Chinese Journal of Liquid Crystals and Displays, Volume. 39, Issue 6, 801(2024)
Improved YOLOx-based vehicle detection method for low light environment
Fig. 1. Image enhancement in normal light（a）Original image；（b）Enhanced image.
Fig. 2. Image enhancement effect under extremely weak light（a）Original image；（b）Enhanced image.
Fig. 3. Schematic diagram of CSPLayer（a），HorBlock（b）and gnConv（c）.
Fig. 4. Structure of Swin-Transformer-based backbone feature extraction network
Fig. 7. Thermogram results of ablation experiment SCORE-CAM.（a）Original image；（b）YOLOx；（c）YOLOx+Swin-Transformer；（d）YOLOx+Swin-Transformer+gnConv；（e）YOLOx+Swin-Transformer+gnConv+CBAM.
Fig. 8. Vehicle detection results when there are many vehicle targets.（a）Original image；（b）Image enhancement rendering；（c）Basic model inspection rendering；（d）Detection effect diagram of the method in this article.
Fig. 9. Detection effect of models under normal lighting.（a）Original image；（b）Image enhancement rendering；（c）Basic model inspection rendering；（d）Detection effect diagram of the method in this article.
Fig. 10. Detection effect of models under low lighting.（a）Original image；（b）Image enhancement rendering；（c）Basic model inspection rendering；（d）Detection effect diagram of the method in this article.
Table 1. Detection performance comparison results of different algorithms
View tableView in ArticleTable 1. Detection performance comparison results of different algorithms
Dataset Model Backbone mAP@（IOU=0.5） Time/ms UA-DETTUN SSD VGG16 0.782 17 Cascade R-CNN ResNet-101· 0.798 49 YOLOv3 Darknet-53 0.876 24 YOLOx Darknet-53 0.896 15 Mask-RCNN Swin-T 0.924 73 Swin-YOLO（ours） Swin-ransformer 0.961 67
Table 2. Comparative results of ablation experiments of the proposed method
View tableView in ArticleTable 2. Comparative results of ablation experiments of the proposed method
Methods mAP@（IOU=0.5） AP@（IOU=0.5） Car Bus Van Others YOLOx 0.896 0.969 0.979 0.902 0.731 YOLOx+image enhancement 0.912 0.963 0.982 0.903 0.787 YOLOx+gnConv 0.906 0.968 0.979 0.892 0.784 YOLOx+CBAM 0.905 0.955 0.971 0.869 0.823 YOLOx+Swin-Transformer 0.924 0.960 0.975 0.886 0.876 YOLOx+Swin-Transformer+gnConv 0.948 0.964 0.979 0.909 0.936 YOLOx+Swin-Transformer+gnConv+CBAM 0.961 0.975 0.982 0.951 0.936
Get Citation
Copy Citation Text
Xiaohan YANG, Jun WANG, Zhongxing DUAN, Leilei HUI. Improved YOLOx-based vehicle detection method for low light environment[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(6): 801
Category: Research Articles
Received: May. 6, 2023
Accepted: --
Published Online: Jul. 30, 2024
The Author Email: Zhongxing DUAN (zhx_duan@xauat.edu.cn)
