Automated detection of similar human actions using motion descriptors

As computing becomes ubiquitous in our modern society, automated recognition of human activities emerges as a crucial topic where it can be applied to many real-life human-centric scenarios such as smart automated surveillance, human computer interaction and automated refereeing. In this research study, a motion descriptor is constructed based on the extraction of optical flow features across consecutive frames for the classification of human activities. A histogram of features is derived from images taking into account the solely local properties embedded within the motion map. Feature selection which is based on the proximity of instances belonging to the same class is performed to obtain the most distinctive features. Experimental results carried out on the Weizmann dataset confirmed the potency for the proposed method with a high recognition rate of 95.02 % to distinguish between different basic human action classes such as running, walking, waving and jumping. The dataset is made of 19 basic actions for 9 different subjects. Further experiments are conducted to assess the ability of the proposed approach to recognize similar actions based on the intra and inter class distribution analysis.