Efficient Model Based Grid Intersection Detection for Single-Shot 3D Reconstruction

Abstract

The crucial role of pixel correspondences in the process of projector-camera calibration and 3D reconstruction is to determine the relationship between camera and projector view. Consequently, acquiring accurate pixel correspondences with a robust detection method can result in accurate 3D reconstruction for a given scene. Single-shot structured light (SSSL) binary patterns are now commonly projected onto target scenes, where they usually consist of white grid lines separating different square tags ordered in such a way to encode the projector pixels. This paper introduces a new technique for detecting the intersection of grid lines to obtain the pixel correspondences of a given SSSL system, where ultimately the detected intersections represent the pixel correspondences. Our SSSL projected pattern contains horizontal and vertical grid lines which allows us to parameterize the lines fitting the grid lines. Experimental results show that the proposed method offers at least 50 times faster computation than a recent connected component analysis-based method at the cost of sparser pixel correspondences and a slight decrease in the accuracy of the 3D reconstructed point cloud.