UAV Semantic Localization Using Reference Tiles Without GNSS
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Unmanned aerial vehicles (UAVs) face challenges in determining their location when Global Navigation Satellite System (GNSS) signals are unavailable, rendering traditional location determination methods ineffective.
Innovation Solution
UAVs employ a camera-based, real-time semantic localization algorithm that uses machine learning to generate a feature mask from captured images, correlating semantically labeled pixels with geolocated reference tiles to determine geographic location.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If GNSS-based location determination is used, then location accuracy is improved, but the system fails in GNSS-denied environments
Solution Approach 1:
The patent introduces an intermediary system (semantic localization algorithm using machine learning models) that bridges the gap between visual imagery and geographic location. The system processes captured images through trained models to generate feature masks, which are then correlated with reference tiles to determine UAV location, serving as a mediator when direct GNSS signals are unavailable
Solution Approach 2:
The patent creates a visual copy of the geographic environment through reference tiles that contain pre-processed semantic information. By comparing the feature mask (a processed copy of the current image) against these reference tile copies, the system can locate the UAV's position without relying on GNSS signals
2Device complexity
If traditional location determination methods are used, then the system is simple, but location determination becomes impossible without GNSS signals
Solution Approach 1:
The patent makes the location determination system universal by enabling it to function in multiple environments - both GNSS-available and GNSS-denied conditions. The machine learning-based semantic localization provides a universal solution that works across different operational contexts, replacing the need for environment-specific location methods
3Measurement precision
If machine learning-based semantic localization is implemented, then location accuracy in GNSS-denied environments is improved, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models offline and generating reference tiles with pre-computed semantic information before actual UAV operation. This shifts the computational burden to the training phase, allowing the deployed system to perform lighter real-time inference operations while maintaining high accuracy
Solution Approach 2:
The system creates simplified copies of complex geographic features through semantic segmentation. Instead of processing raw pixel data in real-time, the system uses pre-generated feature masks and reference tiles that contain extracted semantic information, reducing the computational complexity of real-time location determination
Data Source
AI summary
A computer-implemented method comprises receiving an image captured by a camera on an unmanned aerial vehicle (UAV). The image depicts an environment below the UAV. A feature mask associated with the image is generated via a machine learning model that is trained to identify and semantically label pixels representing the environment depicted in the image. One or more reference tiles associated with the environment are retrieved. The reference tiles are associated with particular geographic locations and specify semantically labeled pixels representing the geographic locations. The semantically labeled pixels of the feature mask are correlated with the semantically labeled pixels of at least one of the one or more reference tiles to determine the geographic location of the UAV in the environment.