Study On Traffic Sign Text Detection And Recognition Based On Video

With the progress of science and technology and the development of social urbanization, the intelligent transportation system has become an important method to solve the urban road traffic. The research of vehicle intelligent auxiliary driving system, which is based on the visual information, has become a hot topic recently. These studies mainly focus on traffic sign detection and recognition of various traffic sign, and very few people research traffic sign detection and recognition which contains text (especially Chinese words). But in all kinds of traffic signs, those that contain text can provide the driver with rich and important road information. And the extraction and application of this information is very important for vehicle intelligent auxiliary driving and unmanned driving. Especially traffic sign text detection and recognition based on video has broad application prospects.The aim of this thesis is to apply text traffic sign detection and recognition in static image into video, and to design traffic sign text detection and recognition system based on video. The main work of this thesis includes the following aspects:1. First of all, a key frame extraction method from video based on content is proposed. After that, the key frame, which is suitable for subsequent text traffic sign detection and segmentation, is extracted by the color and shape features of traffic sign furthermore, the relationship of the front and rear frame.2. In the detection of text traffic sign, the fusion features of DfC and DtB are used to describe the shape of traffic sign, and then the rectangular traffic sign can be detected with Support Vector Machine. Following by that, the black and white points and projection curve fitting method in the vertical direction are searched to judge the traffic sign as text traffic sign, indicated traffic sign or not. This method can detect text traffic sign accurately.3. The traffic sign can be divided into different layers with K-means color clustering algorithm based on the strong contrast of text and background color feature, and the text can be detected and segmented according to the method of analysis of the connected domain. Finally, the text segmented is treated as the input of the OCR module, and the recognition of words can be realized.