A classroom student counting system based on improved context-based face detector

Abstract

Classroom student counting based on surveillance video is the basis for important tasks such as student behavior analysis, resource optimization, school security, and intelligent management. In recent years, with the development of deep learning, the research of face detection has been greatly promoted, but it still cannot solve the problems of difficult recognition of different poses such as tiny faces in the back row, occlusion of students, and head-down in the classroom. This paper designed an intelligent student counting system based on classroom surveillance video to solve the problem of counting people in classroom scenarios. Also the paper proposed a face detector which mixes feature enhancement modules and background-based modules, using Background information including shoulders, bodies and desks, and search for unobstructed, less-occluded, and head-up targets in multiple video frames in the surveillance video based on the relative stillness of the students in the classroom scene, which finally greatly improves the accuracy of the classroom student counting statistics.