Human Joint Orientation Descriptor Based on Geometric Algebra and Its Application

Wen-ming CAO, Yi-tao LU

Abstract


Motion recognition is becoming more and more widely used in various applications. In this paper we propose a novel descriptor to describe human skeleton based on geometric algebra (GA) that decomposes the skeleton posture into the rotations of skeleton parts. In this model, all body bones are rotated from the same original states. We formulate the rotation operator in 3D GA space, which can be used to describe the rotations of human body bones. Then we select the most informative rotations of body bones and joint angles to represent the skeleton. We train a Gaussian Naïve Bayes classifier which can recognize the motion type of a single input frame captured from video sensors. After the motion type is determined, we find the most similar posture in the motion sequence database using the distance based on posture orientations and joint angles. And finally, we calculate the posture difference to give users the calibration advice. Our experimental results have shown the high accuracy and effectiveness of our method.

Keywords


Motion recognition, Gaussian Naïve Bayes, Geometric algebra, Motion calibration


DOI
10.12783/dtcse/cscbd2019/30081

Full Text:

PDF

Refbacks

  • There are currently no refbacks.