A Comprehensive Analysis on Unconstraint Video Analysis Using Deep Learning Approaches

Abstract

Unconstraint video analytics are important in visual learning. Unconstrained videos contain complex content with various artifacts, variable lengths, and different operating environments. Human activity plays an important role in the video, abundant in archives and becoming more prevalent on the Internet. Various methods are employed for action recognition under constrained conditions, but huge attention to complex actions and realistic applications is substantially needed. Complex movements consist of sequences of simple movements that have long temporal structures. Complex actions in the same class exhibit large variations in class-interior appearance and behavior due to complex temporal structures, confusing backgrounds, changes in viewpoint, and changes in movement speed. Thus, feature representation and classification of complex motions are determined to be challenging in unconstrained video analysisdue to complex temporal structures. This article presents a comprehensive analysis to solve this problem using deep learningapproaches.