Multiscale Monocular Three-Dimensional Object Detection Algorithm Incorporating Instance Depth

Fengsui Wang; Lei Xiong; Yaping Qian

doi:10.3788/LOP222627

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 16 >Page 1612002 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 16, 1612002 (2023)

Multiscale Monocular Three-Dimensional Object Detection Algorithm Incorporating Instance Depth

Fengsui Wang^1,2,3,*, Lei Xiong^1,2,3, and Yaping Qian^1,2,3

Author Affiliations

¹School of Electrical Engineering, Anhui Polytechnic University, Wuhu 241000, Anhui, China

²Anhui Key Laboratory of Detection Technology and Energy Saving Devices, Wuhu 241000, Anhui, China

³Key Laboratory of Advanced Perception and Intelligent Control of High-End Equipment, Ministry of Education, Wuhu 241000, Anhui, China

show less

DOI: 10.3788/LOP222627 Cite this Article Set citation alerts

Fengsui Wang, Lei Xiong, Yaping Qian. Multiscale Monocular Three-Dimensional Object Detection Algorithm Incorporating Instance Depth[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1612002 Copy Citation Text

show less

Fig. 1. Instance depth learning module

Download full size

Fig. 2. Multiscale sensing module

Download full size

Fig. 3. Network structural diagram

Download full size

Fig. 4. Visualization results of KITTI

Download full size

Method	Extra data	AP40（3D@R_IOU≥0.7）			AP40（BEV@R_IOU≥0.7）
Method	Extra data	Easy	Moderate	Hard	Easy	Moderate	Hard
AM3D^［24］	Depth	16.50	10.74	9.52	25.03	17.32	14.91
PatchNet^［4］	Depth	15.68	11.12	10.17	22.97	16.86	14.97
DDMP-3D^［5］	Depth	19.71	12.78	9.80	28.08	17.89	13.44
Reference［25］	Depth	20.28	13.12	9.56
Kinematic3D^［26］	Multi-frames	19.07	12.72	9.17	26.69	17.52	13.10
CaDDN^［27］	LiDAR	19.17	13.41	11.46	27.94	18.91	17.19
MonoRUn^［28］	LiDAR	19.65	12.30	10.58	27.94	17.34	15.24
MonoGRNet^［11］	None	9.61	5.74	4.25	18.19	11.17	8.73
MonoDIS^［23］	None	10.37	7.94	6.40	17.23	13.19	11.12
Reference［29］	None	20.89	14.49	12.19	29.57	20.77	17.88
MonoPair^［6］	None	13.04	9.99	8.65	19.28	14.83	12.89
FADNet^［30］	None	16.37	9.92	8.05	23.00	14.22	12.56
MonoDLE^［9］	None	17.23	12.26	10.29	24.79	18.89	16.00
MonoGround^［31］	None	19.48	14.36	12.62	30.07	20.47	17.74
MonoFlex^［7］	None	19.94	13.89	12.07	28.23	19.75	16.89
MonoEF^［32］	None	21.29	13.87	11.71	29.03	19.70	17.26
Reference［33］	None	21.65	13.25	9.91	29.81	17.98	13.08
GUPNet^［8］	None	22.26	15.02	13.12	30.29	21.19	18.20
MonoCon^［34］	None	22.50	16.49	13.95	31.12	22.10	19.00
Proposed method	None	22.50	16.19	13.49	32.44	22.97	19.82
Improvement	Depth	+2.22	+3.07	+3.32	+4.36	+5.08	+4.85
	Multi-frames	+3.43	+3.47	+4.78	+5.75	+5.45	+6.72
	LiDAR	+2.85	+2.78	+2.03	+4.5	+4.06	+2.63
	None	+0	-0.3	-0.46	+1.32	+0.87	+0.82

Table 1. Performance of the Car category on the KITTI test set

Method	AP40 /%（3D@R_IOU=0.7）			AP40 /%（BEV@R_IOU=0.7）			AP40 /%（3D@R_IOU=0.5）			AP40 /%（BEV@R_IOU=0.5）			Runtime /ms
Method	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard	Runtime /ms
Improvement	+5.14	+4.13	+3.31	+5.25	+3.98	+3.03	+8.81	+4.07	+4.73	+7.41	+4.49	+4.19
CenterNet^［10］	0.60	0.66	0.77	3.46	3.31	3.21	20.00	17.50	15.57	34.36	27.91	24.65
MonoGRNet	11.90	7.56	5.76	19.72	12.81	10.15	47.59	32.28	25.50	48.53	35.94	28.59	60
MonoDIS	11.06	7.60	6.37	18.45	12.58	10.66
M3D-RPN	14.53	11.07	8.65	20.85	15.62	11.88	48.53	35.94	28.59	53.35	39.60	31.76	161
MonoPair	16.28	12.30	10.42	24.12	18.17	15.76	55.38	42.39	37.99	61.06	47.63	41.92	57
MonoDLE	17.45	13.66	11.68	24.97	19.33	17.01	55.41	43.42	37.81	60.73	46.87	41.89	40
Proposed method	22.59	17.79	14.99	30.22	23.31	20.04	64.22	47.49	42.72	68.47	52.12	46.11	45

Table 2. Performance of the Car category on the KITTI validation set

Method	AP40（3D）			AP40（BEV）
Method	Easy	Moderate	Hard	Easy	Moderate	Hard
Improvement	+5.14	+4.13	+3.31	+5.25	+3.98	+3.03
Baseline	17.45	13.66	11.68	24.97	19.33	17.01
+ASPP	18.04	14.72	12.47	25.60	20.84	18.20
+PSP	18.98	14.73	12.38	25.61	20.75	18.06
+MSS	20.48	15.89	14.06	28.58	22.41	19.46
+IDLM	21.76	16.19	14.24	28.71	22.44	19.43
+MSS+IDLM	22.59	17.79	14.99	30.22	23.31	20.04

Table 3. Adding Performance Comparison of Different Modules on KITTI validation set

Fengsui Wang, Lei Xiong, Yaping Qian. Multiscale Monocular Three-Dimensional Object Detection Algorithm Incorporating Instance Depth[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1612002

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享