Frostbyte is a computer vision system built to monitor cold storage facilities without any cloud dependency. It runs a YOLOv8 object detection model on an NVIDIA Jetson Nano, processing camera feeds in real time to detect open freezer doors, missing stock, and personnel without PPE.
Temperature anomalies are correlated with visual events. If a door is left open and the zone temperature starts rising, the system escalates from a local buzzer to an MQTT alert sent to the facility management dashboard.
All inference runs on-device at around 25 FPS. Frames and events are logged locally, with optional sync to an S3 bucket over cellular for remote sites. The whole unit runs on a single 10W power budget.
Key Features
- Real-time object detection at 25 FPS on edge hardware
- Open-door and missing-stock anomaly detection
- Temperature correlation with visual events
- Local logging with optional S3 sync over cellular
- Runs on a 10W power budget
- MQTT alerts to facility management dashboard