High performance set of features for human action classification

[abstract]

Human action classification has many applications ranging from providing descriptive labels to video segments to recognition of common dance moves and suspicious behavior in video surveillance cameras. The most common method for handling human action classification is to determine a common set of optimal features and then apply a machine-learning algorithm to classify them. In this paper we explore combining sets of different features for training an ensemble using random subspace with a set of support vector machines. We propose two novel descriptors for this task domain: one based on Gabor filters and the other based on local binary patterns (LBPs) which are applied on the mask images in the well known Weizmann dataset. We then combine these two sets of features with the histogram of gradients. Applying this approach, we have obtained an accuracy of 97.8% using the 10-class Weizmann dataset and a 100% accuracy rate using the 9-class Weizmann dataset. These results are comparable with the state of the art. The main result of this study is to show that by combining sets of relatively simple descriptors it is possible to obtain results comparable to those obtained using more sophisticated approaches. Our simpler approach, however, offers the advantage of being less computationally expensive.