Stereo Vision for 3D Measurement in Robot Systems

Abstract

The major obstacle in the application of stereo vision to extract 3D information is the error in disparity map and highly computational cost at conventional computers. This paper is concerned with finding software solution to obtain tradeoff between disparity map accuracy and reductions in execution time through developing the stereo vision algorithm. This algorithm is based on block matching technique, in which an image is partitioning into blocks. An analysis of the essential parameter of this technique (the size of block) is performed to obtain the optimal solution. Adaptive block size with vertical edge detection has been adopted to implement the proposed algorithm. The significance of this work is reduction of the execution time of the stereo vision algorithm, where the execution time of the proposed algorithm is about (2%) of the required time to execute standard vision algorithm, when running at conventional computers