A Study On3D Reconstruction Of Tree Based On Stereo Vision

With the development of stereo vision, and the wide application of forestry robot, the research and optimization on stereo vision system is very meaningful. In this paper, several key technologies of3D reconstruction of tree are studied based on stereo vision, which can be divided into five parts:image preprocessing, feature extraction, camera calibration, stereo matching and3D-points reconstruction.Image preprocessing is done in order to acquire the most real image, which includes the image gray-scaling and enhancement. On the basis of Harris corner detection algorithm in the feature extraction part, a local corner detection algorithm is presented.The purpose of camera calibration is to establish the correspondences between the3D points and their projected points. We choose the Tsai two-step approach and only consider the radial distortion, then the high accuracy of camera projection matrix is obtained.The purpose of stereo matching is to establish the correspondence between the feature points on left and right image. We choose the feature matching algorithm based on gray-scaling and combine affine transformation constraint and epipolar constraint, so an improved feature matching algorithm is presented finally.In the3D-points reconstruction, we apply the midpoint of vertical line between non-coplanar lines algorithm which can certify for the precision, and3D point cloud of the tree is obtained finally.