DEVLOG #01

Hi, long time no see.

Over the past few weeks, I have been developing two new approaches to
presence detection. The main issue was that the system could not reliably
distinguish between an empty room and a room with me standing inside :(.
To solve this, I first developed D5 and then D6. D5 is already available on
GitHub. D6 is not available yet because it is still under development and
has not been fully validated with real measurements and blind tests.

D5 — Presence Detection

D5 was designed to determine whether a room is empty or occupied. It compares
current CSI measurements with a previously recorded empty-room reference.
However, D5 mainly evaluated changes in one direction. In practice, a person’s
presence can cause CSI values to either increase or decrease, depending on the
receiver and propagation path. As a result, D5 behaved more like a motion
detector and could miss a person who remained still. While it significantly
improved presence detection, it was not robust enough for reliable real-world
use.

D6 — Presence Detection and Position Classification

D6 builds on D5 by evaluating deviations from the empty-room reference in
both directions, making it much more effective at detecting people who
remain stationary.
Once a person has been detected, D6 attempts to classify their position.
Instead of calculating arbitrary coordinates, it identifies one of nine
predefined locations (P01–P09) that were previously trained using CSI
recordings.
A position is only reported if:

• the hardware setup is valid and sealed;
• all receivers provide fresh and valid CSI data;
• the correct transmitter has been verified;
• a person has been detected;
• a valid position index is available; and
• the classification result is sufficiently confident.
  If any of these conditions are not met, D6 deliberately reports no
  position instead of inventing one.
  The software pipeline is largely complete, but the system still needs to be
  validated using a new sealed setup, fresh training recordings, and independent
  blind tests before D6 can be released on GitHub.

New Interface

I also redesigned the standard RuView interface as a clean, white,
Samaritan-inspired UI based on the AI system from the TV series Person of
Interest.