Human Action Recognition and Localization in Video at Contextual Level

Abstract

This paper presents an investigation into the design of a computer based human action recognition system aimed at localizing and recognizing moving objects in a controlled environment. A system based on the object identifier and shape descriptor techniques is proposed. Automated visual perception of the real world by computers requires classification of observed physical objects into semantically meaningful categories (such as ‘animal’ or ‘person’ or ‘objects’).

This paper proposes a partially-supervised learning framework for classification of the moving objects especially vehicles and pedestrians that are detected and tracked in a variety of far-field video sequences, captured by a static camera. Introduction of scene-specific context features (such as image-position of objects using xml) is done to improve classification performance in any given scene. Along with this, a scene-invariant object annotation has been done to adapt this contextual model for new scenes.