Research on Self-driving Based on Dynamic Recognition of Traffic Signs
Shukai Ding, Qu Jian · 2022
Traffic sign recognition is important for self-driving cars, and several approaches use deep neural networks to solve the problem. However, most of them use pre-collected datasets and are fed into cluster computing for testing traffic sign recognition, which we call traffic sign detection in a static environment. These approaches ignore the noise interference of the realworld environment on traffic signs and reduce the difficulty of implementing the task. Therefore, we built one agent which detects real-world traffic signs in realtime. We call it dynamic detection. To implement the real-time traffic sign recognition task, we chose the lightweight MobilenetV2 as the baseline model and filtered the optimal activation function and optimizer for MobilenetV2. The experiments show that our method achieves similar results to existing methods in terms of Accuracy-Static. However, our proposed MAMobilenetV2 achieves 95.9% in terms of AccuracyDynamic, which is significantly higher than existing models.