RVM-based human action classification through Gabor and Haar feature extraction

Human action recognition plays a vital role in surveillance applications. Human action recognition is motivated by some of the applications such as video retrieval, video surveillance systems, human robot interaction, to interact with deaf and dumb people etc. The aim is to analyse the role of Adaboost in the process of recognising the human action by extracting the motion features using optical flow. Adaboost is a supervised learning method used to select the subset of frames with most discriminatory motion features. Saliency point computation is performed to assign a measure of interest to each visual unit. Mean shift algorithm is then used for tracking the objects. Gabor feature is the global feature that includes more detailed information of frequency and orientation. Haar feature is used to show the variation in the pixel. Relevance vector machine classification gives a probabilistic output through Bayesian inference. The proposed system reduces the computation time and provides a higher recognition rate in comparison with existing gentle boost-based recognition system.