Examination of improvement of traffic sign identification using CNN with attention mechanism

Abstract

For the traffic sign recognition problem in the advanced driving assistance system (ADAS) and self-driving system, this study uses YOLO, one of the deep learning methods of target detection, to detect traffic signs based on the Chinese traffic sign dataset TT100K and the self-made Japanese traffic sign dataset. In addition, by introducing attention mechanism into YOLO model, the recognition performance of traffic signs is improved. Subsequently, using the weights trained from the TT100K of the large Chinese traffic sign dataset, the self-made Japanese traffic dataset is trained by the method of fine-tuning of transfer learning. The experimental results show that the convergence speed of the model using transfer learning is improved and a better recognition effect is obtained.