| Text traffic signs contain a lot of semantic information related to traffic conditions such as location,distance,direction,warning,etc.This information provides important road tips for drivers and pedestrians.Accurate detection of text traffic signs and accurate recognition of textual content can not only provide safe driving for drivers or intelligent transportation systems,but also provide auxiliary decision-making information for drivers or intelligent transportation systems,and further assist in solving traffic safety driving and traffic congestion and other issues.This article explores and studies the detection and recognition of text traffic signs.A traffic sign text detection method based on video stream is proposed.The method makes full use of the correlation between the pixels of the coding block and the time correlation of the traffic sign video stream,and includes three functional modules: a traffic sign text edge detection module,a traffic sign text preliminary detection module and a traffic sign text correction detection module.Through intra-frame prediction mode,CBP value and pixel residual information,the edge detection of traffic signs is performed.Preliminary detection of traffic sign text based on the geometric characteristics of traffic signs,the background color of traffic signs and the characteristics of the text of traffic signs.Through intra-frame prediction to predict the probability distribution of the mode,the sum of the absolute value of the vector residual value and the CBP value statistics of the pixel residual distribution,text traffic signs are corrected and detected.This method greatly shortens the detection time while maintaining high detection accuracy,which is beneficial to the real-time detection of text traffic signs.In order to further improve the detection accuracy,a traffic sign text detection algorithm based on video stream and deep learning is proposed.Through the analysis of the video stream,the judgment confidence probability of the text traffic sign is obtained,and it is processed according to the different classification of the judgment confidence probability.If the confidence probability of the text traffic sign is higher than 0.78,it is directly judged as a text traffic sign;when the confidence probability of the traffic sign is lower than 0.3,it is directly judged that it does not include the traffic sign area;the confidence probability is in the interval of [0.3,0.78] to do further judgment.The judgment is based on the video stream traffic sign detection classification algorithm to give the traffic sign candidate area,and the traffic sign detection algorithm based on deep learning is used to make the judgment.The feature extraction network Res Net is improved,and the YUV residual data and YUV data are used for feature extraction together to improve the efficiency of model detection.Experimental results show that the detection algorithm in this paper has a good detection effect of text traffic signs in terms of detection rate,false detection rate,and average detection time.A CRNN-based text recognition algorithm for traffic signs is implemented.The CRNN-based natural scene text recognition method is applied to text traffic sign recognition.First,the traffic sign text area is obtained by the traffic sign text detection method based on video stream and deep learning,and the text recognition algorithm based on CRNN is used to recognize the traffic sign text in this area. |
