Lane Detection Aided Online Dead Reckoning for GNSS Denied Environments

Area of Science:

• Robotics and Autonomous Systems
• Navigation and Positioning
• Computer Vision

Background:

• Autonomous vehicles (AV) increasingly rely on robust navigation systems.
• Global Navigation Satellite System (GNSS) is the primary navigation method but fails in signal-denied environments like urban canyons and tunnels.
• Existing methods struggle with positioning accuracy and error drift in GNSS-denied scenarios.

Purpose of the Study:

• To propose a novel method for vehicle dead reckoning in Global Navigation Satellite System (GNSS)-denied situations.
• To enhance the performance and reliability of autonomous vehicle navigation.
• To bound error drift effectively compared to standalone Inertial Navigation System (INS).

Main Methods:

• Developed a method for simultaneous vehicle dead reckoning using a learning-based lane detection model.
• Fused Inertial Navigation System (INS) data with the lane detection model for global position estimation.
• Employed the Unscented Kalman Filter (UKF) for integrating INS and the lane model, minimizing linearization errors and computation time.

Main Results:

• The proposed method demonstrated effective bounding of error drift compared to standalone Inertial Navigation System (INS).
• Real-vehicle experiments on highway driving validated the performance of the fused navigation system.
• Comparative analysis showed advantages over other dead-reckoning algorithms under similar system configurations.

Conclusions:

• The fusion of INS with a learning-based lane detection model offers a viable solution for autonomous vehicle navigation in GNSS-denied environments.
• The UKF integration provides an efficient and accurate method for dead reckoning.
• This approach significantly improves positioning reliability where satellite signals are challenged.