Room Occupancy Monitoring with Hellbender's Tendryl Camera
Wynne Technologies partnered with Hellbender to build RoomState, a privacy-protecting occupancy sensor on the Tendryl, turning a single edge AI camera into the data source facilities teams can use to efficiently manage their HVAC systems.
Partner Profile
Wynne Technologies is an AI-first product engineering consultancy led by Adam Wynne, who previously directed the innovation accelerator at Bosch’s Azena camera platform. RoomState is Wynne Technologies’ camera-based occupancy sensor, built to give facilities managers the accurate data they need to build heating and cooling schedules around real usage instead of a fixed timetable.
Project Challenges
Commercial buildings heat and cool rooms on a fixed schedule, regardless of occupancy, and traditional motion sensors don’t fix that problem. Passive infrared sensors miss people who sit still, can read a room full of people as empty a few minutes into a meeting, and can’t tell the difference between one occupant and nine. Wynne Technologies needed a vision platform that could solve what motion sensors can’t, while meeting a much higher bar for privacy since a camera, unlike a PIR sensor, is capable of capturing identifiable images of the people in the room. Key requirements included:
- Accurate, Continuous Counting: Deliver a reliable and accurate person count with position and confidence every few seconds.
- Privacy by Structural Design: No image data could leave the device, and the protections needed to be enforced at every level of software.
- No Gateway for a Single Room: The camera needed to be both sensor and collector, with no extra hardware to purchase, install, and secure.
- Real Compute Headroom: Continuously run a person-detection model with enough compute margin for more sophisticated use cases later.
- Field-Ready Reliability: Operate unattended over PoE, survive power loss without corrupting data, and keep operating through network outages.
Results and Recap
Working with the Tendryl — Hellbender’s integrated edge AI camera that combines a Raspberry Pi CM5 with a Hailo-8 AI accelerator — Wynne Technologies built RoomState as a single-device occupancy sensor with no second box to install. Hailo’s runtime, hailo-apps, and the Raspberry Pi camera stack arrived already installed and working together, and Hailo’s own detection demo ran live from the camera at over 26 frames per second with zero dropped frames on the first attempt. That headroom let Wynne Technologies run a full YOLOv11-large detector against a five-second sampling interval with room to spare, using standard Python tooling throughout: uv for packaging, the Pi camera stack for capture, Hailo’s runtime for inference, and Streamlit for the dashboard. No platform-specific code was required.
Privacy was enforced structurally rather than by policy: detections carry a class, bounding box, and confidence — no images, faces, names, or tracking across time.
Wynne Technologies had a stable, deployable proof-of-concept in ten days, including product definition, architecture design, and a full dashboard rebuild around a schedule-versus-measured comparison view.
Key Features
- Integrated Single-Device Design: The Tendryl’s Raspberry Pi CM5 and Hailo-8 accelerator serve as sensor and collector in one unit, requiring no other hardware.
- Privacy-Protecting by Structure: No image data ever leaves the device; the record format has no field a frame could occupy, and detections carry no identity information.
- PoE-Ready Field Deployment: One cable for power and data simplifies installation in a new building to little more than running power.
- Headroom for What’s Next: With ample compute power for RoomState, the deployment has a clear path to classroom occupancy, desk-level density, and dwell-time use cases built from the same data.
Impact
Partnering with Hellbender let Wynne Technologies focus entirely on the RoomState application without integrating and fighting unfamiliar hardware. Wynne Technologies describes the Tendryl development experience as free of the surprises that often stall developers on other smart camera platforms, taking only ten days to build and deploy a fully engineered, privacy-verified proof-of-concept. With their initial problem now solved, RoomState now has a clear path to the room-level energy savings that motivated the project, and a hardware platform ready to support future enhancements.