Method and system for activity classification
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.
1. A method for classifying an activity of a subject comprising:
receiving input data of the subject resulting from the activity, the input data including pressure data;
applying a deep learning neural network to the input data based on weights and biases, resulting in classified activity data;
training the deep learning neural network for updating the weights and biases; and
communicating the classified activity data to a user.
2. The method of claim 1 wherein the input data further comprises data from an accelerometer, a gyroscope, a seismograph, a thermometer, or a humidity sensor, an altimeter, a GPS, a video camera, a heart rate sensor, an oxygen sensor, a breathing rate sensor, a blood glucose sensor, a fatigue measuring device, a limb position measuring device, a blood pressure monitor, an ECG, a lung function meter, an alcohol level sensor, or drug level sensor.
3. The method of claim 1 wherein the input data comprises separate sensors on a support matrix located to receive the input data from different portions of the subject; and the weights and biases are initially determined with reference to locations of the separate sensors on the support matrix.
4. The method of claim 1 wherein training the deep learning neural network comprises:
confirming the activity, resulting in a defined activity and corresponding classified actual activity data;
defining a loss function between the classified actual activity data and the classified activity data; and
updating the weights and biases for mitigating the loss function; and
wherein confirming the activity comprises on or many of the following operations:
prompting the subject to perform the defined activity;
receiving a confirmation input that the subject performed the defined activity;
receiving the confirmation input from the subject; and
receiving the confirmation input from an individual other than the subject.
5. The method of claim 1 wherein the deep learning neural network comprises a translationally invariant neural network,
the translationally invariant neural network comprising:
at least one convolutional layer; and
at least one fully connected layer subsequent to the at least one convolutional layer;
and wherein the fully connected layer comprises attribute data of the subject, the attribute data of the subject concatenated with the fully connected layer.
6. The method of claim 1 wherein training the deep learning neural network comprises receiving training data, the training data comprises synthetic input data generated from empirical input data.
7. The method of claim 6 wherein the synthetic input data is generated from the empirical input data by applying to the empirical input data one or many of the following operations: time-shifting, magnitude-scaling and spectral magnitude-scaling.
8. The method of claim 1 wherein communicating the classified activity data to the user comprises one or many of the following operations: displaying the classified activity data; providing tactile stimulus to the subject; and storing the classified activity in a database.
9. The method of claim 1 wherein the activity comprises activity indicative of an imminent fall and communicating the classified activity data to the user comprises a tactile or other neuroplastic stimulus to prompt the user to correct the activity and avoid a fall.
10. The method of claim 1 wherein the activity further comprises a fall and further comprising communicating the classified activity data of the fall to a third party.
11. The method of claim 1 further comprising applying a time window to the input data to provide time-segmented input data; and wherein applying the deep learning neural network to the input data comprises applying the deep learning neural network to the time-segmented input data.
12. The method of claim 11 further comprising weighting the time-segmented input data to provide weighted input data and wherein applying the deep learning neural network to the input data comprises applying the deep learning neural network to the weighted input data.
13. The method of claim 1 further comprising applying an event detection filter to the input data to provide event-classified input data; and wherein applying the deep learning neural network to the input data comprises applying the deep learning neural network to the event-classified input data.
14. A system for classifying activity of a subject comprising:
a sensor module comprising a pressure sensor, the sensor module for generating input data during the activity, the input data including pressure data;
a processor configured for receiving the input data, the processor configured for executing a method comprising:
applying a deep learning neural network to the input data based on weights and biases, resulting in classified activity data; and
training the neural network for updating the weights and biases.
15. The system of claim 14 wherein the sensor module comprises at least two pressure sensors; and the weights and biases are based on known relationships between the at least two pressure sensors.
16. The system of claim 14 wherein the sensor module further comprises an accelerometer, a gyroscope, a seismograph, a thermometer, or a humidity sensor, an altimeter, a GPS, a video camera, a heart rate sensor, an oxygen sensor, a breathing rate sensor, a blood glucose sensor, a fatigue measuring device, a limb position measuring device, a blood pressure monitor, an ECG, a lung function meter, an alcohol level sensor, drug level sensor, or any other type of sensor that measures a level of impairment.
17. The system of claim 14 wherein the deep learning neural network comprises a translationally invariant neural network; and
the neural network comprises:
at least one convolutional layer; and
at least one fully-connected layer subsequent to the at least one convolutional layer.
18. The system of claim 14 wherein applying a deep learning neural network to the input data comprises applying a time window to the input data to provide time-segmented input data; and applying the deep learning neural network to the time-segmented input data.
19. The system of claim 14 wherein applying a deep learning neural network to the input data comprises applying an event detection filter to the input data to provide event-classified input data; and applying the deep learning neural network to the event-classified input data.
20. The system of claim 14 wherein training the deep learning neural network comprises:
confirming the activity, resulting in a defined activity and corresponding classified actual activity data;
defining a loss function between the classified actual activity data and the classified activity data; and
updating the weights and biases for mitigating the loss function; and
wherein the training data comprises synthetic input data generated from empirical input data.