As part of the Smart Classroom project, we have developed prototypes of a smart access control system using RFID, a smart airconditioning control system and a smart light control system. These prototypes are working in our lab in Nightowl right now. The central processing unit for each room is a Pine64 1Gb, which runs NodeRED to control the business logic of all room devices. An interesting addition to this system could be to use camera's to detect human presence in the room, and potentially detect unauthorized access to the room. Machine learning and image processing tools could be used to achieve this.
The goal of this thesis project is to research the available techniques to achieve this with minimal hardware requirements. Ideally we could detect the presence of people in the room using the 1.2GHz Quad-core Arm Cortex A53 that runs on the Pine64, with 1Gb of ram and a ARM Mali400MP2 Dual-core GPU. For more info see the Pine64 documentation.
Interesting starting information on using deep learning and OpenCV for human detection is on Medium and in the follow-up article also on Medium.
More information about the product can be found on the Smart Classroom project page.
Tasks
- Research the available technologies for doing near real-time (<=1s) human detection
- Test the most viable solution with minimal hardware requirements
- Implement a proof concept application that detects people and sends an MQTT message to NodeRED when a person is detected.
- Test the application's viability both on a desktop Corei5 quad-core iwht NVidia GTX 560Ti GPU and on the Pine64