Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
Improving Traffic Sign Detection in YOLOv7
Faculty of Science, Kunming University of Science and Technology, China.
ABSTRACT
Aiming at the problem that the existing object detection algorithm has a poor detection effect on traffic signs with non-obvious features and complex backgrounds in the image, a traffic sign detection algorithm based on YOLOv7 is proposed. Firstly, the Kmeans++ algorithm is used to replace Kmeans clustering to obtain anchor frame sizes that match the traffic sign dataset to further improve feature extraction capabilities. Secondly, the CBAM attention mechanism is introduced in the backbone network, this attention mechanism makes the network focus on traffic signs and avoid being affected by the background information. Finally, the spatial pyramid pooling SPPCSPC_C structure is proposed, which can reduce the filtering of the edge information of the convolutional layer on the small targets and extract more accurate small target features. Tested on the TT100K traffic sign dataset, the proposed algorithm achieves 95.51% accuracy, which is 2.11% higher than the original YOLOv7 model. The experimental results show that the improved model can effectively enhance the accuracy and model performance of traffic sign detection.
Keywords: Deep learning, Computer vision, Object detection, Traffic sign detection, YOLOv7, Attention mechanism.
Download PDF
Faculty of Science, Kunming University of Science and Technology, China.
ABSTRACT
Aiming at the problem that the existing object detection algorithm has a poor detection effect on traffic signs with non-obvious features and complex backgrounds in the image, a traffic sign detection algorithm based on YOLOv7 is proposed. Firstly, the Kmeans++ algorithm is used to replace Kmeans clustering to obtain anchor frame sizes that match the traffic sign dataset to further improve feature extraction capabilities. Secondly, the CBAM attention mechanism is introduced in the backbone network, this attention mechanism makes the network focus on traffic signs and avoid being affected by the background information. Finally, the spatial pyramid pooling SPPCSPC_C structure is proposed, which can reduce the filtering of the edge information of the convolutional layer on the small targets and extract more accurate small target features. Tested on the TT100K traffic sign dataset, the proposed algorithm achieves 95.51% accuracy, which is 2.11% higher than the original YOLOv7 model. The experimental results show that the improved model can effectively enhance the accuracy and model performance of traffic sign detection.
Keywords: Deep learning, Computer vision, Object detection, Traffic sign detection, YOLOv7, Attention mechanism.
Download PDF