AccuMotion
Takuya Sakai, Wataru Fujimura, Robert W. Songer, Takayuki Kosaka, Akihiko Shirai · 2012
This article contributes to the improvement of natural user interfaces (NUI) using depth-based kinematics recognition tools like the Microsoft Kinect. The proposed method, "AccuMotion" is comprised of tracking sequential key poses as accumulated motion. The AccuMotion recognition algorithm is based on multiple kinematics evaluation functions that evaluate the dot products of target bone structures with the user's kinematic bone structure. Each function continuously outputs a similarity ratio between its respective target and input from the user's kinematic data. Target bone structures are defined by the developers as ideal or arbitrary values. This method is effective for a wide range of users due to its use of a kinematics data that allows for differences of length in user bones. The same target poses apply to a wide range of users through the use of a generic algorithm and user profiling. As an experiment, the recognition function was tested for four directional inputs indicated by user arm movements. The results suggest AccuMotion is suitable for navigating presentation software such as slideshows and video players with solid stability.