Blind Path Detection And Application Based On Gabor Feature

With the development of science and technology,barrier free facilities have been gradually improved in our lives,but the safe travel of blind people is still an urgent problem.There are a large number of blind people in our country.How to use information technology to improve the quality of life of this group has attracted more and more attention from all walks of life.The blind guide technology based on computer vision improves the blind people's ability to obtain information through blind path detection,which has high application value.The key of blind path detection is to get two boundary lines of blind path accurately.According to the blind path features in the image,this paper uses Gabor features of binary image to design algorithm and implement the blind path recognition software,which solves the problem that blind people can't get blind path information.The main work of this paper are as follows:First of all,this paper based on the analysis of different blind road scene images,the preprocessing process of blind road image is designed according to the difference of texture features between the near end and the far end of blind road.The blind image is corrected by perspective transformation,and the key pixels in the blind image are highlighted through binary processing,so as to reduce the interference of fine texture in the subsequent feature extraction.Then,this paper comparing different texture feature extraction methods,Gabor filter with faster speed is selected for image filtering.After normalization and non-linear transformation of redundant filter images,filter images with large amount of information are selected to increase the representativeness of extracted features?Means + + clustering algorithm can effectively segment the blind path region,extract the edge pixels of the segmented blind path image by Canny operator,add sliding window to screen the pixels belonging to the blind path boundary,and detect the final blind path boundary by Hough.After that,according to the obtained boundary line of blind track,a blind track information judgment model is constructed.Calculate the angle information between the direction of the blind person and the boundary line of the blind path to detect the deviation of the blind path;build the distance measurement model of monocular vision,and increase the calculation of the length of the blind path.At last,Through the demand analysis of blind track recognition software,the function,performance requirements and feasibility of blind track recognition software are determined.The blind track recognition software is designed in detail around the image acquisition,image processing,voice reminder and information display module of blind track recognition system,and the blind track recognition software is realized.It is tested and verified by four groups of different types of blind track video,and the accuracy of blind track recognition is up to 96.8%.