Human activity recognition using shape moments and normalized fourier descriptors

This paper presents a video based system for recognizing different continuous human activities in real time using a single stationary camera. The proposed system has a motion descriptor that contains shape moments as well as normalized Fourier descriptors. The main idea of this system is to detect moving objects in each frame and associate the detections to the same object over time. We employ an efficient thresholding technique to segment the objects of interest. The second stage of the proposed system is to extract the object features. The first feature part is constructed by extracting the object contour and estimate its normalized Fourier descriptors. The other part is extracted by the shape moments. We combined the shape moments to produce an invariant feature that is invariant to rotation, translation and scaling. We adopt the multi-class support vector machines (SVM) and Naive Bayes classifiers. The multi-class SVM shows better performance than the other method with a recognition rate up to 94.46%. We evaluated activity recognition on 325 videos of thirteen distinct Human activities (e.g., Walking, Running, Jumping, Hand-waving, Bending and some suspicious activities like Kicking, Punching, Fall floor and shooting gun, etc.) recorded for 260 different persons. Experimental results on three data set Weizman, KTH and HMDB validate the proposed system reliability and efficiency.