Research On Stereo Matching Algorithm For Blind Travel

Blind people can not get enough external information,which limits the blind people’s travel.The key information when blind travel is the spatial location and shape of the obstacles.But there are many problems in blind equipment such as high cost,low ranging precision and small range of service,which lead to a lot of inconvenience of blind travel.The computer vision technology can not only obtain the distance information of objects,but also the three-dimensional structure information of objects.These advantages make it possible for blind people to travel.However,the use of computer vision technology to serve the blind also needs to consider the following issues: whether the accuracy of the ranging algorithm is sufficient,and the algorithm can meet the real-time needs or not.3D information of the object can be highly recovered or not and how to solve the problem that the specular reflection phenomenon which have a great impact on the image data.Therefore,based on the existing computer vision object ranging algorithm,this paper realized the binocular stereo vision object ranging.The best comprehensive performance algorithm was chosen as the research object in the following chapters.Secondly,the stereo matching algorithm based on image convolution transformation was proposed to improve the computing efficiency and precision of stereo matching algorithm.In order to solve the bad effects of specular reflection,compared the stereo matching quality in different color space at first.Then,an algorithm for the recognition and processing of mirror reflection region based on gray continuity as the main constraint condition was proposed.The main work of the thesis and the research results are as follows:(1)In this paper,three commonly used ranging algorithms,namely: BMP GC and SGBM algorithm,were implemented.And the range accuracy,ranging efficiency and 3D reconstruction quality were compared.Then selected the suitable algorithm for obstacle detection for blind navigation.This algorithm would be the optimization object in the following chapters.(2)For the primary purpose of serving the blind,Obstacle detection for blind people needs not only real-time performance but also high precision in ranging accuracy and 3D structure information of object.Therefore,a stereo matching algorithm based on image convolution variation was proposed in this paper.Although the algorithm reduced the size of the image,However,it retained a lot of objectfeature information which applied to the process of 3D reconstruction.At the same time,compared with the traditional SGBM algorithm,it has obvious improvement in time complexity,which further demonstrates the feasibility of applying this algorithm to blind navigation.(3)Specular reflection i a common optical phenomenon that can have a huge impact on images.And further deviate from the detection of obstacles based on computer vision.In this paper,an algorithm based on HSI color space with gray continuity as constraint was proposed to identify and process the specular reflection region.The experimental results show that the method can recognize and eliminate the error gray value in the mirror reflection region,and finally had a good recovery for the whole contour of the mirror reflection region.