Point Cloud-Based Preprocessing Algorithms for Human Pose Estimationaper

In human pose estimation algorithms, the introduction of point cloud data allows for more accurate geometric information of the human body surface without the need for texture mapping. Point clouds are sparse, which effectively reduces storage and processing costs. However, point clouds also exhibit characteristics such as noise, data incompleteness, and processing complexity. To address these issues, a lightweight point cloud data preprocessing method is proposed in this paper. In this work, the background point cloud, ground point cloud, and human body point cloud are first segmented to obtain a clean human body point cloud dataset. Next, denoising techniques are applied to this dataset, removing random noise around the point sets. Finally, the point cloud is downsampled based on the number of points required for human skinning, thereby reducing the sample size and improving processing speed. Experimental results have verified that this method effectively improves the real-time performance and accuracy of human pose estimation algorithms using point clouds.