The Research On Traffic Sign Recognition For Android

With the development and prosperity of economy and living standards, automobiles have become an indispensable part of human life. Alought automobiles have brought great convince to us, traffic accidents and other safety issues also emerged at the same time. According to the literatures reported recently, the research on traffic sign detection with static images has achieved many promising models. However, virtually all of those approaches are performed on PC. Since smart phones have been widely used in almost every corner of our lives, it is no doubt that intelligence transportation will be significantly boosted by proposing a method which can use smartphones to detect and recognize traffic signs. From this point, this thesis proposes a solution for traffic sign detection and recognition with smartphones. The solution can be grouped into client-end and server-end. The client-end runs on a smartphone, captures pictures of traffic signs by the phone camera, and sends the detected signs to server. After recognition, the server-end will feedback the results to the client-end for display.In the traffic sign detection module, this thesis used the color segmentation and hough transformation. Firstly, HSV color model is used for color segmentation, secondly the traffic signs are preprocessed, including image binarization, noise eliminating and candidate regions selecting. Then we classify the candidate regions based on their shape, finally hough transformation is used to detect the shape and check whether the shape belongs to the area of traffic signs, if it does not belong to anyone, then remove it. Based on the method, the client can detect the traffic signs and transmit them to server for recognition.In traffic sign recognition module, this thesis used a graph embedding algorithm based on sparse representation. The algorithm effectively combinate the graph embedding algorithm and sparse representation algorithm into a unified framework, which can not only to preserve the local manifold structure of the data set, but can preserve the discriminative information on different classes as well. In addition, considerating the special nature of the traffic sign- different signals can be divided into different classes, in the same class, the signs can be further divided into different classes. So we use a hierarchical graph structure to better express the relationship between different samples.Finally, we use two standard data sets GTSRB and BelgiumTSC,and images collected in real scene to test the detection and recognition algorithm as well as the whole recognition system, as we can see from the experimental results, the proposed algorithm is superior to the traditional recognition algorithm, and the recognition system has good accuracy and real-time performance.