Research On Key Technology Of Local Blind Image Navigation Based On Scene Matching

This paper mainly studies the key technologies on local navigation for the blind person by scene matching technology. Scene matching navigation technology is a navigation method, which on the basis of establishing benchmark image information feature database, extract the benchmark image features and match them with feature database to get the corresponding geographic information and output in voice.Firstly, this paper studies preprocessing method of images acquired by the blind person under the certain circumstances, and then for several common scene matching algorithm, we use them in the blind person navigation to simulate their application results. This paper focuses on two types of matching algorithms:an algorithm is the scene matching techniques based on the Scale Invariant Feature Transform (SIFT), namely determine magnitude and direction of descriptors by the gradient information around feature points, generate 128-dimensional descriptors by Gaussian weighted, and match by similarity measures approach. Another improved algorithm is based on Speed Up Robust Features (SURF), namely determine the main direction by calculating the gradient information around feature points using Haar wavelet, generate 64-dimensional descriptors by Gaussian weighted in the sub-region, fast index match by Hessian matrix trace, and then match again by the sum of square of match points distance difference (one of similarity measure methods). Finally, using multiple real-time image to analysis and compare the experimental results of SIFT and SURF algorithms, concluded that SIFT and SURF algorithm is effective for change of rotation, scaling, noise and light and impact of a variety of integrated factors.Synthesis results show that, SURF matching algorithm is better than SIFT matching algorithm. SURF matching algorithm can be fast, robust and accurate, and lay the theoretical basis for achieving local image navigation of the blind person.