Detection and Identification of Text-based Traffic Signs

Abstract

The detection and recognition of text-based traffic signs are important in the field of automatic driving, but these tasks pose problems in practical applications, such as low accuracy in text detection and extraction, poor long-text extraction, and a lack of datasets. To solve these problems and to improve the detection and recognition accuracy of text-based traffic signs so that they can better serve automated driving, we propose an improved Advanced efficiency and accuracy scene test (EAST) model and fixed-size prediction to enhance the capability of extracting features. The text recognition stage features a text preprocessing method that trains convolutional recurrent neural network (CRNN) models using synthetic datasets of Chinese strings. Experimental results show that the improved Advanced EAST model and fixed-size prediction enabled the detection of text on traffic signs to achieve a 96% recall rate and an 88.5% accuracy rate; we also saw better results in the case of dense text and obscuration. Thus, in the absence of targeted datasets, the designed text image preprocessing method can realize print text recognition in different scenarios only by training models using synthetic data, thereby eliminating the need for a large amount of work on training dataset labeling while still meeting requirements of detection and recognition.