Vision-Inertial Navigation Map Search Space Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Vision-Inertial Navigation Systems (VINS) face significant navigation errors and prolonged recovery times when passing over cloud cover or featureless terrain, as they require extensive map searches upon re-emergence, consuming substantial computing resources and time, especially in airborne applications.
Innovation Solution
Implementing a Deep Learning algorithm that classifies aerial imagery into probabilities of different land features, allowing the system to pre-divide the map into subsections and focus searches on areas with similar classification probabilities, reducing the map search area and processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the VINS performs a search over a large map area to re-localize after an outage period, then the navigation accuracy is restored, but the computing resources and processing time required become excessively large
Solution Approach 1:
The patent divides the large map area into multiple smaller sub-areas based on different land feature types (e.g., water bodies, urban areas, forests). Instead of searching the entire map area uniformly, the system segments the search space according to the classified land features, reducing the computational burden while maintaining navigation accuracy.
Solution Approach 2:
The system applies different search strategies and computational resources to different map sub-areas based on their land feature classifications. For example, water bodies with no features require different handling compared to urban areas with dense landmarks. This local differentiation optimizes the balance between search thoroughness and computational efficiency.
2Measurement precision
If the VINS performs a search over a large map area to re-localize after an outage period, then the navigation accuracy is restored, but the computing resources required become excessively large
Solution Approach 1:
The patent segments the map search space into multiple sub-areas based on land feature classifications. By dividing the large search area into smaller, feature-based segments, the system reduces the total computational resources required while maintaining the ability to achieve accurate re-localization.
Solution Approach 2:
Instead of performing a complete exhaustive search over the entire map area, the system performs partial searches focused on specific sub-areas identified through land feature classification. This partial action approach reduces computing resource consumption while still achieving the necessary navigation accuracy.
3Measurement precision
If the VINS searches a large map area to account for free-inertial error growth during an outage period, then the re-localization accuracy is improved, but the processing time becomes impractically long
Solution Approach 1:
The system performs land feature classification and map sub-area segmentation before the actual re-localization search. This preliminary action prepares the search space in advance, allowing the subsequent re-localization process to focus only on relevant sub-areas, thereby improving both accuracy and recovery speed.
Solution Approach 2:
By segmenting the map into feature-based sub-areas, the system enables parallel processing and focused searches. This segmentation allows the VINS to quickly identify and search only the relevant sub-areas where the vehicle is likely to be located, improving recovery speed without sacrificing re-localization accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system and methods for reducing map search area requirements in a navigation system are disclosed. The system includes a vehicle, an imaging device onboard the vehicle configured to generate an image scan, receive at least one image responsive to the image scan, and a processing device configured to receive and store the at least one image. The system further includes a navigation system onboard the vehicle configured to store an image of a map, and a learning network associated with the navigation system and configured to divide the image of the map into a plurality of map subsections, recognize each one of a plurality of images of different landmarks on the map, generate a set of classifications for each map subsection, and associate each classification of the set of classifications with at least one landmark of the different landmarks on the map.