Perception System Resource Allocation via Expected Value of Information
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for inferring human-centric context from multiple sensors face challenges in efficiently fusing low-level streams of raw data into higher-level assessments of activity, often requiring substantial computational resources, which can lead to increased CPU usage and slow down primary applications.
Innovation Solution
The use of expected value of information (EVI) analysis in a greedy, one-step look-ahead approach to determine the next best set of observations, combined with rate-based and random selection techniques, to selectively utilize computational resources and reduce the computational burden while maintaining recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If comprehensive sensor data fusion is performed to infer human-centric context, then recognition accuracy is improved, but computational resource consumption increases
Solution Approach 1:
The system dynamically adjusts the level of data fusion and processing based on contextual needs. Instead of continuously fusing all sensor data at full complexity, the system adapts the fusion depth and computational intensity according to the current situation, maintaining accuracy when needed while reducing computational load during stable states.
Solution Approach 2:
The patent implements selective data fusion where only necessary sensor streams and processing levels are activated based on current contextual requirements. Rather than processing all available sensor data uniformly, the system applies partial processing to relevant data subsets, reducing overall computational burden while preserving recognition accuracy for critical functions.
2Speed
If real-time context awareness is implemented, then system responsiveness is improved, but CPU usage increases and primary applications slow down
Solution Approach 1:
The context awareness system is segmented into multiple independent processing layers with different computational intensities and time scales. Fast response layers handle critical real-time decisions with minimal CPU usage, while slower layers process less time-sensitive contextual information. This segmentation allows the system to maintain responsiveness for urgent matters without continuously consuming high CPU resources that would impact primary applications.
Solution Approach 2:
The system employs periodic context updates rather than continuous real-time processing. Context awareness is refreshed at strategically determined intervals based on system state and priority levels, allowing primary applications to execute without constant perceptual processing overhead while still maintaining adequate system responsiveness for time-critical operations.
Data Source
AI summary
The present invention leverages analysis methods, such as expected value of information techniques, rate-based techniques, and random selection technique, to provide a fusion of low-level streams of input data (e.g., raw data) from multiple sources to facilitate in inferring human-centric notions of context while reducing computational resource burdens. In one instance of the present invention, the method utilizes real-time computations of expected value of information in a greedy, one-step look ahead approach to compute a next best set of observations to make at each step, producing “EVI based-perception.” By utilizing dynamically determined input data, the present invention provides utility-directed information gathering to enable a significant reduction in system resources. Thus, of the possible input combinations, the EVI-based system can automatically determine which sources are required for real-time computation relating to a particular context.