System and method for activity classification
One or more computing devices, systems, and/or methods are provided. In an example, a method comprises receiving, by a device, incoming motion data from a motion sensor, generating, by the device, an incoming embedding vector based on the incoming motion data, generating, by the device, a predicted embedding vector based on the incoming embedding vector, assigning, by the device, an activity classification based on the predicted embedding vector, and modifying an operating parameter of the device based on the activity classification.
1 . A method comprising:
receiving, by a device, incoming 3-axis time series motion data from a motion sensor;
projecting, by a processor in the device, the 3-axis time series motion data into an incoming embedding vector using an embedding model;
generating, by the processor in the device, a predicted embedding vector based on the incoming embedding vector and using a Kalman filter;
assigning, by the processor in the device, an activity classification based on the predicted embedding vector; and
modifying, by the processor in the device, an operating parameter of the device based on the activity classification.
2 . The method of claim 1 , comprising:
filtering the incoming 3-axis time series motion data to generate filtered 3-axis time series motion data;
sampling the filtered 3-axis time series motion data into a window of predetermined length; and
generating the incoming embedding vector based on the window.
3 . The method of claim 1 , comprising:
rejecting the incoming embedding vector based on a distance between the incoming embedding vector and a previous value of the predicted embedding vector.
4 . The method of claim 1 , wherein projecting, by the processor in the device, the 3-axis time series motion data into the incoming embedding vector using the embedding model comprises:
generating using the embedding model a mean component of the incoming embedding vector and a variance component of the incoming embedding vector.
5 . The method of claim 1 , comprising:
training the embedding model using a quadruplet loss technique.
6 . The method of claim 1 , wherein assigning, by the processor in the device, the activity classification comprises:
assigning the activity classification using a k-nearest neighbor model.
7 . The method of claim 1 , wherein receiving the incoming 3-axis time series motion data comprises:
receiving at least one of accelerometer data, gyroscope data, or magnetometer data.
8 . A device, comprising:
a motion sensor configured to generate incoming 3-axis time series motion data; and
a processor configured to execute instructions to facilitate performance of operations comprising:
projecting the 3-axis time series motion data into an incoming embedding vector using an embedding model based on the incoming motion data;
generating a predicted embedding vector based on the incoming embedding vector and using a Kalman filter;
assigning an activity classification based on the predicted embedding vector; and
modifying an operating parameter of the device based on the activity classification.
9 . The device of claim 8 , wherein the operations comprise:
filtering the incoming 3-axis time series motion data to generate filtered 3-axis time series motion data;
sampling the filtered 3-axis time series motion data into a window of predetermined length; and
generating the incoming embedding vector based on the window.
10 . The device of claim 8 , wherein the operations comprise:
rejecting the incoming embedding vector based on a distance between the incoming embedding vector and a previous value of the predicted embedding vector.
11 . The device of claim 8 , wherein the operations for projecting the 3-axis time series motion data into the incoming embedding vector using the embedding model comprise operations for:
generating using the embedding model a mean component of the incoming embedding vector and a variance component of the incoming embedding vector.
12 . The device of claim 8 , wherein the operations comprise operations for:
training the embedding model using a quadruplet loss technique.
13 . The device of claim 8 , wherein the operations for assigning the activity classification comprise operations for:
assigning the activity classification using a k-nearest neighbor model.
14 . The device of claim 8 , wherein the motion sensor comprises at least one of an accelerometer data, a gyroscope, or a magnetometer.
15 . A non-transitory computer-readable medium storing instructions that when executed facilitate performance of operations comprising:
receiving incoming motion data from a motion sensor;
projecting the 3-axis time series motion data into an incoming embedding vector using an embedding model based on the incoming motion data;
generating a predicted embedding vector based on the incoming embedding vector and using a Kalman filter;
assigning an activity classification based on the predicted embedding vector; and
modifying an operating parameter of a device based on the activity classification.
16 . The non-transitory computer-readable medium of claim 15 , wherein the operations comprise:
rejecting the incoming embedding vector based on a distance between the incoming embedding vector and a previous value of the predicted embedding vector.
17 . The non-transitory computer-readable medium of claim 15 , wherein the operations for projecting the 3-axis time series motion data into the incoming embedding vector using the embedding comprise operations for:
generating using the embedding model a mean component of the incoming embedding vector and a variance component of the incoming embedding vector.
18 . The non-transitory computer-readable medium of claim 15 , wherein the operations comprise operations for:
training the embedding model using a quadruplet loss technique.