Optical Flow Uncertainty Mapping for UAV Pose and Attitude Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques fail to utilize uncertainty maps to determine the reliability of optical flow output and lack the ability to enhance the efficiency and accuracy of pose, velocity, and attitude estimation for unmanned aerial vehicles (UAVs) and urban air mobility (UAMs), which are crucial for autonomous flight and safety.
Innovation Solution
A method and system that employs a trained machine-learning model to generate an uncertainty map alongside an optical flow map, allowing for the selection of a best set of optical flow estimates by applying an uncertainty map to a feature map, followed by transformations and optimizations to estimate pose, velocity, and attitude values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional techniques are used to estimate pose, velocity, and attitude without uncertainty maps, then the processing is simpler and faster, but the reliability and precision of the estimation are reduced due to inability to filter outliers
Solution Approach 1:
The patent segments the optical flow estimation process by separating reliable estimates from unreliable ones using uncertainty maps. The machine learning model divides the feature map into high-confidence and low-confidence regions, allowing selective processing that improves precision without requiring complete reprocessing of all data.
Solution Approach 2:
The uncertainty map serves as an intermediary element between the optical flow estimation and the final pose/velocity/attitude calculation. This intermediary provides reliability information that mediates the selection of optimal feature sets, enhancing measurement precision while maintaining manageable system complexity through structured intermediate representation.
2Reliability
If uncertainty maps are generated and applied to select best optical flow estimates, then the reliability and precision of estimation improve, but the processing time and computational cost increase
Solution Approach 1:
The uncertainty map is generated in advance as a preliminary step before the actual pose, velocity, and attitude estimation. By pre-computing reliability information and identifying high-confidence feature regions beforehand, the system avoids time-consuming trial-and-error processing during the main estimation phase, thus reducing overall processing time while maintaining high reliability.
Solution Approach 2:
The patent applies partial action by focusing computational resources only on high-confidence regions identified by the uncertainty map. Instead of processing the entire feature map with full computational effort, the system selectively processes only the most reliable portions, achieving high reliability with reduced processing time by avoiding excessive computation on low-confidence data.
3Quantity of substance
If all optical flow estimates are processed without selection, then more data is available for estimation, but outliers reduce the overall precision and reliability of the results
Solution Approach 1:
The patent applies local quality by treating different regions of the feature map differently based on their uncertainty characteristics. High-confidence regions are selected for processing while low-confidence regions are excluded, creating a quality-filtered dataset. This local differentiation ensures that the quantity of processed data consists only of high-quality estimates, thereby maintaining measurement precision.
Solution Approach 2:
The system changes the parameter of data selection by using uncertainty values as a filtering criterion. By transforming the raw optical flow data into uncertainty-weighted selections, the system dynamically adjusts which data points are included based on their reliability parameters, ensuring that the quantity of processed data reflects optimal precision without including detrimental outliers.
Data Source
AI summary
Disclosed are methods and systems for utilizing an uncertainty map to select a best set of optical flow estimates for one or more vehicles. For instance, a method may include receiving one or more images captured by one or more cameras of at least one vehicle, analyzing the one or more images by inputting the one or more images into a trained machine-learning model, wherein the trained machine-learning model is configured to process the one or more images to determine an optical flow map and an uncertainty map, determining a feature map based on the optical flow map, determining an improved feature map by applying the uncertainty map to the feature map, transforming the improved feature map by applying at least one transformation or optimization to the improved feature map, and based on the transformed improved feature map, estimating a pose value, a velocity value, or an attitude value.