Vision-Based Vehicle Localization for GNSS-Obstructed Navigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing navigation systems struggle to accurately determine a vehicle's position, direction of travel, and orientation in real time, especially in conditions where GNSS signals are obstructed or weak, leading to inaccuracies that hinder Augmented Reality capabilities.
Innovation Solution
A computer-implemented method using vehicle-mounted cameras and artificial neural networks to analyze images and compute angular changes, combined with GPS and IMU data, to determine vehicle position and heading vector in real time, with confidence adjustments based on data consistency, and training to predict navigation key points like road signs and turn points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If GNSS trilateration is used to determine vehicle location, then accurate position determination is achieved under open sky conditions, but the system fails in tunnels or urban canyons where satellite signals are obstructed
Solution Approach 1:
The patent combines multiple positioning methods (GNSS, dead reckoning, and visual odometry from camera images) into a unified navigation system. The system dynamically switches between and integrates these different positioning approaches to maintain accurate vehicle position determination whether the vehicle is in open sky conditions or signal-obstructed environments like tunnels and urban canyons.
Solution Approach 2:
The system changes the active positioning parameters based on environmental conditions. When GNSS signals are available, it uses satellite-based trilateration; when signals are obstructed, it transitions to using camera-based visual odometry and dead reckoning parameters, thereby adapting to different operational contexts and maintaining positioning accuracy.
2Duration of action of stationary object
If dead reckoning processes are used to estimate vehicle position during signal outages, then position estimation continues without satellite signals, but the estimates become noisy and experience drift over time
Solution Approach 1:
The system uses camera-based visual odometry to provide continuous feedback on vehicle position and orientation. By capturing images at different time intervals and computing angular changes between them, the system generates corrective feedback that compensates for the drift and noise accumulation inherent in dead reckoning processes, thereby maintaining measurement precision over extended periods without satellite signals.
Solution Approach 2:
The patent replaces the purely inertial mechanical sensing approach of dead reckoning with an optical-based visual odometry system using cameras. This substitution uses image analysis and angular change computation to determine vehicle position and orientation, providing a non-mechanical alternative that avoids the drift and noise problems of traditional inertial methods.
3Speed
If IMU is used to measure vehicle orientation and direction of travel, then real-time orientation data is obtained, but the measurements are noisy and drift over time
Solution Approach 1:
The system uses camera-based visual odometry to provide continuous feedback on vehicle orientation and heading. By computing angular changes between sequential camera images, the system generates corrective feedback that compensates for the noise and drift in IMU measurements, thereby maintaining orientation measurement accuracy while preserving the high refresh rate capability of the IMU.
Solution Approach 2:
The patent creates a composite sensing system that combines IMU data with visual odometry data from cameras. This composite approach integrates the high-speed orientation measurements of the IMU with the drift-corrected orientation information from image analysis, producing a hybrid measurement system that achieves both high refresh rates and sustained accuracy.
4Quantity of substance
If GNSS update rates are used for position determination, then position data is obtained, but the update rate of 0.1 to 1.0 Hz results in slight delays in determining vehicle position
Solution Approach 1:
The patent replaces the satellite-based GNSS positioning system with a camera-based visual odometry system for real-time position determination. By using images captured at high frame rates and computing angular changes between them, the system achieves position updates at the camera's frame rate (typically 30-60 Hz or higher), eliminating the 0.1 to 1.0 Hz update rate limitation and time delays of GNSS.
Solution Approach 2:
The system uses periodic camera image capture at high frame rates to continuously update vehicle position information. This periodic visual sampling occurs much more frequently than GNSS updates, providing real-time position data without the delays inherent in satellite-based systems.
Data Source
AI summary
A computer-implemented method and apparatus for accurately determining a vehicle's position and predicting navigation key points in real time. The method including instructing a vehicle mounted camera to capture a plurality of images, computing the angular changes between a first image and a subsequent second image, determining the vehicle's position and heading vector from the computed angular changes, collecting at least one image of the plurality of images as a first training subset, obtaining image-related coordinates of navigation key points as a second training subset related to the at least one image of the first training data subset, supplying the first and second training data subset to an artificial neural network as a training dataset, and training the artificial neural network on the training dataset to predict image-related coordinates of navigation key points indicative of road sign locations and/or turn points.