JARDCS

Human Activity Recognition through Localized Motion Analysis

Awadhesh Kr Srivastava, K.K. Biswas and Vikas Tripathi

Abstract:

Human activity recognition is one of the most dynamic and eminently useful research areas of computer vision due to its massive range of applications like surveillance, health support system etc. Despite of a lot of contribution by various researchers over a decade, the recognition of various activities is still a challenging task. There are various surveillance approaches available in literature for recognizing activities. In this paper, we present a robust framework for human activity recognition which can be used in numerous applications. In the proposed framework features are extracted from motion image, generated from inter-frames motion analysis. Due to imbalanced nature of data, random forest is used as a classifier. To show the robustness of proposed framework it is applied on the publicly available benchmark dataset Human Motion Database (HMDB). The classification performance of proposed method is 49.6% achieved in terms of accuracy.

Year: 2018

Pages: 264-274