Human Activity Recognition using Depth Body Part Histograms and Hidden Markov Models

Abstract

This paper proposes a novel approach for human activity recognition based on body part histograms and Hidden Markov Models. From a depth video frame, body parts are segmented first using a trained random forest. Then, a histogram for each body part is combined to represent histogram features for a depth image. The depth video activity features are then applied on hidden Markov models for training and recognition. The proposed method was superior when compared with other conventional approaches.