Human Action And Activity Recognition Using Video Sequences

Abstract

Recently, the vision-based understanding in video sequences entices numerous newlinereal-life applications such as gaming, robotics, patients monitoring, content-based newlineretrieval, video surveillance, and security. One of the ultimate aims of artificial newlineintelligence society is to develop an automatic system that can be recognized and newlineunderstand human behaviour and activities in video sequences accurately. Over newlinethe decade, many efforts are made to recognize the human activity in videos but newlinestill, it is a challenging task due to intra-class action similarity, occlusions, view newlinevariations and environmental conditions. newlineTo analyse and address the issue involved in the recognition of human activity newlinein video sequences. Initially, we have reviewed the most popular and prominent newlinestate-of-the-art solutions, compared and presented. Based on the literature newlinesurvey, these solutions are categorized into handcrafted features based descriptors newlineand automatically learned feature based on deep architectures. In this thesis work, newlinethe proposed action recognition framework is divided into handcrafted and deep newlinelearning-based architectures which are then utilized throughout this work by embedding newlinethe new algorithms for activity recognition, both in the handcrafted and newlineautomatic learned features domains. newlineFirst, a novel handcrafted feature based descriptor is presented. This newlinemethod addressed the major challenges such as abrupt scene change phenomena, newlineclutter background and viewpoints variations by presented a novel visual cognizance newlinebased multi-resolution descriptor for action recognition using key pose newlineframes. This descriptor framework is constructed by computation of textural and newlinespatial cues at multi-resolution in still images obtained from videos sequences. A newlinefuzzy inference model is used to select the single key pose image from action video newlinesequences using maximum histogram distance between stacks of frames.