Method and system for activity classification
Inventors
Cheng, Chun Hing • EVERETT, Julia Breanne • PURDY, MICHAEL TODD • STEVENS, Travis Michael • VIBERG, DAVID ALLAN • YEE, DALE BARRY
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12307368-B2
Publication Date
2025-05-20
Expiration Date
Abstract
A method and system for activity classification. A pressure sensor receives input data resulting from physical activity of a subject performing an activity. The input data includes pressure data from at least one pressure sensor, and may include other data acquired through other types of sensors. A deep learning neural network is applied to the input data for identifying the activity. The neural network is trained with reference to training data from a training database. The training data may include empirical data from a database of previous data of corresponding activities, synthesized data prepared from the empirical data or simulated data. The training data may include data from physical activity of the subject being monitored by the system. Different aspects of the neural network may be trained with reference to the training data, and some aspects may be locked or opened depending on the application and the circumstances.
Core Innovation
The invention classifies an activity of a subject by receiving input data resulting from the activity, where the activity includes an activity indicative of an imminent fall and the input data includes pressure data. A deep learning neural network is applied to the input data based on weights and biases to produce classified activity data, wherein the classified activity data includes classified predicted fall data. The approach generates a fall prediction query using the classified predicted fall data, where the fall prediction query is negative or positive, and communicates an alert to a user when the fall prediction query is positive.
The invention further includes training the deep learning neural network for updating the weights and the biases. Training uses monitoring of the subject and a confirmation input during training of the deep learning neural network for a defined activity indicative of an imminent fall. A loss function is defined between classified actual activity data and classified predicted fall data, and the network weights and biases are updated to mitigate the loss function.
The invention supports refinement of classification timing by using a rolling time window or an event window for event detection, optionally with a weighted time window. It also supports verifying and responding to an imminent-fall indication by collecting additional input data after the activity indicative of an imminent fall, detecting a recent fall using the additional input data, and communicating a recent fall alert to emergency services or a caregiver. Alerts can be communicated visually, audibly, or by generating a tactile or other neuroplastic stimulus intended to elicit an instinctive response in the subject.
Claims Coverage
The disclosed claim set includes two independent claims, each centered on pressure-based activity classification using a deep learning neural network to generate a positive/negative fall prediction query and communicate an alert when positive. Across the dependent claims for each independent claim, the document specifies confirmation during training, event-window control, and optional alert/output modalities and downstream actions.
Pressure-based imminent-fall activity classification with deep learning
Receiving input data including pressure data from an activity indicative of an imminent fall, applying a deep learning neural network based on weights and biases to produce classified activity data including classified predicted fall data, and generating a fall prediction query that is negative or positive.
Communicating a positive fall alert to a user
Communicating an alert to a user if the fall prediction query is positive, in the context of the deep learning-based classified predicted fall data.
Neural-network training using confirmation input and loss function
Training the deep learning neural network for updating the weights and the biases using a confirmation input and a loss function between classified actual activity data and classified predicted fall data, with weights and biases updated to mitigate the loss function.
System architecture with sensor module and processor implementing the classification
Providing a sensor module comprising a pressure sensor for collecting input data during an activity indicative of an imminent fall, using a processor to receive the input data, apply a deep learning neural network based on weights and biases to produce classified activity data including classified predicted fall data, generate a negative or positive fall prediction query, and communicate an alert to a user when the fall prediction query is positive.
Both independent claims require pressure data input, deep learning neural network classification using weights and biases to produce classified predicted fall data, generation of a negative/positive fall prediction query, and alert communication to a user upon a positive prediction. The document’s dependent claim set further emphasizes confirmation-based training with a loss function, controllable use of an event window for event detection, and verification via additional input data with optional emergency-services/caregiver alerting and multiple alert modalities.
Stated Advantages
Communicating an alert to a user when the fall prediction query is positive.
Risk level can be identified representing the likelihood of a fall occurring.
Detecting a recent fall using additional input data and communicating a recent fall alert to emergency services or a caregiver.
Corrective interaction can be prompted using visually, audibly, or tactually/neuroplastically communicated alerts.
Documented Applications
Fall prediction and imminent fall alerts, including generating tactile or other neuroplastic stimulus to prompt correction of an activity indicative of an imminent fall.
Detecting a recent fall and communicating a recent fall alert to emergency services or a caregiver.
Interested in licensing this patent?