Keypoint Extraction And Matching Algorithms In Stereo Vision Measurement

Image local keypoint extraction and matching is an important technology in the field of computer vision, it provide basic technical support for many of the more advanced visual research and application. With the target recognition and tracking, stereo vision measurement, robot vision navigation and path planning, three-dimensional reconstruction of computer vision technology widely used in industrial production, aerospace, medical and even military fields, image local keypoint extraction and matching algorithms more and more attention by the academic staff of research and engineering applications. In addition to the classical algorithm like Harris, SIFT and SURF, many of the new excellent algorithm such as FAST, AGAST, BRIEF, BRISK are emerging constantly, image keypoint extraction algorithm is increasingly becoming one of the hot field of computer vision research and application.The luminance variation is an important factor affecting the keypoints extraction and matching, against the brightness defect image keypoints extraction and matching number less problems, joined the LCC and SCB brightness color correction method to shift image brightness and color to a similar level, and significantly improve the keypoints extraction and matching number. In the SIFT and SURF keypoints matching strategies, we did a deep research on the K-D tree algorithm and BBF search strategy which joined priority queue based on K-D tree. For keypoints mismatching problem, an optimization algorithms ORSA based on RANSAC was taken to filter the mismatches.SIFT and SURF algorithm has excellent performance, but there are also very significant speed and efficiency defects, greatly limiting their application in image and video retrieval, real-time target tracking, visual navigation and so on. The new algorithms like BRISK and ORB, although there are faster execution speed, the performance has to sell at a discount greatly. In this paper, a detailed comparison and analyses aim at the scale transformation, perspective transformation, brightness change and other factors of the SURF, ORB and BRISK three algorithms was performed. Finally, a compromise algorithm SU-BRISK whose performance range between SURF and BRISK was given, greatly improve the computation speed, while maintaining good performance. And finally, we established a set of stereo vision location system based on SU-BRISK on Visual Studio2010combination of OpenCV platform.