Confidence-Weighted Data Fusion for Object Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object discrimination techniques often result in overconfident classifications due to the use of additive or multiplicative methods, failing to accurately represent uncertainty in discrimination results, leading to binary-like probabilities that do not effectively utilize variations in data independence and confidence levels.
Innovation Solution
A system and method that incorporate a communication module, object identification module, confidence rules module, object confidence module, and data fusion module to generate a probability of identification by selecting confidence rules and fusing data based on alpha and beta parameters, which represent the degree of expected independence and confidence, allowing for flexible and accurate object classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If additive or multiplicative techniques are used to combine discrimination data, then the classification confidence is increased, but the uncertainty in discrimination results is not represented and overconfident behavior occurs
Solution Approach 1:
The patent transforms the probability combination approach from additive/multiplicative methods to a weighted combination method that incorporates confidence weights. This changes the parameters of the fusion process by introducing confidence-weighted probabilities that maintain both confidence representation and uncertainty quantification through the formula P(fused) = Σ(wi * Pi) where wi are confidence weights and Pi are class probabilities.
2Measurement precision
If confidence weights are introduced to represent uncertainty, then the accuracy of probability representation is improved, but the complexity of the fusion process increases
Solution Approach 1:
The patent applies preliminary action by calculating confidence weights before the probability fusion process. The confidence weight for each sensor or data source is determined in advance based on its reliability metrics, then these pre-computed weights are used in the fusion step. This separates the complexity of confidence assessment from the probability combination, making the overall process more manageable.
3Adaptability or versatility
If multiple confidence rules are applied to select appropriate fusion methods, then the adaptability to different discrimination scenarios is improved, but the computational overhead increases
Solution Approach 1:
The patent applies local quality by selecting different confidence rules and fusion methods based on the specific characteristics of each data source or discrimination scenario. Rather than using a single universal fusion method, the system adapts the fusion approach locally to match the reliability and characteristics of individual sensors or data sources, optimizing performance for each specific case.
Data Source
AI summary
The technology described herein includes a system and/or a method for identifying an object. The technology includes determining an alpha parameter that is associated with a fusion function. The technology includes determining a beta parameter that is associated with a degree of expected independence of a received set of data and the received set of data including information associated with a classification of the object. The technology includes fusing the received set of data based on the alpha parameter and the beta parameter. The technology includes generating a probability of identification of the classification of the object based on the fused data.