Semi-global stereo matching method adopting cost fusion and hierarchical matching strategy

Abstract

The invention provides a semi-global stereo matching method adopting a cost fusion and hierarchical matching strategy, and belongs to the field of computer vision. According to the method, a solution is provided for solving the problems that a semi-global stereo matching algorithm is low in precision in an outdoor scene and low in operation efficiency on CPU equipment. Firstly, the fused matching cost is used as the initial matching cost of stereo matching, which is more robust; a four-path clustering method is adopted to optimize the matching cost, so that the matching cost is more accurate; then, calculating and generating a disparity map by using the steps of disparity calculation, disparity optimization and the like; finally, the matching precision and the matching efficiency are optimized in a layered iterative optimization mode, a parallel computing strategy is introduced for a specific CPU in specific matching of each layer, and the computing efficiency is greatly improved. The method provided by the invention not only improves the stereo matching precision of a semi-global stereo matching algorithm in an outdoor scene, but also improves the operation efficiency of the semi-global stereo matching algorithm on a CPU platform, so that the semi-global stereo matching algorithm has more practical significance when being applied to edge CPU equipment.

Description

technical field [0001] The invention belongs to the field of computer vision, and mainly relates to an improved semi-global stereo matching algorithm and its accelerated realization method on a CPU platform. Background technique [0002] Stereo matching is a technique to recover depth information from planar images. Because the binocular stereo vision system simulates the principle of human visual perception, it only needs two cameras, and it can be put into use after distortion correction and stereo correction. It has the advantages of simple implementation, low cost, and can realize measurement under non-contact conditions. Stereo matching algorithms play an important role in binocular vision systems. Stereo matching algorithms can be divided into four types: local, global, semi-global, and deep learning-based. Among them, stereo matching algorithms based on deep learning have gradually increased in recent years. With their ultra-high precision, they perform well on publ...

AI Technical Summary

Problems solved by technology

However, the matching accuracy of the traditional semi-global stereo matching algorithm is still low in complex scenes such as outdoors, and the stereo matching algorithm is similar to a brute force search, and the real-time performance on the edge CPU hardware device is poor. Therefore, in binocular In the stereo vision system, there is an urgent need for a stereo matching algorithm with high matching accuracy and computational efficiency on the CPU platform

Images