Robust Human Motion Estimation Using Detailed Shape and Texture Models (特集 川人学習動態脳プロジェクト)

There is a vast application field for a low cost vision system that can track and estimate human motion. While specialized optical and magnetic tracking devices are in regular use nowadays, human motion capture from video recordings is still considered to be a very difficult problem. Although textured polygonal models were used in the past for the estimation of human finger movements [5] and for the head tracking [2] [6], they have not been applied yet to the problem of capturing full-body movements. The main reason is that texture of fingers and faces remains similar over a wide range of persons. On the other hand, textured body models must take into account the person's clothing and therefore always require an initial calibration phase to ensure the accuracy of the model. Recent progress in the acquisition of textured models [3] makes such models more practicable also for full-body tracking.