運動中發生部分遮蔽之動態手勢辨識

Abstract

我們提出一種利用電腦視覺 (Computer Vision) 來辨識動態手勢的方法. 在我們的系統中, 使用攝影機 (CCD) 來抓取手部影像,利用手掌及手指的 幾何形狀和特 性作特徵抽取,得到手指端點的位置,利用運動調和法 (Motion Correspondence), 得到運動軌跡再造出向量,再使用類神 經網路中的 supervised fuzzy adaptive Hamming net (SFAHN)來學習和 辨識.針對十六種手勢作測試,結果顯示辨識率相當高; 其中 包括了運動起始點和終點相同,僅運動軌跡不同的相似手勢, 以及手指會發生部份遮蔽 (occlusion) 的手勢. We propose a method for using Computer Vision to recognize dynamic gestures. In our system, we use a charge-coupled device (CCD) camera to capture hand images, we then find the positions of fingertips using the geometrical shapes and characteristics of the human hand, taking advantage of ``Motion Correspondence'' to find the motion trajectories, and calculate vectors from start to stop positions of motion trajectories. We interpret the vectors using a supervised fuzzy adaptive Hamming net (SFAHN). Simulation results show the proposed method has a recognition rate approaching 100% for two pairs of patterns with different motion trajectories but the same start and stop positions, even when fingertips in some gesture patterns were partially occluded.