Virtual Reference Images for Terrain Navigation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing camera-based navigation methods for autonomous vehicles in extreme environments, such as planetary surfaces, face challenges in accurately determining position and orientation due to lighting variations and the lack of illumination-independent feature extraction techniques, leading to reduced accuracy and robustness.
Innovation Solution
A method that generates virtual reference images using computer graphics to create a feature database, allowing for the identification of features in real-time images, even under varying lighting conditions, by pre-calculating expected image sequences based on topographic maps and planned flight trajectories, enabling the use of robust feature extraction techniques during navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional feature extraction methods are used on real camera images, then navigation can be performed in real-time, but accuracy deteriorates under varying lighting conditions
Solution Approach 1:
The patent applies preliminary action by pre-generating virtual reference images from topographic maps and extracting features from these controlled virtual images before actual navigation. This allows the system to prepare a stable feature database under known lighting conditions, which then serves as a reliable reference during real-time navigation regardless of actual lighting variations.
Solution Approach 2:
The patent uses copying by creating virtual reference images that replicate the expected visual appearance of the terrain based on topographic maps. These virtual images serve as proxies for actual terrain images, allowing feature extraction to be performed on idealized copies rather than on noisy real images affected by lighting variations.
2Adaptability or versatility
If illumination-independent feature extraction is attempted, then robustness to lighting changes improves, but feature extraction accuracy deteriorates
Solution Approach 1:
The patent introduces virtual reference images as an intermediary between the topographic map data and the real camera images. Features are extracted from the virtual images with precise localization, and these extracted features then serve as the basis for matching against real images. This intermediary allows precise feature extraction in the virtual domain while maintaining adaptability to lighting changes through the matching process.
3Ease of operation
If crater-based navigation methods are used, then navigation capability is achieved on planetary surfaces, but precision deteriorates compared to pinpoint features
Solution Approach 1:
The patent applies parameter changes by transitioning from using large-scale geological features (craters) as navigation references to using pinpoint features with precisely defined coordinates. This change in the spatial parameter of reference features dramatically improves positioning precision while maintaining navigation capability through the use of virtual reference images that can incorporate any identifiable terrain feature.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
One method describes the architecture of a vision-based navigation system in terrain with available topographic maps, based on existing methods for determining prominent terrain features. An offline process used in this method creates a feature database from existing topographic maps, which is then used in an online process to recognize sighted features. Virtual reference images of the expected situation are generated using computer graphics methods. These images are analyzed using feature extraction methods, and a feature catalog consisting of vision-based feature vectors and the corresponding 3D coordinates of the located features is derived from this analysis. This data is stored in a model database, for example, in a flight system, enabling navigation near a reference trajectory and under planned lighting conditions.The ability to use any feature extractor, not necessarily one that is invariant to illumination, enables high localization properties and correspondingly high navigation accuracy.