Rapid speckle image matching method based on GPU

Abstract

The invention discloses a rapid speckle image matching method based on a GPU. According to the invention, a speckle pattern is projected to a to-be-measured object by using a projector; two cameras are used for synchronously collecting projection patterns and carrying out epipolar correction, a computer graphics processor is used for calculating correlation coefficients of the two images pixel bypixel, pixel matching points of the left camera and the right camera are found, then quadratic polynomial fitting is used for finding sub-pixel matching points of the left camera and the right camera,and parallax values are obtained through the sub-pixel matching points. Compared with traditional speckle image matching, the GPU acceleration unit is adopted, so that the image matching time is greatly shortened, and the matching efficiency is improved.

Description

technical field [0001] The invention belongs to the technical field of image registration, in particular to a GPU-based fast speckle image matching method. Background technique [0002] The 3D imaging based on speckle projection is a kind of method based on structured light, which is generally divided into two matching methods: spatial correlation method and time correlation method. Among them, the spatial correlation method is to project a single speckle image to the target to perform 3D reconstruction. The main range of measurement is the dynamic scene or the target object in motion, but its accuracy is low, and this method is easy to be detected in the target. Points are missing in areas with high curvature or edge areas, which is not conducive to obtaining relatively complete 3D reconstruction results; compared with the former, the time correlation method has higher accuracy, but in the process of 3D reconstruction, multiple different images need to be projected to the t...

AI Technical Summary

Problems solved by technology

Among them, the spatial correlation method is to project a single speckle image to the target to perform 3D reconstruction. The main range of measurement is the dynamic scene or the target object in motion, but its accuracy is low, and this method is easy to be detected in the target. Points are missing in areas with high curvature or edge areas, which is not conducive to obtaining relatively complete 3D reconstruction results; compared with the former, the time correlation method has higher accuracy, but in the process of 3D reconstruction, multiple different images need to be projected to the target to be measured. Therefore, it is mainly suitable for static scenes and targets, and is not suitable for the measurement of dynamic targets.

Comparing the two, each has advantages and disadvantages, but no matter how the 3D reconstruction technology develops in the future, the matching efficiency and matching accuracy of the matching algorithm are the focus of its research

However, in the actual application process, it is difficult to achieve real-time stereo matching effect when the calculation amount of stereo matching is large

Images