Suspicious Activity Detection from Videos using YOLOv3

Human activity detection for video system is an automated way of processing video sequences and making an intelligent decision about the actions in the video. It is one of the growing areas in Computer Vision and Artificial Intelligence. Suspicious activity detection is the process of detecting unwanted human activities in places and situations. This is done by converting video into frames and analyzing the activities of persons from the processed frames. Human detection has always been a challenging problem as human bodies are non-rigid and changes shape arbitrarily. Human recognition and detection in indoor as well as outdoor environment is a quite complex task due to the various problems that arises such as- poor lightening conditions, variation of poses etc. In this paper, YOLOv3 is used to detect different suspicious activities like bag-snatching, lock-breaking etc. Our system has a very good processing speed as well as good accuracy of detection.