Low-Power Gesture Recognition and Airborne Object Sensing Using A Sparse Convex PMUT Array

This work demonstrates a novel motion recognition strategy using a low-cost 2D airborne object sensing system based on a convex, sparsely populated PMUT array with a 120o Field-of-View for real-time gesture detection. A dual-imaging pipeline is proposed to process the received data and reconstruct a 2D image, either with classical Delay-and-Sum or a modified adaptive variant of the Delay-Multiply-and-Sum processing with multipliers of signal pairs. By incorporating bilateration prior to image reconstruction, the proposed sensing system achieves a 120o coverage within a 10-15 cm range using less than half the typical number of PMUTs, reducing blind spots while enhancing cost efficiency and scalability. Dynamic Time Warping is employed to process the images and align temporal patterns across uniquely grouped motion recordings with the features used to train a Random Forest classifier. The proposed system is evaluated with unlabeled gestures considering eight distinct hand movements: upward, downward, down-up-down, left-to-right, right-to-left, wave, stationary hand position and a random class. Compared to dense planar arrays, the proposed 5-channel convex PMUT array system reduces hardware complexity by 75% while maintaining a 95% classification accuracy. The sparse PMUT array design enables a compact, low-power, and scalable solution without the need for bulky sensor infrastructure. This approach represents a promising alternative for gesture-based control in wearable human-machine interface applications.