Method and system for symmetric recognition of handed activities
Inventors
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Assignees
MemberHinge HealthHinge HealthHinge Health delivers a musculoskeletal care platform that combines digital physical therapy, individualized exercise programs, in-person clinical services, and multidisciplinary care for conditions such as pain, mobility loss, pelvic health, gastrointestinal disorders, and menopause-related symptoms. The platform utilizes remote monitoring, motion analysis, and evidence-based care, prioritizing validated clinical outcomes, data security, and integration with benefits systems for employers and health plans.
Hinge Health delivers a musculoskeletal care platform that combines digital physical therapy, individualized exercise programs, in-person clinical services, and multidisciplinary care for conditions such as pain, mobility loss, pelvic health, gastrointestinal disorders, and menopause-related symptoms. The platform utilizes remote monitoring, motion analysis, and evidence-based care, prioritizing validated clinical outcomes, data security, and integration with benefits systems for employers and health plans.
Publication Number
US-12333432-B2
Publication Date
2025-06-17
Expiration Date
Abstract
This disclosure describes an activity recognition system for asymmetric (e.g., left- and right-handed) activities that leverages the symmetry intrinsic to most human and animal bodies. Specifically, described is 1) a human activity recognition system that only recognizes handed activities but is inferenced twice, once with input flipped, to identify both left- and right-handed activities and 2) a training method for learning-based implementations of the aforementioned system that flips all training instances (and associated labels) to appear left-handed and in doing so, balances the training dataset between left- and right-handed activities.
Core Innovation
The patent discloses an activity recognition approach for left/right asymmetric activities that uses mirror symmetry. A handed activity classifier is deployed for a first handedness and is executed for inference on original input and on horizontally flipped input. A class flip module flips a predicted class from the flipped input back to opposite handedness, and a class aggregator aggregates the results into final left/right outputs.
The patent also discloses a training method that converts training instances associated with one handedness into training instances associated with an opposite handedness. Each training instance is associated with a corresponding class label and a corresponding segment of activity data, and the modifying step transforms the corresponding class label to a class label of a second handedness opposite the first handedness. The modifying step also transforms the corresponding segment of activity data using a transformation associated with the first handedness.
As a result of the modifying, none of the modified training instances are labeled with the first handedness. In particular implementations, the activity data can include pose keypoints, images, and/or video, and the handedness transformation can be specified as a horizontal transformation that reverses an X coordinate while keeping a Y coordinate unchanged.
Claims Coverage
The partial claim set includes three independent claims, each centered on training by transforming class labels and transforming associated activity-data segments for an opposite asymmetry/handedness, such that none of the modified training instances remain labeled with the first asymmetry.
Training activity classifier by opposite-handedness label and activity-data transformation
Identifying training instances associated with a first handedness, modifying each training instance by transforming the corresponding class label to a class label of a second handedness opposite the first handedness and transforming the corresponding segment of activity data, and training the activity classifier using the modified training instances, wherein none of the modified training instances are labeled with the first handedness.
Training a machine learning model by opposite-asymmetry class label and activity-data transformation
Identifying training instances associated with a first asymmetry, modifying each training instance by transforming the corresponding class label to a class label of a second asymmetry opposite the first asymmetry and transforming the corresponding segment of activity data, and training the machine learning model using the modified training instances, wherein as a result of said modifying none of the modified training instances are labeled with the first asymmetry.
Non-transitory medium training by opposite-asymmetry transformation with parameterized weights
A non-transitory medium with instructions that cause operations comprising identifying training instances associated with a first asymmetry, modifying each training instance by transforming the corresponding class label to a class label of a second asymmetry opposite the first asymmetry and transforming the corresponding segment of activity data, and training a machine learning model such that one or more weights are parameterized based on the modified training instances and the corresponding transformed segments of activity data.
Across the independent claims, the core inventive coverage is training based on transforming class labels and transforming corresponding activity-data segments into an opposite asymmetry/handedness representation, with the resulting modified instances not remaining labeled with the first asymmetry/handedness, and with model weights parameterized based on the modified instances.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?