View invariant human action recognition using fourier-based and radon-based point cloud analysis

Human action recognition has been one of the most challenging topics in computer vision during the last decade. This paper presents a novel approach for recognizing view independent human actions based on analysis of Fourier transform and Radon transform of self similarity matrix of features obtained from the action. The proposed feature descriptor is extracted from human point cloud over the time and is based on the key idea that some parts of human body which have a longer distance from the body center are more discriminative for human action recognition purposes. The effectiveness of the proposed method is demonstrated with the experiments on i3DPost dataset.