CNN Vector Map Navigation in GPS-Denied Vehicle Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional imaging navigation and positioning systems for autonomous vehicles are complex, costly, and resource-intensive, and rely on GNSS signals, which can be unreliable in resource-constrained environments and affected by atmospheric conditions, leading to processing challenges and reduced accuracy.
Innovation Solution
Utilizing convolutional neural networks (CNNs) for image-based navigation and vector map recognition, combined with edge detection filters, to create and compare vectored maps without GPS/GNSS signals, and employing FPGA control blocks to manage processing and resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multispectral imaging devices are used to create layering and vectorization maps, then mapping precision is improved, but device complexity increases
Solution Approach 1:
The patent uses conventional EOI video systems to capture optical images that serve as copies or representations of the terrain, replacing the need for complex multispectral imaging devices. The CNN then processes these optical images to extract terrain information, achieving mapping precision without requiring specialized sensors.
Solution Approach 2:
The patent replaces complex optical-mechanical multispectral imaging systems with a computational approach using conventional cameras combined with CNN-based image processing. The mechanical complexity of multispectral devices is substituted with algorithmic processing of standard optical images.
2Measurement precision
If multispectral imaging devices are used to capture data across different wavelengths, then material identification accuracy is improved, but energy consumption increases
Solution Approach 1:
The patent captures terrain information using conventional optical images instead of energy-intensive multispectral data acquisition. The CNN processes these optical copies to identify materials and terrain features, reducing energy consumption while maintaining identification accuracy.
Solution Approach 2:
The patent replaces energy-intensive multispectral sensing with computational processing of standard optical images. The CNN algorithm substitutes for the physical multispectral sensors, achieving material identification with lower energy requirements.
3Measurement precision
If GNSS signal synchronization is used for navigation and positioning, then positioning accuracy is improved, but reliability deteriorates in resource-constrained environments
Solution Approach 1:
The patent introduces CNN-based terrain recognition as an intermediary system that works in conjunction with or替代 GNSS. By comparing processed optical images with reference terrain data, the system determines position without relying solely on satellite signals, improving reliability in environments where GNSS is unavailable or unreliable.
4Ease of manufacture
If conventional EOI video systems are used for imaging, then system cost is reduced, but navigation and positioning processing becomes extremely difficult
Solution Approach 1:
The patent replaces complex mechanical and algorithmic processing systems with a CNN-based computational approach. The CNN automatically extracts terrain features and performs navigation processing from standard optical images, reducing processing difficulty while maintaining low system cost.
Data Source
AI summary
An onboard navigation system for vehicles in GPS-denied environments using image-based mapping. A mapping vectorization and navigation system uses a convolutional neural network (CNN) to improve the speed of recognition, orientation, and navigation while avoiding the use of GPS/GNSS signals.