ASCENTA.Camera
On-device person detection
A quantized FOMO model processes 240 × 240 images on the STM32H747. The documented test setup reached an 80.0% F1 score.
Back to product
Edge AI · ToF · IMU
Technology without the cloud.
The prototype divides perception and speech between the glasses and a smartphone. Camera images remain on the glasses; processed coordinates, distance and query audio move within the local system.
Sensing and compute
Four systems in one frame.
Time of flight
Distance for each detection
The current warning logic distinguishes observation, proximity warning below one metre and collision alert below 40 centimetres.
IMU
Double-tap interaction
An LSM6DSOX detects a deliberate double tap using acceleration change and a timing window to start a voice query.
Power
Two-cell prototype supply
Two 620 mAh LiPo cells are connected in parallel behind charging and protection circuitry and charge over USB-C.
Process
From light to speech.
The glasses detect a person and measure distance. Prefixed TCP packets carry processed coordinates, distance and query audio over the local network. The smartphone performs local Vosk speech recognition, combines the signals and returns a discreet text-to-speech cue.
BLE and UDP were rejected for the main stream because the prototype encountered excessive audio latency and packet loss. TCP is the current transport; BLE remains for Wi-Fi provisioning.