Research Of Image Processing Basing On Stereo Vision

Stereo vision is to reconstruct the three-dimensional coordinates of objects from several 2D images. It has the characteristics of non-contact measure and simple actualization. As an important branch of computer vision, stereo vision is a discipline-crossing field and attracts many researchers who adopt different technologies, including vision psychology, physic, mathematics and computer science and so on, to research. The technology has important academic significance and a wide application prospect in the areas national defense, medicine, spatial technology.The integrated stereo vision system generally includes six parts: image acquisition, camera calibration, feature extraction, stereo matching, 3D information recovery and post-processing. This paper firstly introduces the basic concept of computer vision and characteristics of the visual process. Then we mainly discuss camera calibration, feature extraction, and stereo matching.Camera calibration is to ascertain parameters of camera and is the important step of 3D reconstruction from images in computer vision research. Based on analyzing ideal camera projection model, this paper utilizes the orthogonality of rotation matrix and the constraint relationship among camera parameters to design the camera calibration system and compute the inner and outer parameters of the camera. The experiment results show that the algorithm could reduce computation burden and has high precision.The pre-processing is the precondition for image matching. This paper briefly introduces image denoise, image enhancement, researches and compares features'characters and extraction methods in detail, and do experiments to validate the algorithms.Image matching is based on feature extraction and image description. This paper discusses the important constituent part of stereo matching, including feature units selection, matching strategy, restriction, and similarity measurement. Aiming at stereo images with different specialties, this paper presents a matching algorithm based on image feature control points and a matching algorithm regional relativity. The matching results have good performance.