Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 5, 680(2023)
Road object detection algorithm based on improved YOLOv5s
Figures & Tables(14)
Fig. 3. Overall network structure diagram of YOLOv5s after improvement
Fig. 10. （a）and（b）comparison diagram of detection effect under the scene
Table 1. Ablation experimental results of the algorithm on KITTI dataset
View tableView in ArticleTable 1. Ablation experimental results of the algorithm on KITTI dataset
网络模型 MobileNet-V3轻量化主干网络 Bi-FPN特征提取 CIoU损失优化 模型大小/MB P/% R/% mAP/% 每帧推理时间/ms YOLOv5s 46.0 90.51 91.70 79.60 34 A √ 14.1 88.67 90.3 76.93 30 B √ 44.0 91.93 92.02 81.28 35 C √ 45.3 91.50 92.60 80.56 35 D √ √ 15.7 92.71 93.03 82.63 31 E √ √ 15.3 92.47 93.51 83.10 31 F √ √ 44.7 94.03 94.19 83.52 36 本文算法 √ √ √ 13.6 95.12 96.03 85.40 28
Table 2. Ablation experimental results of the algorithm on the BDD dataset
View tableView in ArticleTable 2. Ablation experimental results of the algorithm on the BDD dataset
网络模型 MobileNet-V3轻量化主干网络 Bi-FPN特征提取 CIoU损失优化 模型大小/MB P/% R/% mAP/% 每帧推理时间/ms YOLOv5s 50.5 61.53 62.9 48.75 41 A √ 18.0 55.71 56.07 46.41 38 B √ 49.7 62.45 63.00 50.88 43 C √ 50.3 62.07 62.83 50.42 42 D √ √ 19.2 63.48 63.24 52.75 37 E √ √ 19.6 63.09 63.97 52.61 37 F √ √ 49.0 64.23 65.38 53.94 42 本文算法 √ √ √ 16.7 65.84 66.35 54.20 37
Table 3. Comparison of detection results under different network models
View tableView in ArticleTable 3. Comparison of detection results under different network models
目标检测算法 主干网络 模型大小/MB mAP/% 每帧推理时间/ms Faster RCNN ResNet50 160.2 76.3 30 SSD VGG16 112.1 69.7 95 YOLOv3 Darknet-53 240.0 73 77 YOLOv4 CSPDarknet-53 255.0 75.1 64 YOLOv4_tiny CSPDarknet53_tiny 29.6 65.8 49 YOLOv5s CSPDarknet 46 81.0 34 文献［ 20 ］FGHDet 2.6 74.4 35 文献［ 21 ］MobileNetV1 26.3 81.1 36 本文算法 MobileNetV3 13.6 85.4 28
Table 4. Test result data of two network models under scenarios（a）and（b）
View tableView in ArticleTable 4. Test result data of two network models under scenarios（a）and（b）
目标类别 场景（a）下检测目标个数 场景（b）下检测目标个数 Dontcare Person Truck Car Traffic light Traffic sign Person Car 真实标注值 4 7 2 2 4 7 1 13 YOLOv5s算法 6 7 0 1 3 8 3 11 本文算法 4 7 2 2 3 7 1 12
Tools
Get Citation
Copy Citation Text
Qing ZHOU, Gong-quan TAN, Song-lin YIN, Yi-nian LI, Dan-qin WEI. Road object detection algorithm based on improved YOLOv5s[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(5): 680
Paper Information
Category: Research Articles
Received: Jul. 30, 2022
Accepted: --
Published Online: Jul. 4, 2023
The Author Email: Gong-quan TAN (tgq77@126.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20