Depth segmentation using disparity and self-adaption fringe modulation analysis

Abstract

In this paper, we propose a depth segmentation method combining disparity information with fringe modulation information, which aims at improving the robustness under the noisy scene. Firstly, we investigate the continuity characteristics in the U-V disparity map and distinguishing modulation distribution in the wrapped phase images, respectively. Then, the disparity-based method is optimized to detect and classify the valid regions containing objects and their supporting surfaces. Besides, we provide an evaluation criterion to reduce the noise represented as scattered points. Furthermore, through the analysis of the different fringe modulation distributions between objects and their shadow areas, a method with adjustable thresholds is presented to suppress shadow noise. Our approach is evaluated on both simulation scenes and real-world scenes to verify the availability and robustness. Sufficient experimental results indicate that the proposed method can effectively separate the valid region into objects and detect their incomplete supporting surfaces.