Computer Vision Based Method For 3D Reconstruction Of The Standing Trees

For China, the demand for the wood is great, while the output of the wood is relatively less. Developping the plantation forest is a potent way of approaching this problem. For plantation forest, the correct cultivation can improve the quality of the wood and the output of the wood for every hektare plantation forest. The intelligent cultivating machines with the machine vision system including intelligent pruning machine on trees, intelligent pruning machine on ground and the selective cleanng machine are the key equipments for the cultivation. For the future, the forest harvesting machine will be equipped with the computer vision system. The method of the 3D reconstruction for the standing trees is an important technology for the computer vision system of the intelligent forestry machine.The main content of this thesis is to study the method of 3D reconstruction for the standing trees from the images got from different orientations by the camera on the intelligent forestry machine, that is to say to get the 3D information of the standing trees by using the binocular stereo vision. Getting the trunk and branches of the standing tree, extracting the corners and stereo matching, and projective reconstruction are discussed detailly in this thesis; also the camera self-calibration and 3D Euclidean reconstruction are presented in detail. Getting the trunk and branches, and eliminating the little branches and leaves are all done for the following 3D reconstruction. Extracting the corners and stereo matching are done to get the corresponding points on the different images. In the projective part, the fundamental matrix of two images can be computed firstly from the corresponding points of the two images, and then 3D reconstruction is made in the projective space according the cordinates of the position where the camera is to get the first image. In the camera self-calibration and 3D Euclidean reconstruction part, the intrinsic parameters K can be get from the fundamental matrix, and the rotation matrix R and transferation vector t can be got from the the essential matrix E computed from the the intrinsic parameters K and the fundamental matrix F. At last, the 3D information of the standing tree in Euclidean space can be computed from the intrinsic parameters and motion of the camera. Following main conclusions are drawn:1. Two methods on standing tree image segmentation are proposed. One is level set method in standing tree image segmentation based on particle swarm optimization, the other is tree image segmentation based on mathmatical morphology. For the level set method on standing tree image segmentation based on particle swarm optimization, the image segmentation based on C-V model is considered as an optimal problem that was traditionally considered as a PDE problem before, and then is approached by the particle swarm optimization. The experimental results show that it is effictive for the two class-image segmentation with differnet backgrounds. For the tree image segmentation based on mathmatical morphology, the tree image is segmentated by the watershed algrithm and then segmentated by the auto-threshold method to remove the over-segmentated problem. The experimental results show that the segmentation effct of the mathematical morphology method is better than those of the Sobel and other operation method. So the two methods are all useful for the standing tree image segemtation with tanglesome background.2. One method on corner detection and matching is proposed. The method combine the scale invariant strongpoint for corner detection of the SIFT and the merit for fastly computing of the NCC matching rule. The experimental results show that the method is more accurate than the Harris and has higher computing velocity than the traditional SIFT.3. One method on computing the fundamental matrix based on particle swarm optimization is proposed.The method can solve the problem how to select eight groups of the corresponding points in eight-point method. It is an optimal approaching to computing the fundamental matrix in the global space. But it is researched deeply that the fast computing algrithm of the method, because the method with higher time-complexity.4. One method on estimating the intrinsic parameters based on genetic algrithm is proposed. For the method, the camera self-calibration is considered as an optimal problem to get the minal value of the cost function, and it is solved by the genetic algrithm in this thesis.5. The method on 3D reconstruction or getting the 3D information of the standing tree by using the principle of computer vision is valid. The experimental results show that the method can be used in the vision system of the intelligent forestry machine to solving the problem of getting the 3D information of the standing trees.