Research on Natural Interactive 3D Registration Algorithm Based on the Detection of Human Hand Features

摘要

Aiming at hand feature detection and recognition problem in the natural interaction, especially for the problem that finger was ignored or undetected in the current method, we propose, an improved Adaptive Mean-SHIFT(CamShift) algorithm which combined skin color detection to track and detect hand information accurately. In terms of in the natural interaction, the hand's position and posture could be acquired effectively when it's moving and flipping. We proposed a hand feature point extracting method that based on improving corner feature curvature detection method, then divided these feature points into different group effectively, using weighted algorithm to describe these feature point. This method improve the accuracy of detection and ensure the completion of 3D registration. Experimental result shows than this algorithm detect more accurate than traditional mean-shift algorithm, has stronger robustness for the target occlusion and posture change video sequence.