The Design Of Dormitory Access Management System Based On Machine Learning

Today China is in a stage of continued expansion and keeping open of universities, the campus security issues become increasingly prominent, universities are actively laying the campus security system to ensure the safety of teachers and students in all aspects. There are many universities gradually building and improving of dormitory access management solutions, which is seen as an important part of campus security. Most of the existing access management system rely on Radio Frequency Identification (RFID) technology, such as campus card, to identify students, but in practice, fraudulent use of campus card occurs easily; the data generated during the use of the system is mainly used for querying information, but not utilized effectively. This paper presents a machine learning-based access management system, it learns and analyzes image and behavior data from authorized users to achieve intelligent management of dormitory, and to create a smart and safe campus with state-of-art computer technology.The access management system in this paper consists of a passageway controller, a management client and a server, these parts work together to verify the user's identity according to campus card and face characteristics, learn the user's regular access pattern and companions, and detect abnormal access by learning the user's regular access pattern. The passageway controller is the combination of a microcontroller, a card reader and supplement circuit, it does the preliminary verification of the user's identity. The client also verifies the user's regular access pattern and queries the user's companions. It also uploads the user's data regularly to the server and fetch latest models; the server trains the face recognition model, analyze user's regular access pattern and companions from recent data, and sends those models and parameters together with authorized user list and daily access data to the clients via network communication.This paper first describes the overall design and work flow of each sub-system, followed by a detailed description of the training and application of facial recognition model, regular access pattern model and querying for companions including detailed design and implementation of the client and server.