Systems and methods for analyzing time series data based on event transitions
Inventors
Chen, Jixu • Tu, Peter Henry • Chang, Ming-Ching • Kim, Yelin • Lyu, Siwei
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10839009-B2
Publication Date
2020-11-17
Expiration Date
Abstract
A method for analyzing time series data to identify an event of interest is provided. The method includes receiving, at a computing device, time series data that includes the event of interest, identifying, using the computing device, a start time of the event of interest and an end time of the event of interest by modeling at least one transitional pattern in the time series data, and categorizing, using the computing device, the event of interest based on the at least one transitional pattern.
Core Innovation
The invention analyzes image time series data to identify an event of interest corresponding to an action performed by a subject. The method receives the image time series data at a computing device and performs identifying and categorizing using modeling and estimated poses. The event is characterized by what occurs and by how the event transitions in time within the image time series data.
A start time of the event of interest and an end time of the event of interest are identified by modeling at least one transitional pattern in the image time series data based on one or more estimated poses of the subject. The transitional pattern includes onset segments and offset segments tied to transitions between neutral events and the event of interest. This treats event boundaries as part of the modeled transitional pattern rather than as independent time cuts.
After modeling the transitional pattern, the event of interest is categorized based on the at least one transitional pattern. The categorization includes categorizing the event of interest as suspicious activity and generating an alert to notify a user. The approach is implemented as a computing-device method, a computing device with a memory device and processor, and computer-executable instructions on at least one non-transitory computer-readable storage media.
Claims Coverage
The independent claims cover three implementation forms: a method, a computing device, and non-transitory computer-readable storage media. Each includes modeling at least one transitional pattern from estimated poses to identify start and end times and then categorizing the event based on the transitional pattern. Four inventive features are identified across the claims.
Transitional-pattern-based event localization and categorization
Receiving image time series data that includes the event of interest corresponding to an action performed by a subject; identifying a start time and an end time of the event of interest by modeling at least one transitional pattern in the image time series data based on one or more estimated poses of the subject; and categorizing the event of interest based on the at least one transitional pattern.
Onset/offset transitional pattern between neutral and event states
Modeling transitional patterns by modeling onset segments and offset segments from estimated poses of the subject that describe transitions between neutral events and an event of interest.
Probability-based start/end identification using neutral events
Identifying a start time and an end time for an event of interest by computing likelihood that the event occurs after a first neutral event and that a second neutral event follows the event of interest.
Suspicious activity categorization with user alert
Categorizing the event of interest as suspicious activity and generating an alert to notify a user.
Across the independent claims, the core claimed coverage is unified: model at least one transitional pattern from one or more estimated poses to identify event start and end times, then categorize the event based on that transitional pattern. Dependent refinements include onset/offset transitional modeling between neutral events and the event of interest, probability-based timing relative to neutral events, and optionally classifying suspicious activity with an alert.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?