Towards Classification Based Human Activity Recognition in Video Sequences

Recognizing human activities is an important component of a context aware system. In this paper, we propose a classification based human activity recognition approach. This approach recognizes different human activities based on a local shape feature descriptor and pattern classifier. We have used a novel local shape feature descriptor which is integration of central moments and local binary patterns. Classifier used is flexible binary support vector machine. Experimental evaluations have been performed on standard Weizmann activity video dataset. Six different activities have been considered for evaluation of the proposed method. Two activities have been selected at a time with binary classifier. These are walk-run, bend-jump, and jack-skip pairs. Experimental results and comparisons with other methods, demonstrate that the proposed method performs well and it is capable of recognizing six different human activities in videos.