3D Scene Reconstruction Based On Binocular Stereovision

Binocular Stereovision is a very important technique in 3D reconstruction of scene. When use it, we first get two or more pictures by camera, then we could get the 3D information included in these pictures by geometry and other relationship. This skill is widely used in many parts of industry or other areas, and also it doesn't need to project light to the scene, which could introduce secret problems .So it has some advantages that initiative vision couldn't compare. This paper mainly deals with the matching problem, together with some de-noising problems.Binocular Stereovision proceeds with the reconstructing process based on disparity. One important problem in it is how to achieve good matching between the according points in the two pictures. As the scene will lose the information of one degree when projected to the 2D picture, this technique reconstructs this lost information with two or more pictures. But because of the irregularity of the scene, there is much uncertainty in matching. Also as the sheltering and blurring in the scene, the according points in the two pictures maybe could not turn up at the same time, or maybe don't have the same color. In order to proceed with the matching more effectively, this paper raises a new matching algorithm: matching two according polar lines in two directions. This technique achieves every point's matching by matching according leaping points in according polar lines. With this technique, we could solve the sheltering problem to some extent. Also compared with point's color value, the leaping points are less easily disturbed by the changing of light. So it could be more robust. Moreover, we could solve the problem about the initiative points preferably.In view of the matching problem, there have been many computing or optimizing algorithm, for example iterative algorithm, which first decides some initiative matching points, then optimizes the result according to some rules until the result is up to some ending rule. Also some papers use dynamic programming algorithm to get the optimized matching. These algorithms usually will take too much time, so they could not answer for the need of real-time reconstruction. In the method raised in our paper, we first match the leaping points, not every picture point. In this way, we could greatly reduce the quantity of the elements that need to match. Then we use the interpolation method to match the points between the leaping points. If the initiative matching points are given, the matching process could be finished in linear time. Besides, this paper also provides a new method in detecting noise in order to get rid of the matching parts which have large warps. According to the experiment, we could see that with this method, we could get good matching result.In short, with the algorithm which matches the two according polar lines in two directions raised in this paper, we could get preferable points matching, and then get preferable reconstruction result. And also we could get a very high speed.