Evaluation of human action recognition techniques intended for video analytics

Human Action Recognition (HAR) in video plays a vital role in today's world. The aim of the HAR is to build a self-analysis system for on-going events from video data and understand the behavior of a person. This is the key functionality of intelligent video surveillance system and has wide range of applications. The applications include visual surveillance systems, robotics, health-care systems, human-computer interaction, ambient intelligence, video indexing, traffic management etc. that include interaction between humans and objects. Smart surveillance systems that can aid the human operator in real-time threat detection can be developed by applying video Analytics. Video analytics helps in interpreting the video to identify and decide spatial & temporal events not based on a one image. There are various methods to recognize actions and complex activities in a video which are reviewed in this paper. Actions are distinguished by simple motion pattern executed by a single human such as walking, running, hand-waving etc. Methods like bag of features, multiple instance markov model, MRF method are evaluated with experiments on these actions. Activities are more complex and involve harmonized actions among group of humans such as working on laptop, entering/exiting a room etc. SFG method is used with experiments on these activities. This paper collectively generalizes and interprets the various challenges, methodologies of HAR and brings out the best technique for HAR. Results of related papers are discussed elaborately with performance parameters like Precision, Recall and Accuracy.