The method to detect one's action by analysis of sequential view images in everyday life

摘要

In this report, we propose the method to detect and recognize one's antion by analysis of sequential view images in everyday life. In this method, we estimate user's head motion using optical flow by multi resolution images with gaussian filter to reduce the noises. And to acquire a user's hand position and track it from input images, we use the color feature of human hand by the Lab color space. Observing user's head motion and hand motion, we recognize user's action and it's kind, and detect target objects in automatically.