Dense Stereo Matching Optimization Algorithm based on Image Segmentation and ORB Gravitational Field

This paper reference the data field ideology from Cognitive Physics, presents a novel dense stereo matching method, in which the cost metric of robustness to texture-less and occluded regions. In order to correct the mismatching and constrain error propagation in cost aggregation, Oriented FAST and Rotated BRIEF (ORB) gravitational field is introduced to the traditional matching cost metrics on the basis of previous studies. Strong feature points can be regarded as correct depth estimation points, which can be used as the center to establish gravitational field for effective energy transmission of each pixel. Then the combined matching cost metrics are aggregated exploiting image segmentation structure. Finally, the experiments on the stereo images of datasets and ours prove that the proposed algorithm has better performance in matching accuracy and robustness.