Sensor unit with on-device unsupervised learning and classification
AssigneeSTMicroelectronics S.r.l.
InventorsSwapnil Sayan SAHA, Mahesh CHOWDHARY
Life of this patent
Estimating the term…
Abstract
An electronic device includes a sensor unit. The sensor unit includes a sensor and low power, low area sensor processing unit. The sensor processing unit performs an unsupervised machine learning processes to learn to recognize an activity or motion of the user or device. The user can request to learn the new activity. The sensor processing unit can request that the user remain stationary for a selected period of time before performing the activity. The sensor processing unit records sensor data while the user performs the activity and generates an activity template from the sensor data. The sensor processing can then infer when the user is performing the activity by comparing sensor signals to the activity template.
Claims22
A method, comprising:
- receiving, with an electronic device, a request from a user to train a sensor unit of the electronic device to recognize a first motion;
- prompting, with the electronic device, the user to remain stationary for a selected period of time;
- prompting, with the electronic device, the user to perform the first motion after remaining stationary for the selected period of time;
- generating, with the sensor unit, sensor data while the user performs the first motion;
- calculating a variance for each of three axes from the sensor data;
- selecting, from the three axes, two axes having a largest variance;
- generating, with the sensor unit, a first template for the first motion based on the sensor data from the two axes;
- and
- storing the first template in a memory of the sensor unit.
An electronic device, comprising:
- a user input;
- a display;
- and
- a sensor unit including:
- a sensor;
- a learning module configured to perform an unsupervised machine learning process to generate a template representing a first motion of the user based on first sensor data from the sensor by calculating a variance for each of three axes from the sensor data, selecting, from the three axes, two axes having a largest variance, and generating the template based on the sensor data from the two axes having the largest variance;
- and
- an activity detection module configured to detect whether the user is performing the first motion by comparing new sensor data from the sensor to the first template.
A method, comprising:
- receiving a request to train a sensor unit of an electronic device to recognize a motion of the electronic device;
- generating, with the sensor unit, a template for the motion by performing an unsupervised machine learning process including:
- recording sensor data during the motion;
- calculating a variance for each of three axes from the sensor data;
- selecting, from the three axes, two axes having a largest variance;
- generating, with the sensor unit, the first template for the first motion based on the sensor data from the two axes;
- and
- inferring, with the sensor unit after generating the template, that the electronic device is undergoing the motion based on sensor data recorded after generation of the template.
Dependencies are inferred from the claim text.