Analysis of Grey-Level Features for Line Segment Stereo Matching

In order to recover 3D-information from stereo image pairs, a number of stereo matching methods are available. These methods mainly differ in various tokens which are applied to solve the correspondence problem between left and right image. In most of the line segment based stereo matching algorithms, geometrical and structural information is used to estimate the correspondence. Unfortunately these features are dependent on the accuracy of segmentation and the angle of view. This might cause an ambiguity regarding the solution. Grey-level features (GLF) are introduced as robust features independent of segmentation errors and the angle of view. Concerning different line segmentation algorithms, e.g. sequential-oriented or global-oriented algorithms, two methods are proposed for a proper estimation of grey-level information. Finally this additional information will be applied in a line segment based stereo matching algorithm. Regarding these different grey-level features the uniqueness of the solution and the computational effort will be compared with geometrical features. Experimental results are presented.