human detection and person counts

[kartikdhoke9923]

Hi, I want to use RuView to count how many people are in a room and show the number on a dashboard.

I can get [2-3] ESP32-S3 boards and I have a WiFi router [in/near the room]. The room is about 2000 sqft and I expect [0 to 20] people. Is this realistic with the current code, and which counter should I use (server persons or cog-person-count)?

Any advice on node placement or known limits for counting 2+ people would help. Thanks!

[have-you-tried-restarting]

Short version: it's realistic for rough headcounts in that range, but not for exact counts, and I'd start with the server counter.

Which counter

• Use the server's persons value first. It uses an adaptive P95 normalisation, so it calibrates to your room on its own, and you can tune the multi-node dedup with /api/v1/config/dedup-factor (feat(sensing-server): adaptive person count — RollingP95 + dedup_factor runtime API #491).
• cog-person-count is the learned counter from ADR-103. That ADR is still marked "Proposed", and it says v0.1.0 ships with "modest absolute accuracy" because of limited training data (ADR-079 P8 follow-up: PCK@20 3.0% → ≥35% requires more paired data + multi-room framing #645). I wouldn't rely on it yet. Compare both on your own room.

Limits to expect (from the README)

• Each AP distinguishes about 3-5 people, and multi-AP scales roughly linearly. 0-20 people in 2000 sqft is at or past the edge for 2-3 nodes, so expect undercounting once people cluster or overlap.
• Treat the number as an estimate, not a precise headcount. Calibrate it against a manual count and tell the dashboard to smooth it (for example a rolling median).

Placement

• Spread nodes around the room perimeter instead of bunching them together, so the paths cross the whole space. More distinct paths help more than more nodes in one spot.
• Keep the router in view of every node.

I haven't tested this on hardware, so check the numbers in your own room. A quick manual count at 2, 5 and 10 people would show how much it drifts.