Deep Learning-Based Traffic Sign Identification Using Multi-Scale Feature Fusion

Roadside traffic signs are used to enforce vehicle safety laws, appropriately convey traffic conditions, and aid in safe driving. Therefore, accurate traffic sign identification is essential for automated driving. In this article, we suggest a simple and reliable one-stage object detection model for detecting and identifying traffic signs by using ResNet50 and a Feature Pyramid Network (FPN). We evaluate our network using the Tsinghua-Tencent 100K (TT100K) benchmark for traffic sign recognition, and the results demonstrate that the detection network can obtain overall Mean Average Precision (mAP) of 46.012%.