3P1 Case Study.
AI Gesture Recognition.
During COVID-19 lockdown, we were approached to develop a gesture recognition product. Using a 6GHz radar, it would sense distance to a target (human hand, no contact), which would be waved around as a gesture.

Our software development team’s approach
To allow the sensor to be placed at some distance from the device, it used Bluetooth LE HID to present the gestures to a PC or cell phone as a Bluetooth keyboard. To collect the data, we connected the sensor to a PC. We used a custom application to record the sensor output while we recorded gestures over it. We used this to generate a machine learning model. The model was then compressed down to fit into a microcontroller running the TensorFlow Lite MCU engine. This required the AI model to fit into MCU RAM, alongside a BLE stack, which proved to be technically challenging. The model went through a number of iterations to prove itself capable of the performance level the customer desires. Another significant constraint was the CPU of the target – a single M4 core without AI acceleration.
Despite these challenges, the concept was proved, and this was used to begin development of a custom ASIC to handle the task.