Vision-based People Counting for Attendance Monitoring System

In this paper, an automatic people detection and counting system using data collected from an over-head camera is proposed. The purpose of this research is to develop a fast and accurate intelligent people counting technique for attendance monitoring systems in offices and lecture rooms. The proposed method includes two stages working sequentially. First, the detection task is executed to find any person presented in the current frame. A deep learning architecture, MobileNetv2-SSD, was used to carry out the detecting phase. If there is any detected person, the tracking phase, which based on visual-tracking techniques, will be initialized, and keeps track of the people's position. Based on the tracked motion path of the detected people, we can determine if there is any person who has entered or exited the room. Therefore, we can monitor the number of attending people. The testing hardware was a Raspberry Pi computer and a camera. This work has been tested on different stages of a day and achieved real-time performance with sufficient accuracy.