Improved lightweight traffic sign detection method with YOLOv7-tiny

A lightweight traffic sign detection method MCBAM-F-YOLOv7-tiny, which improves YOLOv7-tiny, is proposed to address the issues of low computational power, difficulty in improving real-time performance, low detection accuracy in low light environments, and detection omission errors in vehicle terminals and other devices. Firstly, the CBAM-driven MCBAM network is used as the backbone feature extraction network in the algorithm, which effectively saves computational resources and reduces the size of the network model; Then, a spatial pyramid pooling SPPFCSPCSIM module based on cascaded pooling layers was designed to reduce computational time and obtain an efficient and concise detection algorithm model. Finally, combining CIoU Loss with Focal Loss to construct an optimized Focal CIOU Loss function for bounding boxes, speed up the model training convergence speed and improve the learning ability of the network. The experimental results show that the improved algorithm achieves 89.4% precision, 76.7% recall, and 85.0% mAP on the CCTSDB traffic sign detection dataset, which are comparable to the original model, and the model size is reduced to 69.3% of the original model, and the detection speed reaches 101.1f·s
^-1
, which is a 19.4% improvement and meets the real-time requirement. The visualization experiment further verifies that the improved algorithm can improve the detection accuracy in low light and avoid wrong detection and leakage.