Human Action Recognition from Motion Trajectory using Fourier Temporal Features of Skeleton Joints

Spatial and temporal dynamics of human being create a rich set of information to process and analyze very important human activities that can attract the attention of various discipline of real-life applications. Finer view of data modality for human body can be characterized by skeleton, contour, silhouette and articulated geometrical shapes. All modalities of a video are being affected by challenging vision problems like view invariance, occlusion and camera calibration at varying scale. In this work, we focused skeleton based human activity recognition and proposed motion trajectory computation scheme using Fourier temporal features from the interpolation of skeleton joints of human body. This is accomplished by considering the human motion as a trajectory of skeleton joints. Experimental observations ensure that this approach outperforms many of state of the arts. The proposed algorithm is tested on the MSRAction3D benchmark dataset. For this we have experimented on three action sets AS1, AS2 and AS3 categorized from the dataset. After different training and testing samples, this gives overall accuracy 95.32 % for human action recognition.