Laser & Optoelectronics Progress, Volume. 61, Issue 24, 2415001(2024)
Photon-Counting Lidar Point Cloud Filtering Using a Backpropagation Neural Network
Figures & Tables(11)
Fig. 2. Evenly distributed simulated point cloud noise. (a) Simulated noise photons; (b) statistical histogram of noise photons
Fig. 3. Simulated point cloud signals with a Gaussian distribution. (a) Simulated signal photons; (b) statistical histogram of signal photons
Fig. 4. Distance-based point cloud data discrimination. (a) Signal photon; (b) noise photon
Fig. 5. Point cloud data discrimination based on point cloud density. (a) Signal photon; (b) noise photon
Fig. 6. Analysis of denoising results. (a) Simulated data (1 MHz/0.5); (b) signal extraction result (1 MHz/0.5); (c) simulated data (5 MHz/0.05); (d) signal extraction result (5 MHz/0.05)
Fig. 7. Denoising effect in a scene with high background noise. (a) Simulated noise photons; (b) simulated signal photons; (c) noise mixing with signal; (d) signal extraction result
Fig. 8. Denoising results of UAV-borne radar point cloud data. (a) 2D point cloud of the original data profile; (b) signal extraction result
Fig. 9. ICESat-2 surface data denoising results. (a) 2D point cloud of the original data profile; (b) signal extraction result
Table 1. Eigenvalue description
View tableTable 1. Eigenvalue description
Feature Description DBSCAN_circle_1 Number of photon points in DBSCAN clustering radius R （R =1 m） DBSCAN_circle_3 Number of photon points in DBSCAN clustering radius R （R =3 m） DBSCAN_circle_5 Number of photon points in DBSCAN clustering radius R （R =5 m） DBSCAN_circle_10 Number of photon points in DBSCAN clustering radius R （R =10 m） KNN_dist_10 K of the photon point is the sum of the approaching distances （K =10） KNN_dist_20 K of the photon point is the sum of the approaching distances （K =20） KNN_dist_50 K of the photon point is the sum of the approaching distances （K =50） KNN_dist_100 K of the photon point is the sum of the approaching distances （K =100）
Table 2. Comparison of algorithm accuracy in different scenarios
View tableTable 2. Comparison of algorithm accuracy in different scenarios
Noise rate /MHz p DBSCAN KNN BP neural network R P F R P F R P F 1 0.05 0.9958 0.7411 0.8508 0.9986 0.7411 0.8508 0.9140 0.9191 0.9166 0.1 0.9969 0.8738 0.9327 1.0000 0.8738 0.9327 0.9901 0.9672 0.9329 0.5 1.0000 0.9697 0.9847 1.0000 0.9697 0.9847 0.9994 0.9911 0.9952 2.5 0.05 0.9517 0.6720 0.8008 0.9905 0.6720 0.8008 0.9375 0.8199 0.8747 0.1 0.9855 0.8310 0.9072 0.9987 0.8310 0.9072 0.9933 0.9149 0.9525 0.5 1.0000 0.9608 0.9800 1.0000 0.9608 0.9800 1.0000 0.9758 0.9877 5 0.05 0.7905 0.4081 0.5694 0.9415 0.6476 0.7861 0.8551 0.6269 0.7235 0.1 0.9943 0.7706 0.8681 0.9939 0.7874 0.8681 0.9986 0.7326 0.8452 0.5 1.0000 0.9406 0.9695 1.0000 0.9406 0.9695 1.0000 0.9558 0.9773
Tools
Get Citation
Copy Citation Text
Shiao Yu, Wei Kong, Rujia Ma, Genghua Huang. Photon-Counting Lidar Point Cloud Filtering Using a Backpropagation Neural Network[J]. Laser & Optoelectronics Progress, 2024, 61(24): 2415001
Paper Information
Category: Machine Vision
Received: Mar. 4, 2024
Accepted: Apr. 25, 2024
Published Online: Dec. 19, 2024
The Author Email:
CSTR:32186.14.LOP240793
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20