Stereo matching algorithm for adaptive window based on pixel expansion

Yubo MEN,Guoyin ZHANG,Chaoguang MEN,Xiang LI,Ning MA

DOI: https://doi.org/10.11990/jheu.201705056

2018-01-01

Abstract:Aiming at the foreground expansion in the disparity map of the local stereo matching method,this study uses the adaptive window stereo matching method based on pixel expansion to reduce the foreground expansion in the disparity map according to the Gestalt vision theory.First,the method establishes the window model for the pix-el to be matched by the extending constraints;second,the pre-matching judgment is conducted using a relative co-incidence rate of the window of the point pairs to be matched,and the window that meets the matching condition is selected point-by-point in the disparity searching range.Finally,the final disparity result is calculated through the normalized cross correlation of window regularization and the smooth disparity plane segmentation.The experimental results show that the proposed algorithm can obtain a more accurate dense disparity map and improve the matching accuracy of the depth discontinuities and occlusion regions.The average mismatching rate in Bicycle2, Class-room2,Hoops,and Staircase is 14.9%.