オプティカルフローを用いた方向分類による人物動作の同期解析

摘要

For the human motion detection by camera, it is usually needed to use sensors and/or markers. We propose a method that does not handle any sensors nor markers, instead, that adopts optical flows. By applying this method, we can analyze synchronization between human motions and classify moving direction of crowds. The optical flows are estimated by the Lucas-Kanade method. The result of the optical flows is assigned to a hue circle, and the histogram of the hue gives information to identify human motions. Utilizing this color information makes our method robust to the environmental changes. We defined the combined Bhattacharyya coefficient and use it to detect the synchronization of the human motions. In this paper, two persons dance choreography and the results before and after the practices are quantitatively evaluated. The crowds' classification is performed depending on the moving directions of the crowds. We also describe the method to detect unusual acts. These functions are available to score the group performances in games and to detect the persons doing normal and abnormal actions.%計算機の高速化や高画質カメラの普及により，複雑な画像処理を，高速に実現できる環境が整ってきた．そのような環境下で，人の行動に着目した研究が多く行われている．たとえば，画像処理による複数人物の追跡・解析手法として，Mぬm8♂1）が，簡潔な尺度を用いたものを提案している．また，加mらのは，群集追跡法についての展望を論じている．複数人物間にお