Temporal Combining for Mobile Activity Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing mobile device technologies cannot effectively utilize accelerometer signals to provide sophisticated assistance to users, as they often face a trade-off between accurate and timely activity classification, leading to delays and inaccuracies, especially when users transition between activities quickly.

Innovation Solution

A method and apparatus that determine activity likelihood function values for past epochs based on sensor signals, combining them to infer the present activity with adjustable latency settings, allowing for simultaneous classifications with different latency and accuracy demands using techniques like Temporal Voting, Maximum Likelihood, and Infinite Impulse Response filtering.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If activity classification is performed with high accuracy using sensor signals, then measurement precision is improved, but processing time increases causing delays in timely activity detection

Engineering Contradiction:

Improveactivity classification accuracyVSAvoidprocessing delay

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary computation of activity likelihood function values for multiple activity classifications across multiple past epochs before the present epoch. By pre-calculating and storing these likelihood values in advance, the system reduces the processing time required at the present epoch while maintaining accurate activity classification through temporal combining of pre-computed values.

Inventive Principle:

Principle #10Preliminary action

2Reliability

If the system processes sensor data to detect user activities, then activity classification capability is improved, but the system may overlook short-duration activities due to processing delays

Engineering Contradiction:

Improveactivity detection reliabilityVSAvoiddetection delay

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system continuously computes activity likelihood function values for multiple past epochs without interruption, maintaining an ongoing record of activity probabilities. This continuous computation ensures that even brief activities are captured in the temporal sequence of likelihood values, preventing oversight of short-duration activities while maintaining reliable detection.

Inventive Principle:

Principle #20Continuity of useful action

Solution Approach 2:

The system extends the analysis from a single time point to multiple temporal dimensions by computing likelihood values across several past epochs. This temporal dimensionality allows the system to detect activities that occur briefly within the epoch sequence, as the multi-epoch analysis captures transient states that single-point analysis would miss.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Measurement precision

If the system uses multiple sensor signals to classify activities, then measurement precision is improved, but device complexity increases

Engineering Contradiction:

Improveactivity classification accuracyVSAvoidprocessing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system merges the computation of activity likelihood function values across multiple past epochs and multiple activity classifications into a unified temporal combining process. By integrating these computations systematically, the system handles multiple sensor signals and classification categories without proportionally increasing complexity, as the merging process consolidates redundant operations and shared calculations.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS8930300B2Systems, methods, and apparatuses for classifying user activity using temporal combining in a mobile device

Publication Date: 2015.01.06 QUALCOMM INC

AI summary

Components, methods, and apparatuses are provided for determining activity likelihood function values for an activity classification for two or more past epochs based, at least in part, on signals from one or more sensors of a mobile device. A method may comprise, for each of a plurality of activity classifications, determining activity likelihood function values for each of the plurality of activity classifications for two or more past epochs. The activity likelihood function values may be based on signals from one or more sensors of a mobile device. The method may also include combining the activity likelihood function values to determine a likelihood function for an activity classification at a present epoch. The method may also include inferring a present activity of a user co-located with the mobile device to be one of the activity classifications based on the determined likelihood functions for the activity classifications at the present epoch.