Solving Line-Feature Based Stereo Matching Problem By Genetic Algorithms

• Main Authors: Ping-Chun Wu, 吳炳諄
• Other Authors: Fang-Chih Tien
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/73483531316304781404

碩士 === 朝陽科技大學 === 工業工程與管理系碩士班 === 89 === Stereo vision has been popularly used in many industrial applications such as autonomous guided vehicle (AGV), robotics, automated visual inspection, and so forth. Stereo vision employs a pair of cameras at separate locations and simultaneously takes images of an object from different angles. The depth information perceived in stereo vision systems depends on disparity between the two images. The key to this information is to determine which feature in one image corresponds to a given feature in the other image; that is called the "correspondence problem." In this paper, a GA line-feature-based stereo matching formulation is proposed in order to obtain the 3D information. Instead of the “point” features, line features are used to facilitate the entire matching process. When using point features, matching speed would be very slow because of a huge number of matching candidates. Also, a feasible number of matching point is hard to select. Another disadvantage is a reconstruction of the object from these 3D points must be done. In the contrary, using line features reduces the number of matching entities. Also, the feature selection and reconstruction processes can be avoided. Similar to the point-feature-based method, the correspondence problem will be formulated as a “0-1” integer programming problem, in which the objective function is a correlation between two sets of line segments and the constraints include uniqueness, epiplor, disparity and other possible constraints. This problem is first solved as an assignment problem under the ideal condition. When the numbers of matching candidates in two images are different or constraint conflict happens, an energy function is constructed and Genetic Algorithm is applied to solve this problem. At the end, a false target removing process is proposed to screen out the unfeasible matches. A designed experiment is conducted to verify the proposed method. The final comparison is based on the criteria of matching speed, computation storage, matching correctness and others. It shows that the proposed method has effectively solved the stereo matching problem, and is capable to derive the 3D dimension of objects.