Laser & Optoelectronics Progress, Volume. 61, Issue 12, 1228008(2024)
Real-Time Traffic Sign Detection Based on Yolov5-MGC
Figures & Tables(12)
Fig. 1. Structure of lightweight backbone. (a) Structure of Mobile-Bottleneck; (b) depthwise separable convolution; (c) structure of MobileCSP
Fig. 7. Visualization of detection results of different models in different scenes
Table 1. Results of ablation experiments
View tableTable 1. Results of ablation experiments
Model M G C mAP@0.5 /% Speed /（frame·s-1） Param /MB FLOPs /109 Yolov5m 91.76 55.18 21.12 50.81 Yolov5m-M √ 92.20 69.16 12.09 26.36 Yolov5m-G √ 93.16 52.08 21.37 52.59 Yolov5m-C √ 92.92 52.51 21.66 52.69 Yolov5m-MG √ √ 92.62 67.22 12.08 26.28 Yolov5m-MC √ √ 92.19 67.91 12.11 26.59 Yolov5m-GC √ √ 94.02 48.78 21.90 54.45 Yolov5m-MGC √ √ √ 94.34 62.59 12.36 28.38
Table 2. Comparison of attention models
View tableTable 2. Comparison of attention models
Model mAP@0.5 /% Speed /（frame·s-1） Param /MB FLOPs /109 Yolov5-SE 91.69 37.58 47.06 115.92 Yolov5-CBAM 91.78 53.26 21.42 51.62 Yolov5-ECA 92.84 51.09 21.38 51.62 Yolov5-MS-CAM 92.72 52.67 21.31 51.30 Yolov5-GLFA 93.16 52.08 21.37 52.59
Table 3. Performance comparison of different models on CMTSD
View tableTable 3. Performance comparison of different models on CMTSD
Model Backbone mAP@0.5 /% Speed /（frame·s-1） Param /MB FLOPs /109 Faster R-CNN Resnet50 63.64 10.25 137.02 402.31 SSD VGG16 90.37 54.50 23.879 274.05 Yolov4 CSPDarknet53 89.12 34.46 64.02 141.72 Yolov5m CSPDarknet53 91.76 55.18 21.12 50.81 Yolov5m Mobilenetv3 92.18 58.19 11.76 21.13 Yolov5m-6.0 CSPDarknet53 91.65 57.04 20.94 48.24 Yolov7 ‒ 92.26 43.69 37.28 105.04 Yolov5-MGC CSPDarknet53 94.34 62.59 12.36 28.38
Table 4. Performance comparison of different models on CCTSDB
View tableTable 4. Performance comparison of different models on CCTSDB
Model Backbone mAP@0.5 /% Speed /（frame·s-1） Param /MB FLOPs /109 Yolov4 CSPDarknet53 90.30 36.82 64.02 141.72 Yolov5m CSPDarknet53 89.82 44.19 21.12 50.81 Centernet Resnet50 91.34 38.69 32.66 70.21 Yolov5m-6.0 CSPDarknet53 92.19 46.67 20.94 48.24 Model in Ref.［ 28 ］CSPDarknet53 84.35 ‒ ‒ ‒ Model in Ref.［ 29 ］‒ 92.47 20.0 ‒ ‒ Yolov5-MGC CSPDarknet53 94.31 51.37 12.36 28.38
Table 5. AP comparison for a target with defferent sizes
View tableTable 5. AP comparison for a target with defferent sizes
Model AP for a target with defferent sizes Small Medium Large Yolov5 0.247 0.616 0.764 Yolov5-MGC 0.482 0.696 0.783
Tools
Get Citation
Copy Citation Text
Ningke Zhu, Qing Ge, Hanwen Wang, Pengfei Yu. Real-Time Traffic Sign Detection Based on Yolov5-MGC[J]. Laser & Optoelectronics Progress, 2024, 61(12): 1228008
Paper Information
Category: Remote Sensing and Sensors
Received: Jul. 12, 2023
Accepted: Sep. 18, 2023
Published Online: Jun. 5, 2024
The Author Email: Pengfei Yu (pfyu@ynu.edu.cn)
CSTR:32186.14.LOP231703
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20