balenaLocating-ML

PoC to show indoor BLE triangulation using RaspberryPi sensors and a KNN classifier model

The above floorplan shows the placement of devices. Training data is captured for each room. The training data is then used to sample sensor values to predict, via the KNN, where a BLE beacon is located.

Simple KNN classifier

1. Finds the distance from the unknown item to the training tuples
2. Orders the result from closest to furthest away
3. Finds the K nearest items
4. Identifies the most frequent class in the result set

PoC working

[Click to watch the video]

Future Steps

• Remove the need for a cloud connection/account. All on the edge!
• adding more tags and displaying a grid of where they all are.
• Streaming the telemetry through the KNN classifier and storing tag locations in a heatmap:
• Using the heatmap data to drive a K-Means model to detect abnormal movements + alerting.