Traffic sign recognition based on improved YOLOv4
Gan Zhang, Wenju Li, Chu Wanghui, Su Pan · 2021
The current driverless technology has great security risks. The traffic sign recognition technology equipped in the roof sensor is easily affected by light and occlusion, resulting in poor detection effect. To solve this problem, this paper proposes a traffic sign recognition model based on improved yolov4. According to the characteristics of weighted bidirectional feature pyramid network, a cross layer connection is added to the traditional yolov4 network, and the weight of the transferred feature map is adjusted in the process of feature fusion. This method can enhance the feature extraction ability of the network. The experimental results show that the proposed method can detect more types of traffic signs, the mAP of TT100K dataset reaches 87.85%, which is 1.03% higher than the traditional yolov4 algorithm, and the frame processing speed reaches 31.25fps, which meets the requirements of real-time detection.