Method and apparatus for recognition of sensor data patterns
AssigneeBosch Sensortec GmbH
InventorsAdam K. Tilton, Shane T. Ghiotto, Prashant G. Mehta
Life of this patent
Estimating the term…
Abstract
Methods and systems for learning, recognition, classification and analysis of real-world cyclic patterns using a model having n oscillators, with primary frequency ω1, ω2, . . . , ωn. The state of the oscillators is evolved over time using sensor observations, which are also used to determine the sensor characteristics, or the sensor observation functions. Once trained, a set of activity detection filters may be used to classify a sensor data stream as being associated with an activity.
Claims19
A method comprising:
- receiving sensor data comprising a plurality of sequential sensor measurements associated with an unknown activity;
- obtaining, for each of a plurality of activity filters, activity parameters including a set of activity frequencies, wherein the set of activity frequencies corresponds to a plurality of sets of particles, each activity frequency associated with a respective set of particles configured to oscillate in accordance with the associated activity frequency, each set of particles having an associated set of particle phases;
- updating, for each set of particles, the associated set of particle phases by updating the phase of each particle of the set of particles according to i) its associated activity frequency and ii) a prediction error based on a difference between at least one of the plurality of sequential observation measurements and an observation prediction from an observation model operating on the associated set of particle phases;
- determining, for each of the plurality of activity filters, an activity probability based on the plurality of sets of particle phases for the corresponding activity filter, wherein each activity filter is associated with a predetermined activity, and wherein the activity probability for each activity filter indicates a probability that the received sensor data is indicative of the predetermined activity associated with that activity filter;
- and
- outputting an indicator that identifies an activity filter having a highest activity probability among the plurality of activity filters.
An apparatus comprising a computer readable medium having instructions stored thereon that when executed cause a processor to perform the functions of:
- receiving sensor data comprising a plurality of sequential sensor measurements associated with an unknown activity;
- obtaining, for each of a plurality of activity filters, activity parameters including a set of activity frequencies, wherein the set of activity frequencies corresponds to a plurality of sets of particles, each activity frequency associated with a respective set of particles configured to oscillate in accordance with the associated activity frequency, each set of particles having an associated set of particle phases;
- updating, for each set of particles, the associated set of particle phases by updating the phase of each particle of the set of particles according to i) its associated activity frequency and ii) a prediction error based on a difference between at least one of the plurality of sequential observation measurements and an observation prediction from an observation model operating on the associated set of particle phases;
- determining, for each of the plurality of activity filters, an activity probability based on the plurality of sets of particle phases for the corresponding activity filter, wherein each activity filter is associated with a predetermined activity, and wherein the activity probability for each activity filter indicates a probability that the received sensor data is indicative of the predetermined activity associated with that activity filter;
- and
- outputting an indicator that identifies an activity filter having a highest activity probability among the plurality of activity filters.
Dependencies are inferred from the claim text.