Human Detection in Video for Security Surveillance Systems

Human detection in video is an important task for security surveillance systems. It could help reducing time and effort when tracking a human in a long video such as finding a thief in video of surveillance camera. This study proposes an approach to build an application of recognizing and detecting human in videos. The purpose of this application is to trace, investigate, or review the events that have taken place from the security camera when the need arises. In this work, we have compared there methods (pre-trained Yolov7, self-defined sequential model, and VGG16 Transfer Learning) for detecting human in the video. Experiments are built using data set which is extracted from security camera with several angles and different resolutions. The approach has achieved results with the highest accuracy of 97% for human recognition, thus, it would be a good solution for practice.