Interacting with Soli - Exploring Fine-Grained Dynamic Gesture Recognition in the Radio-Frequency Spectrum

Explore a novel machine learning architecture designed for radio-frequency based gesture recognition in this 20-minute conference talk from the ACM Symposium on User Interface Software and Technology. Delve into high-frequency, short-range radar sensing using Google's Soli sensor, which offers unique capabilities in resolving fine motions and segmenting gestures in range and velocity spaces. Learn about an end-to-end trained combination of deep convolutional and recurrent neural networks that can detect a rich set of dynamic gestures and resolve small finger motions with high accuracy. Discover how this algorithm achieves an average 87% recognition rate on 11 challenging dynamic gestures, generalizes well across multiple users, and runs efficiently on commodity hardware at 140 Hz using only CPU processing.