3 Method and apparatus for extracting features for machine learning on gesture recognition from 3D skeleton information

摘要

Conventional motion recognition based on skeleton information avoids a method of experimentally extracting a feature value according to a change of absolute position information of a joint and utilizes human body mechanic to extract a useful human body feature for a relative position and an angle between joints. Motion recognition by the existing method is valid only for limited motions. On the other hand, in the present invention, various complicated motions can be recognized and can be easily used as features for machine learning. Moreover, the method of the present invention is a feature extraction method designed to have real-time responsiveness, thereby being also used for motion recognition application requiring urgent action such as safety-related or social problems.