Suspicious Human Activity Recognition for Mobile Robot

Abstract

This work aims to develop a real-time application for surveillance robot in indoor environment, to recognize human suspicious activity using deep learning architecture applied on visual data. To learn different classes of activity, a combination of three deep-learning algorithms have been used, based on the idea that a set of classifiers improves machine learning accuracy. The ensemble of classifier has been trained on a collected public dataset containing videos of human activities with normal and suspicious behavior. The model is then, evaluated in real time scenarios, and the experimental results show that the proposed system has the potential to benefit applications in surveillance robots