Automatic Joints Extraction of Scanned Human Body

This paper presents an automatic method to extract joints accurately from scanned human body shape. Firstly, model is divided into slices by a group of horizontal parallel planes and primary landmarks are determinated according to the slices. Additionally, we propose a model intersection method based on direction of limbs. Model is separated into five segments including torso and four limbs and approximate direction of each segment can be calculated according to the primary landmarks. Five groups of parallel planes perpendicular to corresponding segments are employed to divide each segment into slices, and contours are calculated by intersection detection between planes and triangles. Finally, we propose a circularity function differentiation technique to extract key contours at joints from segments and the joints can be calculated according to centroids of the key contours. The experimental results demonstrate that the method has more advantages than the conventional ones, especially in algorithm's accuracy and robustness.