Application of neural network for human actions recognition

Abstract

In this paper we have proposed human actions recognition methodology. The main novelty of this paper is application of neural network (NN) trained with the parallel stochastic gradient descent to perform classification task on multi-dimensional time-varying signal. The original motion-capture data consisted of 20 time-varying three-dimensional body joint coordinates acquired with Kinect controller is preprocessed to 9-dimensional angle-based time-varying features set. The data is resampled to the uniform length with cubic spline interpolation after which each action is represented by 60 samples and eventually 540 (60 × 9) variables are presented to input layer of NN. The dataset we used in our experiment consists of recordings for 14 participants that perform nine types of popular gym exercises (totally 770 actions samples). The averaged recognition rate in k-fold cross validation for different actions classes were between 95.6 % ± 9.5 % to even 100 %.