Segmented Depth Motion Map Assisted Human Action Recognition from Depth Action Videos

Abstract

Recently, Action Recognition from depth videos has gained a huge research interest due to the discovery of low-cost sensors which can capture the depth information of 3D structural objects in the scene. Based on this inspiration, we have developed an action recognition framework Based on Segmented Depth Motion Map (S-DMM) and Binary Histograms. Unlike the conventional DMM which didn’t focused on the segmentation, this work segments the entire video into a set of segments and then employs DMM to represent the motion and shape cues of human body. Further the DMM is subjected to feature extraction through Local Binary Patterns and finally fed to classification unit. To check the performance, two different classifiers such as K-Nearest Neighbor (K-NN) and Support Vector Machine (SVM) are employed. The developed framework is applied to a standard depth action dataset and the obtained results demonstrate the effectiveness compared to conventional methods.