Single Object Tracking of LiDAR Point Cloud Combined with Auxiliary Deep Neural Network

Xiaoyu Zhou; Ling Wang; Yanxin Ma; Peibo Chen

doi:10.3788/CJL202148.2110001

Chinese Journal of Lasers, Volume. 48, Issue 21, 2110001(2021)

Single Object Tracking of LiDAR Point Cloud Combined with Auxiliary Deep Neural Network

Xiaoyu Zhou¹, Ling Wang^1、*, Yanxin Ma², and Peibo Chen¹

Author Affiliations

¹College of Electronic Science, National University of Defense Technology, Changsha, Hunan 410073, China

²College of Meteorology and Oceanography, National University of Defense Technology, Changsha, Hunan 410073, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(18)

Fig. 1. Structure schematic diagram of P2B algorithm network

Download full size

Fig. 2. Schematic diagram of feature augmentation

Download full size

Fig. 3. Schematic diagram of proposal generation

Download full size

Fig. 4. Structure schematic diagram of proposed algorithm network

Download full size

Fig. 5. Flow chart of network training

Download full size

Fig. 6. Structure schematic diagram of feature propagation layer

Download full size

Fig. 7. Tracking results of class Car.(a)--(h) 1st-8th frames

Download full size

Fig. 8. Tracking results of class Pedestrian. (a)-(h) 1st-8th frames

Download full size

Fig. 9. Tracking results of class Cyclist. (a)-(h) 1st-8th frames

Download full size

Fig. 10. Tracking results of class Van. (a)-(h) 1st-8th frames

Download full size

Fig. 11. Attached segmentation task results of class Car. (a)(e) 1st frame; (b)(f) 2nd frame; (c)(g) 3rd frame; (d)(h) 4th frame

Download full size

Fig. 12. Attached segmentation task results of class Pedestrian. (a)(e) 1st frame; (b)(f) 2nd frame; (c)(g) 3rd frame; (d)(h) 4th frame

Download full size

Fig. 13. Attached segmentation task results of class Cyclist. (a)(e) 1st frame; (b)(f) 2nd frame; (c)(g) 3rd frame; (d)(h) 4th frame

Download full size

Fig. 14. Attached segmentation task results of class Van. (a)(e) 1st frame; (b)(f) 2nd frame; (c)(g) 3rd frame; (d)(h) 4th frame

Download full size

Table 1. Comparison of tracking performance of different algorithms for various targets unit:%

View table

Table 1. Comparison of tracking performance of different algorithms for various targets unit:%

Parameter	Method	Car 6424	Pedestrian 6088	Cyclist 308	Van 1248	Mean 14068
R_S	SC3D	41.30	18.20	41.50	40.40	31.20
	P2B	56.20	28.70	32.10	40.80	42.40
	Proposed algorithm	56.98	28.40	30.44	48.60	43.29
R_P	SC3D	57.90	37.80	70.40	47.00	48.50
	P2B	72.80	49.60	44.70	48.40	60.00
	Proposed algorithm	73.73	52.40	41.67	59.25	62.51

Table 2. Performance comparison under different number of search area point clouds S and template point clouds T unit:%

View table

Table 2. Performance comparison under different number of search area point clouds S and template point clouds T unit:%

Parameter	S	T	Car 6424	Pedestrian 6088	Cyclist 308	Van 1248	Mean 14068
R_S	512	256	56.63	30.48	32.13	37.89	43.11
	512	512	57.55	36.87	32.82	44.92	46.94
	1024	512	56.98	28.40	30.44	48.60	43.29
R_P	512	256	72.60	54.14	42.62	46.31	61.62
	512	512	73.22	66.11	44.53	54.22	67.83
	1024	512	73.73	52.40	41.67	59.25	62.51

Table 3. Number of parameters of each network layer of auxiliary network

View table

Table 3. Number of parameters of each network layer of auxiliary network

Module	Layer name	Number of inputchannels	Number of outputchannels	Kernel size	Stride size	Parameter size
Aux_module	3-NN	256	256			0
	Interpolation	128	256			0
	Conv2d	512	256	(1, 1)	(1, 1)	131072
	BatchNorm2d	256	256			512
	ReLU	256	256			0
	3-NN	512	512			0
	Interpolation	256	512			0
	Conv2d	384	256	(1, 1)	(1, 1)	98304
	BatchNorm2d	256	256			512
	ReLU	256	256			0
	3-NN	1024	1024			0
	Interpolation	512	1024			0
	Conv2d	256	256	(1, 1)	(1, 1)	65536
	BatchNorm2d	256	256			512
	ReLU	256	256			0
Aux_regreesion	Linear	256	3			768
Aux_segmentation	Linear	256	1			256
Total						297472

Table 4. Average speed comparison of various methods
View table
Table 4. Average speed comparison of various methods
Method SC3D P2B Proposedalgorithm
Speed /(frame·s^-1) 2.2 55.6 55.6

Tools

Get Citation

Copy Citation Text

Xiaoyu Zhou, Ling Wang, Yanxin Ma, Peibo Chen. Single Object Tracking of LiDAR Point Cloud Combined with Auxiliary Deep Neural Network[J]. Chinese Journal of Lasers, 2021, 48(21): 2110001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Received: Feb. 22, 2021

Accepted: Apr. 15, 2021

Published Online: Oct. 18, 2021

The Author Email: Wang Ling (wangling@nudt.edu.cn)

DOI:10.3788/CJL202148.2110001

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of tracking performance of different algorithms for various targets unit:%

Table 1. Comparison of tracking performance of different algorithms for various targets unit:%

Table 2. Performance comparison under different number of search area point clouds S and template point clouds T unit:%

Table 2. Performance comparison under different number of search area point clouds S and template point clouds T unit:%

Table 3. Number of parameters of each network layer of auxiliary network

Table 3. Number of parameters of each network layer of auxiliary network

Table 4. Average speed comparison of various methods

Table 4. Average speed comparison of various methods