A Survey on Automated Human Action Recognition Using Multi-view Feature

Abstract

Recognizing the human action plays a significant role in surveillance cameras. Usually cameras are situated at distant place and convey actions in form of signals at one particular place. This paper presents a framework for recognizing a sequence of actions based on multi-view video data. To depict various actions activities performed in various perspectives, view-invariant feature is being used. The features of multi-view are extracted from various temporal scales, which are demonstrated using global spatial-temporal distribution. The proposed system performs is designed to work on cross tested datasets wherein the system doesn’t require retraining for same scenario that occurs multiple times.