Intelligent Counting System for Classroom Numbers Based on Video Surveillance
Mingxi Liu, Xinze Zhang, Yiran Han · 2020 IEEE International Conference on Power, Intelligent Computing and Systems (ICPICS) · 2020
The classroom is the main place for students to attend classes and self-study. Because the school classroom is limited, it often takes more time for students to find a classroom with no classes or few people. Therefore, the real-time statistics of the number of students in the classroom is of great significance for strengthening the school spirit and assisting teachers to understand the classroom situation. And it is meaningful to develop a classroom population statistics system to help students find a suitable self-study room quickly. In order to solve the problem that it is difficult to automatically count the number of students in the classroom and can not be real-time, according to the principle of face detection and referring to the convolution neural network target detection framework, a special single detection architecture and a scale allocation strategy suitable for this framework are proposed. the difficulty of automatic counting of students in the classroom is solved. In this paper, we use a specific data set for verification, achieve advanced detection results, and finally get a statistical accuracy of 97.8%. In classroom surveillance images, individual students are small targets, but the existing R-FCN target detection algorithm based on convolution neural network is difficult to detect small targets. In order to solve this problem, a series of improvements are made on the basis of R-FCN, which greatly improve the ability of R-FCN target detection algorithm to recognize small targets. It is verified on the self-made data set, and the accuracy is up to 89.4%.