Traffic Sign Detection Algorithm Based on Improved YOLOv5

Wei Xing, Huang Hongqiong
Proceedings of 2024 International Conference on Machine Learning and Intelligent Computing, PMLR 245:316-324, 2024.

Abstract

Due to the phenomenon of small size, complex background or high density of traffic signs, there is a certain degree of missing or false detection, which ultimately leads to the problem of reduced detection accuracy. To solve this problem, a real-time traffic sign detection algorithm based on YOLOv5s is proposed. Firstly, feature upsampling is carried out through ContentAware ReAssembly of Features upsampling operator in the neck network, which can aggregate information in the large receptive field, so that the network can get a more accurate feature map. Secondly, the normalized Gaussian Wasserstein distance is used as the similarity measure to construct the NIOU regression bounding box loss function to improve the overall performance of the model. Finally, the FasterNet module is used instead of the C3 module, which is lighter and faster. Experiments were carried out on TT100K data set. Compared with YOLOv5s, CNF-YOLO algorithm reduced Parameters by 1/5, the computing load decreased by 3GFLOPs, the detection speed increased by 18.4 frames/SEC and the weight file decreased by 2.1MB. All models were lighter. And its mAP@0.5 has also been increased by 0.5% to enable rapid detection of traffic signs.

Cite this Paper

BibTeX
@InProceedings{pmlr-v245-xing24a, title = {Traffic Sign Detection Algorithm Based on Improved YOLOv5}, author = {Xing, Wei and Hongqiong, Huang}, booktitle = {Proceedings of 2024 International Conference on Machine Learning and Intelligent Computing}, pages = {316--324}, year = {2024}, editor = {Nianyin, Zeng and Pachori, Ram Bilas}, volume = {245}, series = {Proceedings of Machine Learning Research}, month = {26--28 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v245/main/assets/xing24a/xing24a.pdf}, url = {https://proceedings.mlr.press/v245/xing24a.html}, abstract = {Due to the phenomenon of small size, complex background or high density of traffic signs, there is a certain degree of missing or false detection, which ultimately leads to the problem of reduced detection accuracy. To solve this problem, a real-time traffic sign detection algorithm based on YOLOv5s is proposed. Firstly, feature upsampling is carried out through ContentAware ReAssembly of Features upsampling operator in the neck network, which can aggregate information in the large receptive field, so that the network can get a more accurate feature map. Secondly, the normalized Gaussian Wasserstein distance is used as the similarity measure to construct the NIOU regression bounding box loss function to improve the overall performance of the model. Finally, the FasterNet module is used instead of the C3 module, which is lighter and faster. Experiments were carried out on TT100K data set. Compared with YOLOv5s, CNF-YOLO algorithm reduced Parameters by 1/5, the computing load decreased by 3GFLOPs, the detection speed increased by 18.4 frames/SEC and the weight file decreased by 2.1MB. All models were lighter. And its mAP@0.5 has also been increased by 0.5% to enable rapid detection of traffic signs.} }
Endnote
%0 Conference Paper %T Traffic Sign Detection Algorithm Based on Improved YOLOv5 %A Wei Xing %A Huang Hongqiong %B Proceedings of 2024 International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2024 %E Zeng Nianyin %E Ram Bilas Pachori %F pmlr-v245-xing24a %I PMLR %P 316--324 %U https://proceedings.mlr.press/v245/xing24a.html %V 245 %X Due to the phenomenon of small size, complex background or high density of traffic signs, there is a certain degree of missing or false detection, which ultimately leads to the problem of reduced detection accuracy. To solve this problem, a real-time traffic sign detection algorithm based on YOLOv5s is proposed. Firstly, feature upsampling is carried out through ContentAware ReAssembly of Features upsampling operator in the neck network, which can aggregate information in the large receptive field, so that the network can get a more accurate feature map. Secondly, the normalized Gaussian Wasserstein distance is used as the similarity measure to construct the NIOU regression bounding box loss function to improve the overall performance of the model. Finally, the FasterNet module is used instead of the C3 module, which is lighter and faster. Experiments were carried out on TT100K data set. Compared with YOLOv5s, CNF-YOLO algorithm reduced Parameters by 1/5, the computing load decreased by 3GFLOPs, the detection speed increased by 18.4 frames/SEC and the weight file decreased by 2.1MB. All models were lighter. And its mAP@0.5 has also been increased by 0.5% to enable rapid detection of traffic signs.
APA
Xing, W. & Hongqiong, H.. (2024). Traffic Sign Detection Algorithm Based on Improved YOLOv5. Proceedings of 2024 International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 245:316-324 Available from https://proceedings.mlr.press/v245/xing24a.html.

Related Material