TIER 1
Occivar™ reference SoC
Jetson-class with our pre-flashed model. Recommended for net-new pilots. PoE, ceiling- or wall-mount. From AUD $390 per sensor. See the full occupancy sensor spec.
01 · Technology · Edge AI
On-device intelligence. Nothing leaves the sensor.
Frames are inferenced where they're captured. The model is small enough to run on a Jetson-class SoC, fast enough to count people in real time, and dumb enough not to know what a face is in the first place.
02 · Architecture
Camera. SoC. Counts. That's the whole pipeline.
No cloud video. No optional cloud video. No "would you like to enable cloud video?" Just the three boxes below, in order.
01 · CAMERA
Wide-angle sensor
Standard CMOS. Outputs raw frames at 12–30 fps to the SoC over MIPI. Frames never reach a network interface.
02 · EDGE SoC
Inference on-device
Jetson-class or equivalent. Runs the model. Emits a count vector + density grid. Deletes the frame.
03 · COUNTS
JSON to the platform
Counts, deltas, threshold events. Encrypted in transit. Stored aggregated only. No raw video ever existed to upload.
← < 60ms end-to-end →
03 · Numbers
Latency, accuracy, and what we never collect.
Inference latency
< 60 ms
Camera → counts, per frame, p99
Counting accuracy
97.8%
vs ground-truth, well-lit indoor
Video frames stored
0
On device or in cloud
Faces in any database
0
We don't have a face model
PII fields collected
0
Not stored. Not transmitted.
Bandwidth per sensor
≈ 200 B/s
Counts only. JSON over HTTPS.
04 · Hardware
Hardware-agnostic. We support what your operations team is willing to mount.
TIER 2
Bring-your-own edge
If you have x86/ARM compute already, we'll ship the runtime. Linux-only. Containerised. Same on-device guarantee.
TIER 3
Existing CCTV (read-only)
Where compliance permits, we can run the model against an existing camera RTSP feed on a co-located edge box. Frame-discard contract still holds.
05 · Engineering deep-dive welcome