Object segmentation using stereo images

Ping An,Chaohui Lü,Zhaoyang Zhang

DOI: https://doi.org/10.1109/ICCCAS.2004.1346183

2004-01-01

Abstract:Object segmentation is one of the vital tasks in various three-dimensional applications. The paper presents a hybrid object segmentation algorithm to combine intensity segmentation and disparity segmentation in a stereoscopic vision system. First, the disparity maps of the stereo images are estimated using a foreground-based disparity estimation method. Then, the intensity stereo images and their corresponding disparity maps are separately segmented using a region-growing technique. The real segmentation mask can be obtained and the semantic object be extracted by a fusion processing on the intensity and disparity segments. Computer simulations indicate the reliable performance of the proposed algorithm for stereoscopic segmentation.