A new 6D motion gesture database and the benchmark results of feature-based statistical recognition

A motion gesture can be represented by a 3D spatial trajectory and maybe augmented by additional 3 dimensions of orientation. Depending on the tracking technology in use, the 6D motion gesture can be tracked explicitly with the position and orientation or implicitly with the acceleration and angular speed. In this work, we first present a motion gesture database which contains both explicit and implicit 6D motion information. This database allows us to compare the recognition performance over different tracking signals on a common ground. Our main contribution is to investigate the relative effectiveness of various feature dimensions in motion gesture recognition. Using a simple and primitive recognizer, we evaluate the recognition results of both explicit and implicit motion data. In our experiments, both user dependent and user independent cases are addressed. We also propose two general techniques to improve the recognition accuracy: smoothing and the temporal extension. Our pilot study produces benchmark results that give an insight into the attainable recognition accuracy with different tracking devices.