Eigenspace technique in stereo vision matching processing for autonomous navigation

A new method in correspondence problem for two cameras stereo vision system is proposed in this paper. Reduction noise effect and computation time are the core of our investigation. In direction to reduce noise we transfer two left and right images to two correspondence vectors. The transferred vectors then construct a 2ç2 matrix. The algorithm is mainly based on matrix space concept and eigenspace decomposition of the matrix. The main assumption is that the two cameras are arranged horizontally so that correspondences points are distributed in horizon direction. Some practical examples presented to evaluate our investigation.