Using a Complete Convolutional Network for the Detection and Recognition of Symbolic Traffic Signs
Jinwei Yang · 2024
In this paper, a traffic sign detection and recognition method based on full convolutional network guidance is proposed. First, a full convolutional network (FCN) is used to distinguish between symbolic and non-symbolic traffic sign areas in images. Then, the object candidate Box method-Edge Box is used to locate symbolic traffic signs, find a small number of potential candidate boxes for symbolic traffic signs, and submit them to the deep Convolutional network classifier for classification. Through this detection and recognition process from coarse to fine, most of the background areas are filtered, and almost all the traffic sign areas are retained. This significantly diminishes the search scope for symbolic traffic signs. The experimental results show that the proposed symbol traffic sign detection and recognition based on the full convolutional network guidance can achieve good detection and recognition effect on different symbol traffic signs, achieving an average accuracy of 98.67% and an average recall rate of 93.27%. The recall rate is very high, the accuracy rate is also high, and it can meet the requirements of real-time.