A fast stereo-matching method for a speckle pattern based on grid and gradient constraints

Abstract

In the existing binocular stereo-matching methods for a speckle pattern, the pixel-by-pixel correlation weight calculation across the entire global disparity range is time-consuming. To address this issue, this paper presents a fast stereo-matching method for a speckle pattern based on grid and gradient constraints. The grid pattern is projected onto the measured object. The grid features are extracted to form the local disparity constraints, which narrow the search ranges. Moreover, the image gradient is used in selecting the pixels involved in the correlation weight calculation, which reduces the redundant computations. 3D reconstruction experiments of paper and plaster models were employed to evaluate the matching accuracy and speed of the proposed method. Furthermore, the experiments of measuring displacement were designed to validate the accuracy of our method. In contrast to the existing methods, our approach increased the matching speed by 66% on average while ensuring the matching accuracy.