Binocular Stereo Vision3D-reconstrection Based On Improved SURF

As the hot and difficult issues in computer vision, binocular stereo vision is an important form of computer vision, it is based on two cameras, whose parameters are same or different, to get two images of the same3D scene in a different perspective simulating the human visual mechanism, to restore the3D scene depth information using parallax principles and mathematical methods. It has a broad application prospects in many computer vision field, such as aerial mapping, visual navigation, motion analysis and industrial inspection etc., so it has some theoretical basis and research value to searching for3D reconstruction based on binocular stereo vision. Thesis researched in-depth and improved3d reconstruction based on binocular stereo vision mainly from the binocular stereo camera calibration, image feature extraction and stereo matching, depth information acquisition and so on.In binocular stereo camera calibration module, gets internal parameters of single camera using Zhang Zhengyou checkerboard calibration algorithm, and accurate them using re-projection error. For stereo camera calibration, put forward a improved stereo calibration method based on epipolar rectification, firstly rectify the two images using camera parameters, then calculate the rectifying error, finally optimize the parameters matrix between cameras using the error. The experiments show that it has a higher precision using the improved calibration algorithm.In the aspect of image feature extraction and stereo matching, adopted SURF operator as the local feature descriptor compared with performance between several kinds of typical local operator comparison, then proposed an improved SURF algorithm combined with KD-tree feature matching algorithm. The improved algorithm extended dimension of the SURF feature descriptor vector to128, generally, the longer the length of the feature vector, the greater the amount of information carried by the feature descriptor, and the higher the probability of feature vector detection in matching, but the consideration of time spent will has a corresponding increase in the match. For this problem, adopt to BBF search method based on KD-tree, establish a KD-Tree structure for these extension features description vector. For KD-tree has a has strict requirements on the dimensions of the data set, generally no more than20D, expand it to high-dimensional data set using the improved KD-Tree nearest neighbor query mechanism, executing the BBF nearest neighbor search, then purify the matching points using the ratio between the nearest neighbor and next nearest neighbor. The experimental results proved that the improved SURF algorithm is of less computing time compared with SIFT algorithm, which is widespread used now. It is more applicable for image registration, which has a higher demand of real-time and an obviously change of scale, rotation, and brightness etc. It provides a new train of thought for the research field of panoramic image misaiming, stereo vision and3D reconstruction etc.After completed the feature points matching, we can build the correspondence between matching points and the3D object points using the camera parameters which are calibrated, in order to obtain3D information. The3D reconstruction system is based on binocular stereo vision, so the3D coordinate values of the matching points can be calculated according to the measurement principle of binocular parallax, generate a3D sense of depth image using vc6.0combined with opencv.