Three-Dimensional Vehicle Detection Algorithm Based on Binocular Vision

Jiexiao Yu; Meiqi Zhang; Yuting Su

doi:10.3788/LOP202158.0215004

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 2 >Page 0215004 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 2, 0215004 (2021)

Three-Dimensional Vehicle Detection Algorithm Based on Binocular Vision

Jiexiao Yu, Meiqi Zhang^*, and Yuting Su

Author Affiliations

School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China

show less

DOI: 10.3788/LOP202158.0215004 Cite this Article Set citation alerts

Jiexiao Yu, Meiqi Zhang, Yuting Su. Three-Dimensional Vehicle Detection Algorithm Based on Binocular Vision[J]. Laser & Optoelectronics Progress, 2021, 58(2): 0215004 Copy Citation Text

show less

Fig. 1. Flow chart of Stereo R-CNN

Download full size

Fig. 2. Stages 5,6 of DetNet

Download full size

Fig. 3. Flow chart of Stereo R-CNN left and right consistency network

Download full size

Fig. 4. Detection results of different algorithms. (a) Stereo R-CNN; (b) our algorithm

Download full size

Bottleneck block	Detail structure
A	keneral size 1×1 conv, 256 channelskeneral size 3×3 conv, dilate2, 256 channelskeneral size 1×1 conv, 256 channels
B	keneral size 1×1 conv, 256 channelskeneral size 3×3 conv, dilate2, 256 channels + keneral size 1×1convkeneral size 1×1 conv, 256 channels

Table 1. Detail structure of bottleneck blocks A and B

Algorithm	Left			Right			Stereo
Algorithm	Easy	Mode	Hard	Easy	Mode	Hard	Easy	Mode	Hard
Faster R-CNN	98.57	89.01	71.54	--	--	--	--	--	--
Stereo R-CNN	98.73	88.48	71.26	98.71	88.50	71.28	98.53	88.27	71.14
DetNet only	98.01	88.52	73.44	98.80	88.32	72.67	98.63	88.44	73.16
Ours	98.50	88.92	73.63	98.34	88.25	73.10	98.44	88.76	73.18

Table 2. Average precision of 2D detection unit: %

Algorithm	Bird’s eye view			3D box
Algorithm	Easy	Mode	Hard	Easy	Mode	Hard
3DOP^[5]	55.04	41.25	34.55	46.04	34.63	30.09
Stereo R-CNN	87.13	74.11	58.93	85.84	66.28	57.24
DetNet only	87.02	74.26	58.98	85.12	65.85	57.81
Ours	88.63	75.90	59.46	86.66	66.39	58.18

Table 3. Average precision of 3D detection (IoU is 0.5) unit: %

Algorithm	Bird’s eye view			3D box
Algorithm	Easy	Mode	Hard	Easy	Mode	Hard
3DOP^[5]	12.63	9.49	7.59	6.55	5.07	4.10
Stereo R-CNN	68.50	48.30	41.47	54.11	36.69	31.07
DetNet only	67.42	49.21	42.68	52.12	34.85	31.81
Ours	69.63	49.90	43.46	55.66	36.39	32.14

Table 4. Average precision of 3D detection (IoU is 0.7) unit: %

Jiexiao Yu, Meiqi Zhang, Yuting Su. Three-Dimensional Vehicle Detection Algorithm Based on Binocular Vision[J]. Laser & Optoelectronics Progress, 2021, 58(2): 0215004

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information