Lane Detection Based on a Lightweight Convolutional Neural Network

Jie Hu; Zongquan Xiong; Wencai Xu; Kai Cao; Ruoyu Lu

doi:10.3788/LOP202259.1015012

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 10 >Page 1015012 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 10, 1015012 (2022)

Lane Detection Based on a Lightweight Convolutional Neural Network

Jie Hu^1、2、3, Zongquan Xiong^{1、2、3、*}, Wencai Xu^1、2、3, Kai Cao⁴, and Ruoyu Lu⁴

Author Affiliations

¹Hubei Key Laboratory of Modern Auto Parts Technology, School of Automotive Engineering, Wuhan University of Technology, Wuhan 430070, Hubei , China

²Auto Parts Technology Hubei Collaborative Innovation Center, School of Automotive Engineering, Wuhan University of Technology, Wuhan 430070, Hubei , China

³Hubei Technology Research Center of New Energy and Intelligent Connected Vehicle Engineering, Wuhan 430070, Hubei , China

⁴Dongfeng Yuexiang Technology Co., Ltd., Wuhan 430058, Hubei , China

show less

DOI: 10.3788/LOP202259.1015012 Cite this Article Set citation alerts

Jie Hu, Zongquan Xiong, Wencai Xu, Kai Cao, Ruoyu Lu. Lane Detection Based on a Lightweight Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2022, 59(10): 1015012 Copy Citation Text

show less

Fig. 1. ERFNet-LW network structure

Download full size

Fig. 2. Comparison of core module structure. (a) Non-bottleneck-1D; (b) shuffle-unit

Download full size

Fig. 3. Downsampling structure. (a) Initial downsampling;

Download full size

Fig. 4. Feature fusion module

Download full size

Fig. 5. Dataset distribution

Download full size

Fig. 6. Evaluation based on IoU. (a) Scene 1; (b) scene 2

Download full size

Fig. 7. Visualization of CULane test sets. (a) Original image; (b) predicted results

Download full size

Layer	Type
1	Downsampling block
2	Downsampling block
3-7	5 × Non-bottleneck-1D
8	Downsampling block
9	Non-bottleneck-1D （dilated=2）
10	Non-bottleneck-1D （dilated=4）
11	Non-bottleneck-1D （dilated=8）
12	Non-bottleneck-1D （dilated=16）
13	Non-bottleneck-1D （dilated=2）
14	Non-bottleneck-1D （dilated=4）
15	Non-bottleneck-1D （dilated=8）
16	Non-bottleneck-1D （dilated=16）
17	Deconvolution
18-19	2 × Non-bottleneck-1D
20	Deconvolution
21-22	2 × Non-bottleneck-1D
23	Deconvolution

Table 1. Network architecture of ERFNet

Algorithm	Parameter /MB	GFLOPs	MAC /MB
ERFNet	2.67	11.47	343.30
ERFNet-shuffle	1.02	2.03	167.52

Table 2. Comparison of model indicators

	Normal	Crowded	Night	No-line	Shadow	Arrow	Dazzlelight	Curve	Crossroad	Total
Before	91.8	72.1	69.5	46.4	67.5	86.9	65.8	67.7	2526	73.9
After	91.7	72.0	69.4	46.3	67.4	86.7	65.6	67.7	2526	73.8

Table 3. Comparison of F1-measure on the CULane dataset with IoU of 0.5 before and after the downsampling optimization of the model

Category	Res50-Seg	SCNN	Enet-SAD	ERFNet	ERFNet-LW
Total	66.7	71.6	70.8	73.1	73.9
Normal	87.4	90.6	90.1	91.5	91.8
Crowded	64.1	69.7	68.8	71.6	72.1
Night	60.6	66.1	66.0	67.1	69.5
No-line	38.1	43.4	41.6	45.1	46.4
Shadow	60.7	66.9	65.9	71.3	67.5
Arrow	79.0	84.1	84.0	87.2	86.9
Dazzlelight	54.1	58.5	60.2	66.0	65.8
Curve	59.8	64.4	65.7	66.3	67.7
Crossroad	2505	1990	1998	2199	2526

Table 4. Comparison of F1-measure for different algorithms on CULane dataset with IoU of 0.5

Method	Parameter /MB	Runtime /ms
Res50-Seg	-	-
SCNN	20.72	133.5
Enet-SAD	0.98	13.4
ERFNet	2.68	14.0
ERFNet-LW	1.23	12.2

Table 5. Comparison of parameters and runtime of different algorithms

Jie Hu, Zongquan Xiong, Wencai Xu, Kai Cao, Ruoyu Lu. Lane Detection Based on a Lightweight Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2022, 59(10): 1015012

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information