University of Dayton Research Institute, Sensor and Software Systems Division, 300 College Park, Dayton, Ohio 45469-7640, USA (edward.watson@udri.udayton.edu)
It has been shown that point separations as a feature derived from point clouds can be used to discriminate between two objects of similar class. Here we show that the same feature derived from sparse point clouds can maintain significant discrimination capability. Using the point-separation feature, templates created from random realizations of a point cloud are developed for several vehicles. The templates are then used in two-class discrimination tests. The point-separation feature is shown to produce reliable discrimination using a log-likelihood ratio between two objects.
No available data were generated in the presented research.
Cited By
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
Results of Discrimination Comparison between Two Vehiclesa
Comp Test
Honda CRZ
Hyundai Sonata
Nissan Sentra
Mazda SUV
Dodge Ram
Forklift
Honda CRZ
100%
100%
91%
92%
100%
100%
Hyundai Sonata
98%
100%
82%
84%
100%
100%
Nissan Sentra
91%
76%
100%
68%
100%
100%
Mazda SUV
95%
88%
78%
100%
100%
100%
Dodge Ram
100%
100%
100%
100%
100%
100%
Forklift
100%
96%
93%
97%
100%
100%
The left-hand column lists the test vehicle, and the top row lists the comparison vehicle. The numbers represent the percentage of time that the test vehicles are correctly chosen over the comparison vehicle. For this example, the pixel width was set to 5 cm at the object, and the mean number of photocounts per image was set to 2400.
Table 2.
Same as Table 1 Except the Pixel Width Was Set to 20 cm at the Object and the Mean Number of Photocounts Per Image Was Set to 150
The middle column shows the average percentage of correct discrimination for the vehicle in the left column. The column on the right shows the percentage of time the vehicle was incorrectly selected.
The middle column shows the average percentage of correct discrimination for the vehicle in the left column. The column on the right shows the percentage of time the vehicle was incorrectly selected.
Table 5.
Jensen–Shannon Divergence When Comparing the Point-Separation Template for the Vehicle in the Left-Hand Column with the Vehicle along the Top Rowa
JSD
Honda CRZ
Hyundai Sonata
Nissan Sentra
Mazda SUV
Dodge Ram
Forklift
Honda CRZ
0
0.023
0.013
0.014
0.097
0.035
Hyundai Sonata
0.023
0
0.003
0.003
0.038
0.012
Nissan Sentra
0.013
0.003
0
0.00
0.052
0.013
Mazda SUV
0.014
0.003
0.00
0
0.052
0.013
Dodge Ram
0.097
0.038
0.052
0.052
0
0.037
Forklift
0.035
0.012
0.013
0.013
0.037
0
Note that ${\rm JSD} = 0$ when the distributions are identical. This chart is for the templates created using a pixel width of 5 cm and a mean photocount of 2400.
Table 6.
Same as Table 5 Except This Chart Is for the Templates Created Using a Pixel Width of 20 cm and a Mean Photocount of 150
JSD
Honda CRZ
Hyundai Sonata
Nissan Sentra
Mazda SUV
Dodge Ram
Forklift
Honda CRZ
0
0.018
0.016
0.017
0.094
0.049
Hyundai Sonata
0.018
0
0.001
0.002
0.041
0.015
Nissan Sentra
0.016
0.001
0
0.000
0.047
0.015
Mazda SUV
0.017
0.002
0.000
0
0.046
0.014
Dodge Ram
0.094
0.041
0.047
0.046
0
0.023
Forklift
0.049
0.015
0.015
0.014
0.023
0
Table 7.
Discrimination Performance Using the Same Pixel Size as in Table 1, but with Mean Photocount Set an Order of Magnitude Smaller, to 240
Comp Test
Honda CRZ
Hyundai Sonata
Nissan Sentra
Mazda SUV
Dodge Ram
Forklift
Honda CRZ
100%
94%
89%
91%
100%
100%
Hyundai Sonata
95%
100%
77%
81%
100%
100%
Nissan Sentra
86%
68%
100%
87%
100%
100%
Mazda SUV
93%
69%
32%
100%
100%
99%
Dodge Ram
100%
98%
99%
100%
100%
99%
Forklift
100%
89%
92%
93%
100%
100%
Tables (7)
Table 1.
Results of Discrimination Comparison between Two Vehiclesa
Comp Test
Honda CRZ
Hyundai Sonata
Nissan Sentra
Mazda SUV
Dodge Ram
Forklift
Honda CRZ
100%
100%
91%
92%
100%
100%
Hyundai Sonata
98%
100%
82%
84%
100%
100%
Nissan Sentra
91%
76%
100%
68%
100%
100%
Mazda SUV
95%
88%
78%
100%
100%
100%
Dodge Ram
100%
100%
100%
100%
100%
100%
Forklift
100%
96%
93%
97%
100%
100%
The left-hand column lists the test vehicle, and the top row lists the comparison vehicle. The numbers represent the percentage of time that the test vehicles are correctly chosen over the comparison vehicle. For this example, the pixel width was set to 5 cm at the object, and the mean number of photocounts per image was set to 2400.
Table 2.
Same as Table 1 Except the Pixel Width Was Set to 20 cm at the Object and the Mean Number of Photocounts Per Image Was Set to 150
The middle column shows the average percentage of correct discrimination for the vehicle in the left column. The column on the right shows the percentage of time the vehicle was incorrectly selected.
The middle column shows the average percentage of correct discrimination for the vehicle in the left column. The column on the right shows the percentage of time the vehicle was incorrectly selected.
Table 5.
Jensen–Shannon Divergence When Comparing the Point-Separation Template for the Vehicle in the Left-Hand Column with the Vehicle along the Top Rowa
JSD
Honda CRZ
Hyundai Sonata
Nissan Sentra
Mazda SUV
Dodge Ram
Forklift
Honda CRZ
0
0.023
0.013
0.014
0.097
0.035
Hyundai Sonata
0.023
0
0.003
0.003
0.038
0.012
Nissan Sentra
0.013
0.003
0
0.00
0.052
0.013
Mazda SUV
0.014
0.003
0.00
0
0.052
0.013
Dodge Ram
0.097
0.038
0.052
0.052
0
0.037
Forklift
0.035
0.012
0.013
0.013
0.037
0
Note that ${\rm JSD} = 0$ when the distributions are identical. This chart is for the templates created using a pixel width of 5 cm and a mean photocount of 2400.
Table 6.
Same as Table 5 Except This Chart Is for the Templates Created Using a Pixel Width of 20 cm and a Mean Photocount of 150
JSD
Honda CRZ
Hyundai Sonata
Nissan Sentra
Mazda SUV
Dodge Ram
Forklift
Honda CRZ
0
0.018
0.016
0.017
0.094
0.049
Hyundai Sonata
0.018
0
0.001
0.002
0.041
0.015
Nissan Sentra
0.016
0.001
0
0.000
0.047
0.015
Mazda SUV
0.017
0.002
0.000
0
0.046
0.014
Dodge Ram
0.094
0.041
0.047
0.046
0
0.023
Forklift
0.049
0.015
0.015
0.014
0.023
0
Table 7.
Discrimination Performance Using the Same Pixel Size as in Table 1, but with Mean Photocount Set an Order of Magnitude Smaller, to 240