3D Human Curve Skeleton Extraction Based on Solid-State LiDAR

Abstract

Human curve skeleton extraction is an important research topic in computer vision. In this paper, we propose a 3D human curve skeleton extraction method for point cloud, unlike the skeleton extraction methods in the literature that use an omnidirectional scanned point cloud for processing, the point cloud used in this paper is solid-state LiDAR scanning data with a fixed distance range. First, the human body point cloud is extracted by using geometry constraints. Next, the human body point cloud is contracted using point cloud normal vectors to increase the separability of data from different parts of the body. Then, the point cloud is projected to the image plane for image skeleton extraction and back-projected to the human point cloud to obtain the initial skeleton. Lastly, the final curve skeleton is further optimized using the Laplacian contraction method. The proposed method is faster than other contraction method, and the experimental results on real captured data demonstrate that the proposed method achieves better performance compared to the existing algorithms.

Keywords

• Human skeleton extraction
• point cloud
• Laplacian contraction