A contextual approach for modeling activity recognition

Abstract

In this paper, we propose a contextual approach for modeling human activity recognition. Activity recognition is performed using motion estimation based on context. Here contextual information is derived from motion, which is predicted from previous frames. This greatly enhances the process of activity recognition, by setting up a particular scenario which helps in constructing the activity. Context is acquired with the help of external inputs which surround an activity and help towards accurate reasoning about that activity. Context Modeling for any object can be done in terms of its relationship to other objects, called as contextual associations that lead towards accurate estimate of object position and presence. Here our focus is on vision based activity recognition. This process involves efficient feature extraction and subsequent classification for image representations. Classification accuracy is enhanced through Support Vector Machine (SVM) classifier, used along with Principle Component Analysis.