Machine Learning Based Human Activity Recognition in Video Surveillance

Abstract

In the current time there is a massive shift of technology from analog to digital. According to previous facts, three quarters of world data was in analog form. But now, approximately it is preserved in digital form. Due to this rapid growth of technology there is a massive growth of image data being generated by surveillance cameras. Automated anomaly detection has become necessary in order to detect the presence of any kind of dangerous activities such as robbery, road accidents and many more. Recently, machine Learning approaches have achieved the state of the art results in many tasks related to the automated anomaly detection process. The objective of this paper is to propose one such efficient method for anomaly detection. The proposed approach works with multiple instance learning with I3D feature extractor and crop augmented images. The obtained AUC prove the proposed approach to be superior of previous models.