Combined image block and DBSCAN processing for large field of view structured light point cloud denoising

Stripe projection structured light systems are widely used in fields such as industrial inspection and virtual reality. With the gradual growth of the demand for sizeable field-of-view three-dimensional morphology reconstruction, the expansion of the redevelopment area will trigger additional environmental factor interactions, which may lead to an increase in noise points during the point cloud reconstruction process. To reduce the complexity of operations on 3D point cloud data, we propose a combined preprocessing method for phase field and binocular matching two-dimensional data. First, we divide the absolute phase field data computed from the left and right viewpoints into multiple image blocks and utilize the statistical information of each image block for outlier detection and rejection. Second, calculate the distance information between binocular matching points, construct distance indexed scatter data based on this information, use the DBSCAN clustering algorithm for outlier detection, and backtrack to reject the outlier data. Finally, combined with the camera calibration information, the reconstruction of the 3D shape of the measured object can be realized. The experimental results show that, compared with gaussian filtering, bilateral filtering, and statistical filtering algorithms, our method is more effective in removing the noise points in the point cloud data while retaining the target point cloud more completely, thus improving the reconstruction quality of the point cloud data with a large field of view.