Inspiration
What it does
It uses a recognition model to identify different t-shirt colours to distinguish different members of DurHack stuff on the automated distributed CCTV system that restricts access to certain building areas for the non-authorised personnel. Each node is fully autonomous, meaning the processing isn't happening on the central hub and is instead being done on separate nodes which are interconnected with each other. The system is resilient to partial outages: when a node gets cut off, the next nodes are informed about it and raise their security threat level due to inability to check for an intruder at the level being served by the one.
How we built it
The pre-trained YOLO Pose model with Coco dataset is being fed with samples in different lightning of the targeted t-shirts that are identified real-time on the ip-camera live streaming from phones set at different locations. When the alarm goes off, the security alert is being sent to Telegram bot stored on a different machine.
Challenges we ran into
Our first node was based on Raspberry Pi Zero W with Camera 2 but due to the camera's inability to capture good-quality images of long-distance objects.
Accomplishments that we're proud of
Our recognition model achieved more that 70% accuracy across different lighting settings and different distances from the camera.
What we learned
Throughout the different stages of our project we explored different ideas and changed our initial plan many times starting from the rescue drones system. This has taught us not to focus on one thing, but to be able to learn mid-development and be open to various different ideas
