Study on Action Recognition Based on Kinect and Its Application in Rehabilitation Training

Abstract

We introducted some basic structures and functions of Kinect briefly, analyzed depth images by aquiring camera body movement of the Kinect body feeling device. Further, we aquired 20 group node data by combinning with the SDK, so we proposed algorithm of gesture recognition by extracting feature vector in this paper. Representation method of action has been proposed based on the human body structure information. That is, according to the key points structure human body characteristic vector, we choosed the feature vector of the combination of vector Angle and vector modulus ratio to create human body posture to describe vector, and represented the action by using gesture description vector sequence. DTW algorithm is mainly studied, on the basis of the traditional DTW algorithm, improved and optimized the DTW (dynamic time warping) algorithm in this paper. Order to improve and optimize the DTW algorithm, we builded the template library operation and finished using improved DTW algorithm for identification matching experiments. Experimental data show that the improved DTW algorithm for simple action recognition rate is higher, and recognition rate is low for more complex action. The action recognition which based on access application is significance and feasibility in rehabilitation training.

DOI: 10.1109/BDCloud.2015.38

3 Figures and Tables

Cite this paper

@article{Li2015StudyOA, title={Study on Action Recognition Based on Kinect and Its Application in Rehabilitation Training}, author={Nianfeng Li and Yinfei Dai and Rong-Quan Wang and Yanhui Shao}, journal={2015 IEEE Fifth International Conference on Big Data and Cloud Computing}, year={2015}, pages={265-269} }