US 12,366,925 B2
Robustifying radar-based gesture recognition solution using contextual information
Vutha Va, Plano, TX (US); Priyabrata Parida, Dallas, TX (US); Saifeng Ni, Santa Clara, CA (US); Anum Ali, Frisco, TX (US); and Boon Loong Ng, Plano, TX (US)
Assigned to Samsung Electronics Co., Ltd., Suwon-si (KR)
Filed by Samsung Electronics Co., Ltd., Suwon-si (KR)
Filed on Jan. 16, 2024, as Appl. No. 18/414,200.
Claims priority of provisional application 63/523,474, filed on Jun. 27, 2023.
Prior Publication US 2025/0004561 A1, Jan. 2, 2025
Int. Cl. G06F 3/01 (2006.01); G01S 7/41 (2006.01)
CPC G06F 3/017 (2013.01) [G01S 7/415 (2013.01); G06F 3/011 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
obtaining a gesture sample via a gesture classifier and a prediction output from the gesture classifier indicating a type of gesture that the gesture classifier associated with the gesture sample, wherein the gesture sample includes parameters of features extracted from radar signals based on user motion in performance of a gesture;
determining whether to obtain feedback for a target gesture associated with the gesture sample and the prediction output;
in response to a determination to obtain the feedback, obtaining the feedback for the target gesture, the feedback including a label indicating a type of gesture;
determining whether the gesture sample associated with the label is valid based on comparison to a dataset of pre-existing validated gesture samples and a distance threshold; and
in response to determining that the gesture sample associated with the label is valid, determining whether to update a model for the gesture classifier using the validated gesture sample as a training sample.