Histogram of spatio temporal local binary patterns for human action recognition

Recognizing human action from video sequences has lots of applications that make it an interesting research subject. Motion History Image (MHI) is a good spatio-temporal template to represent the distinctive profile of an action using a single image. However, in this paper, we use Local Binary Patterns (LBP) to extract the highlighted features from the spatio-temporal template and formulate them as a histogram to make the feature vector. Rather than MHL we use Directional MHI (DMHI) for this purpose. We also use shape feature taken from selective silhouettes and concatenate them with LBP histograms. We measured the performance of the proposed action representation method along with some variants of it by employing Weizmann action dataset and found reasonably higher accuracy for practical use.