Recognizing human actions using novel space-time volume binary patterns

In this paper, we propose a novel feature type, namely Motion Binary Pattern (MBP) and different computation strategies for the well-known Volume Local Binary Pattern (VLBP). MBPs are a combination of VLBPs and Optical Flow. By combining the benefit of both methods, a simple and efficient descriptor is constructed. Motion Binary Patterns are computed in the spatio-temporal domain while the motion in consecutive frames is described. Finally, a feature descriptor is constructed by a histogram computation. Volume Local Binary Patterns are a feature type to describe object characteristics in the spatio-temporal domain. But apart from the computation of such a pattern further steps are required to create a discriminative feature. These steps are evaluated in detail and the best strategy is shown. For MBPs and VLBPs, a Random Forest classifier is learned and applied to the task of human action recognition. The proposed novel feature type and VLBPs are evaluated on the well-known, publicly available KTH dataset, Weizman dataset and on the IXMAS dataset for multi-view action recognition. The results demonstrate challenging accuracies in comparison to state-of-the-art methods.