US 11,657,281 B2
Method and system for symmetric recognition of handed activities
Colin Brown, Saskatoon (CA); and Andrey Tolstikhin, Montreal (CA)
Assigned to Hinge Health, Inc., San Francisco, CA (US)
Appl. No. 17/593,270
Filed by Hinge Health, Inc., San Francisco, CA (US)
PCT Filed Mar. 12, 2020, PCT No. PCT/IB2020/052249
§ 371(c)(1), (2) Date Sep. 14, 2021,
PCT Pub. No. WO2020/188424, PCT Pub. Date Sep. 24, 2020.
Claims priority of application No. CA 3036836 (CA), filed on Mar. 15, 2019.
Prior Publication US 2022/0148296 A1, May 12, 2022
Int. Cl. G06V 10/774 (2022.01); G06V 40/10 (2022.01); G06V 40/20 (2022.01)
CPC G06V 10/7747 (2022.01) [G06V 40/107 (2022.01); G06V 40/20 (2022.01)] 20 Claims
OG exemplary drawing
1. An activity classifier system comprising:
a capture device which captures activity data;
a transform module which receives the activity data from the capture device and transforms the activity data using a symmetric transformation;
an activity classifier that identifies activities using received activity data having a determined symmetry and outputs an inferred activity class or plurality of class probabilities;
a class flip module which receives the inferred activity class or class probabilities from the activity classifier using the transformed activity data from the transform module and flips the activity classes using a transformation corresponding to the symmetric transformation to produce flipped inferred activity classes; and
a class aggregator that receives the flipped inferred activity classes and the inferred activity class from the activity classifier using the activity data from the capture device, and generates an output activity class based on which activity class is dominant or a combined plurality of class probabilities.