Deep Gesture Recognition under Data Loss

Authors

Amin Kajbaf, Ehsan Yazdian, Mohammad Ali Akhaee, Ramin Toosi, Saeed Gazor

Abstract

Millimeter-wave radar enables contactless gesture recognition, but real-world sensing often suffers from interference and missing measurements. This paper studies deep gesture recognition under data-loss conditions using FMCW mmWave radar. Building on a multi-environment dataset of 11,161 samples spanning 12 gesture classes and 7 human action classes, collected with an AWR1843BOOST radar across laboratories, corridors, and outdoor settings with static and dynamic secondary targets, we investigate deep learning approaches for robust recognition when radar observations are incomplete or degraded.