Human Activity Recognition Using Mixture of Gaussians and Pair-wise Oriented Local Binary Pattern

Abstract

Tracking and recognizing human activities from video is a very challenging task in the field of Computer Vision. In this paper, we aim to recognize human activities by coping with the existing challenges. At first, the background and the foreground of images in videos are detected using the mixture of Gaussian distributions and the binary silhouette images are obtained. We propose a feature descriptor named Pair-wise Oriented Local Binary Pattern (POLBP) and an improved version of DLBP feature descriptor for images. POLBP is capable of encoding more information than intensity differences of LBP by incorporating orientation information with the intensity difference. This Pair-wise Oriented Local Binary Pattern (POLBP) extracts local orientation information from binary silhouette images. These feature vectors are sent to Support Vector Machine (SVM) classifier for classification. The proposed method has been used in the area of Human Activity Recognition (HAR) and the result of recognition rate is very encouraging.