Research and Performance Optimization of Traffic Sign Recognition Based on Efficientdet

The purpose of this paper is to improve its performance in traffic sign recognition by studying the EfficientDet model, and to support the autonomous driving technology. Traffic sign recognition faces multiple challenges such as complex environment, multi-scale target detection and small target detection. Traditional methods do not perform well in dealing with these problems. By studying the model architecture of EfficientDet, aiming at the characteristics of traffic sign recognition. On the well-constructed verification set, the model achieves an excellent result of 96.6% with [email protected], which shows its excellent detection accuracy. At the same time, under the stricter evaluation standard of [email protected]:0.95, the model can still maintain a high level of performance of 76.94%. For different types of traffic signs, the model also shows balanced and excellent detection performance, and the detection rates of warning signs, prohibitory signs and indication signs reach 93%, 91% and 84% respectively. In addition, the reasoning speed of the model reaches 75.5 milliseconds per frame, and the number of parameters is only 8.1 million. Through a series of in-depth analysis of evaluation curves and confusion matrices, the outstanding ability of EfficientDet in dealing with multi-scale targets, complex scenes and distinguishing different types of traffic signs is further verified. The confusion matrix clearly shows the distinguishing ability of the model to various traffic signs, which further enhances the reliability of the model in traffic sign recognition tasks.