A room, seen by Wi-Fi
A drawing of the idea. The waves and the strip are made up, not recorded CSI, and the three choices below are not the classes the model was trained on.
The person moves, and the signal at the receivers keeps changing with them. The change over time is what a CSI spectrogram holds, and what the classifier reads as activity.
- Teal: the Wi-Fi signal, leaving the router on the left for the four receivers on the right.
- Amber: the person, and what they change in the signal. In the strip under the room, time runs left to right, and the more amber a cell is, the more the signal is changing then.
The problem and the idea
Detects human presence and activity from Wi-Fi channel state information (CSI), with no camera. Team project for Engineering Clinics II (Winter 2025/26).
What I built
- Built the Flutter app that shows live presence and activity
- Helped the team build the custom biquad antenna
- Team pipeline: 4× ESP32-S3 receivers, Hampel filtering, CSI spectrograms and an EfficientNetV2-S classifier
The result
- 4ESP32-S3 receivers
From the signal to the app
A team project. Every link says who built it.
Wi-Fi signalBuilt by the team
It starts with a Wi-Fi signal crossing the room. No camera is involved.
ESP32-S3 and biquad antennaI helped build the antenna
4 ESP32-S3 receivers and a custom biquad antenna pick the signal up.
CSI spectrogramsBuilt by the team
The channel state information (CSI) goes through Hampel filtering and is turned into spectrograms.
EfficientNetV2-SBuilt by the team
An EfficientNetV2-S classifier reads the spectrograms for presence and activity.
Flutter appI built this
The app shows presence and activity, live.
Stack
- Flutter
- Dart
- ESP32-S3
- EfficientNetV2-S