had its leaves plucked off and the area of those leaves measured with a LiCOR 3000C. The vine on the right, with leaves intact, is for comparison. The Grapevine Rover (GRover) is an adaptable proximal-sensing platform designed to provide rapid, high- quality information of plant performance in the field. GRover currently uses light radar (LiDAR) to produce precise (± 3mm), high resolution (300 scans/second), three-dimensional point clouds of vine growth. This poster explains: how LiDAR scans of vine growth have been collected in the vineyard, processed computationally and correlated with plant growth. Results from field trials have shown that LiDAR can accurately capture leaf area and pruning weight. Continued efforts will make data processing more automated and focus on capturing additional vine-traits with LiDAR and other instruments. Matthew Siebers 1 , Mark R. Thomas 1 , Everard Edwards 1 , Jose Jimenez Berni 2 , Rob Walker 1 1: CSIRO Agriculture and Food, Waite Campus, Adelaide; 2: High Resolution Plant Phenomics Centre, Canberra AGRICULTURE AND FOOD Matthew H. Siebers Email: [email protected] Expanding GRover and LiDAR use in vineyards: LiDAR technology appears to be a intriguing and useful tool in vineyards. But before this technology is ready for broader use, there is still a lot of development that needs to be done. For example, further LiDAR scanning needs to be carried out at multiple locations and on a number of varieties under varying environmental conditions to see if the destructive measurements and the scans continue to correlate. Additionally, computational analysis needs to be streamlined. The end aim is to have panels or plots identified and the analysis run automatically. Finally, LiDAR scans are not a solution for imaging every vine trait. GRover will soon be equipped with stereoscopic and hyperspectral cameras to explore yield quantification, bunch quality and disease monitoring. By eye it’s easy to distinguish stem and leaf or bunch material but there are no programs available where one can upload a scan of a vine and it will quantify the amount or volume of material present. The first attempts to correlate vine-scans with known amounts of plant growth have been through voxelization (Fig. 2c). Voxels recursively split the point cloud into regular sized cubes (for more information see [1]). The number of cubes can be counted within each scan and then compared to, say, the measured amount of leaf area. LiDAR scans correlate strongly with destructive measures of vine growth Figures 2d and 2e demonstrate how well voxel counts of LiDAR scans correlate to conventional, destructive measurements of vine growth. Both leaf area (2d) and pruning weight (2e) correlate strongly with voxel count (R 2 = 0.9). Encouragingly, voxelization is a simple computational procedure. With the help from specialists within CSIRO, development of more sophisticated algorithms should improve the richness of the data that can be extracted from point clouds. 2a 2b R² = 0.89 0E+00 1E+05 2E+05 3E+05 4E+05 5E+05 6E+05 0 2 4 6 8 10 12 14 16 18 Voxel Number Leaf area (m 2 ) per panel Voxel number Vs. Leaf area (m 2 ) per panel 2d 2c R² = 0.91 0E+00 1E+04 2E+04 3E+04 4E+04 5E+04 6E+04 7E+04 8E+04 0 2000 4000 6000 8000 Voxel Number Pruning weight (g) per panel Voxel # (pre minus post prune) Vs. Pruning weight 2e Figure 2: These figures show: (a) pictures of the destructive sampling, (b) 3D LiDAR scans, (c) voxelization of a point cloud and the correlations between the voxel counts and (d) leaf area and (e) pruning weight. The point clouds in 2b and 2c are of the same vines pictured in 2a. Pruning weights and leaf area calculations were done on whole panels of vine growth. Those panels contained two or three plants. 1: http://docs.pointclouds.org/trunk/group__octree.html I would like to thank Deidre Blackmore, James Buller, Matt Woodward, Bruce Henderson and Mick Schafer for their help in the field with destructive sampling. Funding was provided by CSIRO and the Australian Grape and Wine Authority. 1 6 5 4 3 2 7 3.2m 3.2m
Transcript of Application of LiDAR to determine vine growth in the vineyard · Application of LiDAR to determine...
FOR FURTHER INFORMATION
Figure 1: Above: The GRover frame in a vineyard. Numbers indicate: 1) The Sick LMS-400 LiDAR and GPS-IMU unit. 2) Hinge point for the collapsible mast. 3) Junction boxes routing power to the instruments. 4) Wheels driven by in-hub electric motors. 5) Wheel encoder that tracks linear distance 6) 48v batteries powering the wheels and a 24v instrument battery 7) Laptop for data collection. Left: GRover in a trailer with the mast retracted, contained in the boom-rest, ready for transportation to a vineyard.
REFERENCES ACKNOWLEDGEMENTS
Application of LiDAR to determine vine growth in the vineyard
The Grapevine Rover (GRover), how LiDAR is deployed in the vineyardLiDAR operates in a similar way as sonar. Instead of using sound, though, it uses light. The LiDAR used on GRover is the Sick LMS-400-200 (Sick, Germany; Fig. 1,1). An eye-safe line-scanner that emits laser pulses 300 times a second in a 70 degree field of view (Fig. 1,1). Those pulses reflect off of a target and are returned to the LiDAR sensor, providing information about the position of the return in 3D space and the reflectivity of the target’s surface.
Scans of vine growth are visualized and analysed using the open source 3D point cloud and mesh processing software Cloud Compare (http://www.danielgm.net/cc/). Figure 2a shows an example canopy and 2b an example of what a 3D point cloud of a vine looks like in Cloud Compare. The colours in 2b illustrate the differences in reflectivity of the different surfaces in the scan: leaf, wood, grass and even the plastic tag.
Developing relationships between LiDAR scans and vine growth. Field trials and computational analysis
After producing 3D point clouds, the challenge is to correlate scans with meaningful biological parameters. To that end, field work was conducted at the South Australian Research and Development Institute (Nuriootpa, SA), scanning vines before and after destructively quantifying the leaf area, cane weight, bunch number and cordon volume. The example canopy (Figure 2a) shown on the left had its leaves plucked off and the area of those leaves measured with a LiCOR 3000C. The vine on the right, with leaves intact, is for comparison.
The Grapevine Rover (GRover) is an adaptable proximal-sensing platform designed to provide rapid, high-quality information of plant performance in the field. GRover currently uses light radar (LiDAR) to produce precise (± 3mm), high resolution (300 scans/second), three-dimensional point clouds of vine growth. This poster explains: how LiDAR scans of vine growth have been collected in the vineyard, processed computationally and correlated with plant growth. Results from field trials have shown that LiDAR can accurately capture leaf area and pruning weight. Continued efforts will make data processing more automated and focus on capturing additional vine-traits with LiDAR and other instruments.
Matthew Siebers 1, Mark R. Thomas1, Everard Edwards1, Jose Jimenez Berni2, Rob Walker1
1: CSIRO Agriculture and Food, Waite Campus, Adelaide; 2: High Resolution Plant Phenomics Centre, Canberra
AGRICULTURE AND FOOD
Matthew H. Siebers
Email: [email protected]
Figure 2: Figures 2a to 2e illustrate the general
Expanding GRover and LiDAR use in vineyards: LiDAR technology appears to be a intriguing and useful tool in vineyards. But before this technology is ready for broader use, there is still a lot of development that needs to be done.
For example, further LiDAR scanning needs to be carried out at multiple locations and on a number of varieties under varying environmental conditions to see if the destructive measurements and the scans continue to correlate.
Additionally, computational analysis needs to be streamlined. The end aim is to have panels or plots identified and the analysis run automatically.
Finally, LiDAR scans are not a solution for imaging every vine trait. GRover will soon be equipped with stereoscopic and hyperspectral cameras to explore yield quantification, bunch quality and disease monitoring.
By eye it’s easy to distinguish stem and leaf or bunch material but there are no programs available where one can upload a scan of a vine and it will quantify the amount or volume of material present. The first attempts to correlate vine-scans with known amounts of plant growth have been through voxelization (Fig. 2c).
Voxels recursively split the point cloud into regular sized cubes (for more information see [1]). The number of cubes can be counted within each scan and then compared to, say, the measured amount of leaf area.
LiDAR scans correlate strongly with destructive measures of vine growth
Figures 2d and 2e demonstrate how well voxel counts of LiDAR scans correlate to conventional, destructive measurements of vine growth. Both leaf area (2d) and pruning weight (2e) correlate strongly with voxel count (R2 = 0.9).
Encouragingly, voxelization is a simple computational procedure. With the help from specialists within CSIRO, development of more sophisticated algorithms should improve the richness of the data that can be extracted from point clouds.
2a
2b
R² = 0.89
0E+00
1E+05
2E+05
3E+05
4E+05
5E+05
6E+05
0 2 4 6 8 10 12 14 16 18
Vo
xel N
um
ber
Leaf area (m2) per panel
Voxel number Vs. Leaf area (m2) per panel2d
2c
R² = 0.91
0E+00
1E+04
2E+04
3E+04
4E+04
5E+04
6E+04
7E+04
8E+04
0 2000 4000 6000 8000
Vo
xel N
um
ber
Pruning weight (g) per panel
Voxel # (pre minus post prune) Vs. Pruning weight2eFigure 2: These figures show: (a) pictures of the destructive sampling, (b) 3D LiDAR scans, (c) voxelization of a point cloud and the correlations between the voxel counts and (d) leaf area and (e) pruning weight. The point clouds in 2b and 2c are of the same vines pictured in 2a. Pruning weights and leaf area calculations were done on whole panels of vine growth. Those panels contained two or three plants.
1: http://docs.pointclouds.org/trunk/group__octree.html I would like to thank Deidre Blackmore, James Buller, Matt Woodward, Bruce Henderson and Mick Schafer for their help in the field with destructive sampling. Funding was provided by CSIRO and the Australian Grape and Wine Authority.
1
65
4
3
2
7
3.2m
3.2m
