Multi-Sensor Fusion Using Transform Learning for Lower Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional multi-sensor fusion techniques are computationally complex and do not perform well in all scenarios due to challenges related to data imperfection and diversity of sensing mechanisms, often resulting in less effective data integration.
Innovation Solution
A processor-implemented method involving joint optimization of sensor-specific transforms, fusing transforms, and weight matrices through a two-stage approach, where individual sensor representations are first learned using dedicated transforms and coefficients, and then fused using a common transform and its associated coefficients to capture correlations between different sensor representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional multi-sensor fusion techniques are used, then data integration is achieved, but computational complexity increases
Solution Approach 1:
The patent segments the multi-sensor fusion process into two distinct stages: (1) individual sensor representation learning using dedicated transforms and coefficients, and (2) fusion of these representations using a common transform. This segmentation reduces computational complexity by breaking down the complex simultaneous optimization into manageable sequential steps, while maintaining effective data integration through the structured two-stage approach.
2Reliability
If conventional multi-sensor fusion techniques are used, then data integration is achieved, but performance varies across different scenarios
Solution Approach 1:
The patent employs dynamic adaptability by allowing the system to learn sensor-specific transforms and coefficients tailored to each sensor type and scenario. The iterative optimization process adapts the fusion weights and transforms based on the specific characteristics of the input data, enabling the system to perform effectively across diverse scenarios including different sensor configurations, data qualities, and application domains.
3Reliability
If sensor-specific transforms and coefficients are learned for each sensor, then data authenticity is enhanced, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by first learning sensor-specific representations using dedicated transforms and coefficients before performing the fusion operation. This preliminary processing enhances data authenticity by capturing the unique characteristics of each sensor type, while the pre-computed representations reduce the complexity of the subsequent fusion step, as the system only needs to fuse already-processed features rather than raw sensor data.
Data Source
AI summary
This disclosure relates to multi-sensor fusion using Transform Learning (TL) that provides a compact representation of data in many scenarios as compared to Dictionary Learning (DL) and Deep network models that may be computationally intensive and complex. A two-stage approach for better modeling of sensor data is provided, wherein in the first stage, representation of the individual sensor time series is learnt using dedicated transforms and their associated coefficients and in the second stage, all the representations are fused together using a fusing (common) transform and its associated coefficients to effectively capture correlation between the different sensor representations for deriving an inference. The method and system of the present disclosure can find application in areas employing multiple sensors that are mostly heterogeneous in nature.