EDGE GESTURE PREDICTION Using Artificial Intelligence

Artificial intelligence for gesture detection and classification has transformative potential across industries, from healthcare to manufacturing and consumer devices.

- ASHWINI KUMAR SINHA

It enables tasks like motion tracking in smartwatches, anomaly detection in machinery, and vehicle predictive analysis. Implementing AI on compact devices like the IndusBoard Coin—a 3cm innovation with built-in sensors—demonstrates how these capabilities can be brought to edge devices with limited resources.

We create a simple model for gesture detection, designed to classify motions commonly used in smartwatches, phones, and exercise-related devices. The model is optimised to run efficiently on a microcontroller (MCU), overcoming constraints like limited RAM and storage while delivering reliable AI performance.

Only a few components, detailed in the accompanying table, are needed to build the device. Fig. 1 shows the prototype testing setup, showcasing how AI can seamlessly execute tasks on the edge.

An online tool named Edge Impulse was used to collect the data, train a machine learning model, and deploy it on the IndusBoard MCU. An Edge Impulse account must be created using the link https://edgeimpulse.​com. The developer and student mode offers a free account, while the enterprise version is available for factories or industrial use.

After creating an Edge Impulse account, the Edge Impulse Data Forwarder must be installed on the PC. For macOS, Linux, or Windows, the installation can be completed using the terminal (cmd for Windows) by running the following commands sequentially:

- Installing Node.js:

node -v

npm -v