Mobile Activity Detection Using Tiered Linear Discriminant Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing mobile device technologies lack an efficient method to accurately determine user activities based on sensor data, often relying on raw data analysis that is computationally demanding and inefficient, especially when distinguishing between similar activities.
Innovation Solution
The implementation of a tiered activity classification system using linear discriminant analysis (LDA) to transform raw sensor data into abstract features, which are then used to identify presumed activities by calculating posterior probabilities based on Gaussian likelihood models and prior contextual data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If raw sensor data is analyzed directly to determine user activities, then measurement precision is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the activity detection process into multiple tiers: first tier analyzes a subset of activities using selected features, while second tier handles remaining activities. This segmentation reduces computational complexity by avoiding full analysis of all activities simultaneously, while maintaining measurement precision through hierarchical refinement.
Solution Approach 2:
The patent extracts and analyzes only the most relevant features for each activity tier rather than processing all raw sensor data uniformly. Linear discriminant analysis identifies and extracts discriminative features that maximize separation between activity classes, reducing computational burden while preserving detection accuracy.
2Measurement precision
If all sensor features are analyzed to distinguish between similar activities, then measurement precision is improved, but processing time increases
Solution Approach 1:
The patent divides activities into tiers based on difficulty of distinction. First tier handles easily distinguishable activities using fewer features, while second tier handles similar activities requiring more detailed analysis. This segmentation reduces average processing time while maintaining precision for difficult cases.
Solution Approach 2:
The patent applies partial analysis by first evaluating a subset of features and activities. Only when confidence is insufficient does it proceed to more comprehensive analysis. This partial action approach reduces average processing time while maintaining measurement precision when needed.
3Measurement precision
If comprehensive feature analysis is performed for all activities, then activity detection accuracy is improved, but energy consumption increases
Solution Approach 1:
The patent segments computational resources and features across activity tiers. First tier uses minimal processing energy for common, easily distinguishable activities. Second tier consumes more energy only when necessary for ambiguous cases. This segmentation optimizes the trade-off between detection accuracy and energy consumption.
Solution Approach 2:
The patent dynamically changes processing parameters including which features to analyze and which activities to consider, based on current sensor data characteristics. Linear discriminant analysis adapts feature selection to maximize separation for current activity patterns, reducing unnecessary computational energy while maintaining detection accuracy.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining a presumed activity associated with a mobile device. A plurality of sensor values detected by one or more sensors onboard the mobile device is received over a period of time. A plurality of derived values is calculated from the plurality of sensor values. The derived values are selectively combined to generate one or more abstract values. A presumed activity is identified from a plurality of possible activities based on a level of similarity between the one or more abstract values and expected values of each of the plurality of possible activities that correspond to the one or more abstract values.