Human Detection in Surveillance Video using Deep Learning Approach

Object detection (OD) is a key challenge in computer vision and received a lot of importance from last decade. The main goal of object identification is to determine location of specific target class object with good accuracy and label object in an image with appropriate class. Basically, deep learning-based OD techniques are splits up into two parts:-1) two-stage identifier and 2) one-stage identifier. In this article, we give a short and vital review of current developments in deep learning-based OD. The purpose of our research work is to identify humans in a surveillance video and divide the video into active and non-active frames to save the time of watching the entire video. We implemented two novel methods of object detection, YOLOv4 and Mask-RCNN because of their good and promising performance. After extensive experiments, it is noted that Mask-RCNN performs better than YOLOv4 on a developed dataset. Mask-RCNN achieves 85% accuracy and exhibit better performance as compared to YOLOv4 which achieve 65% accuracy only.