3D reconstruction based on pose transformation of human joints

Data obtained from depth camera scans of the human body often includes noise, and registering consecutive frames can lead to errors due to inevitable non-rigid movements during scanning. To address this, we propose a method for 3D human body reconstruction based on human joint points. Initially, the Kinect DK sensor captures data from different angles, which is then preprocessed to extract the main body point cloud. Subsequently, segmented human body data, combined with pose-related joint points, guides the point cloud data of the body's surface. Using an enhanced iterative closest point algorithm, we achieve precise registration of the main body point cloud, resulting in an accurate 3D human body model. Experimental results show that this method rapidly generates realistic, finely detailed, and accurately dimensioned 3D human body models.