High precision and fast disparity estimation via parallel phase correlation hierarchical framework

Abstract

When estimating the disparity of remote sensing images, known phase correlation (PC)-based disparity estimation methods are not fast and robust, such as hierarchical structure PC method and fixed window PC method. To tackle this problem, a parallel PC-based hierarchical framework is proposed, which includes two ideas: first, a weighted PC peak fitting algorithm is introduced for estimating the high precise disparity matrix efficiently and stably; second, a graphics processing unit-based parallel PC algorithm is integrated into the hierarchical framework for fast and robustly estimating high precise disparity map. Additionally, many stages of hierarchical framework, such as padding and reliable evaluation stages, are improved for improving the computational efficiency of disparity estimation system. In a large number of experiments, the results have shown that the efficiency of the proposed algorithm is on average 24 times faster than the compared state-of-the-art methods. Meanwhile, the precision of the proposed algorithm is also superior to or very close to the compared algorithms. The proposed algorithm has been successfully used in a unmanned aerial vehicle three-dimensional retrieval system, and the practice effect has also been verified.