A lightweight and high-precision traffic sign identification method
Bo Yin · 2023
This paper presents a lightweight and highly accurate method for identifying traffic signs. This method adopts the object detection method without anchor frame, which greatly reduces the number of parameters of the deep learning network model, reduces the network complexity, and realizes the purpose of lightweight network. In addition, this method greatly improves the accuracy of object detection by improving the backbone network structure, neck structure and detection head, and optimizing the loss function. According to experimental simulation, the parameters of this method is only about 3.3M, which belongs to lightweight network and can be easily deployed on mobile terminals. In addition, the detection accuracy of this method mAP (0.50) is as high as 99%, which is significantly higher than the current mainstream object detection method.