HUMAN ACTIVITY RECOGNITION IN EGOCENTRIC VIDEO USING GIST AND INTEREST POINTS DESCRIPTORS

Abstract

In the contemporary world, recognizing the activity that assists old age people, differently able population has become a challenging task. This paper aims at development of a system for recognition of activities of human in egocentric video. In our proposal from the activity videos, we retrieved different features such as Generalized Search Tree (GiST) and interest point features like Scale Invariant Feature Transform (SIFT), Speeded up Robust Features (SURF) and Space Time Interest Points (STIP). Classification of activity is done by classifiers such as Probabilistic Neural Network (PNN), Support Vector Machine (SVM), k-Nearest Neighbor (kNN) as well as our combined SVM with k Nearest Neighbor (SVM+kNN) classifiers. Also, the input selected is multimodal egocentric activities. The results of the study show that SVM+kNN classifier has enhanced performance in comparison to other classifiers already in use.