Edge AI Gesture Detector

Demonstrates how to perform on-device inference on a Blecon Smart Beacon device

Overview

The Edge AI Gesture Detector example demonstrates how ML workflows can leverage Blecon connectivity. Inference can be performed on a Blecon-enabled device and results logged and transmitted upon connection. Models can then be updated using an OTA firmware update. This application integrates an Edge Impulse-generated model.

Source Code

The source code for this app is provided in the sensor-ml directory of the blecon-oem-device-firmware repository: https://github.com/blecon/blecon-oem-device-firmware/tree/main/apps/sensor-ml

Edge Impulse Model

The application integrates the gesture detection ML model from the corresponding Edge Impulse example. Inference results will also be displayed using the device's LEDs if the device's button has been pressed within a configurable period:

• Red: Up/Down motion

• Green: Wave

Results are reported within log messages in the ml-sensor namespace.

Supported Device

This code supports the MOKO SMART L02S-BCN Smart Beacon. The Zephyr board name is mkbnl02sn/nrf54l15/cpuapp.

Configuration Options

Configurations for each board type is found in the configuration sub-directory.

Reporting Frequency

• REPORTING_PERIOD_SEC - Configures the maximum frequency with which the device submits logged data to the cloud. If connectivity is not available when the device requests a connection, the device will continue to request

LED mapping

LED mappings are configured in the app.overlay file for the selected board under the chosen node. The blecon-led-gesture-1, blecon-led-gesture-2, and blecon-led-status can be assigned to LED device nodes in the devicetree.

Blecon LED Status Library

To display Blecon connection status on LEDs, the BLECON_LED_LIB option must be enabled. For devices with limited memory, this library can be disabled to reduce the application's memory footprint.