Passive Sensor Integration for Vehicle Self-Localization in Urban Traffic Environment

Area of Science:

• Robotics
• Geomatics Engineering
• Computer Vision

Background:

• Vehicular positioning in urban canyons is challenging due to Global Navigation Satellite System (GNSS) signal degradation (Non-Line-Of-Sight, multipath).
• Inertial sensors offer motion data but suffer from cumulative drift over time.
• Existing systems struggle to achieve lane-level accuracy in complex urban environments.

Purpose of the Study:

• To develop an accurate vehicular positioning system with lane-level precision in urban canyons.
• To mitigate GNSS signal issues using a novel positioning technique and sensor fusion.
• To improve the robustness and accuracy of vehicle localization through integrated sensor data.

Main Methods:

• Integration of Global Navigation Satellite System (GNSS) receivers, onboard cameras, and inertial sensors.
• Implementation of a novel GNSS positioning technique leveraging 3D building maps to reduce multipath and Non-Line-Of-Sight (NLOS) effects.
• Development of vision-based lane detection for inertial sensor drift correction and lateral error reduction.

Main Results:

• The proposed integrated system achieves sub-meter mean positioning error in challenging urban scenarios.
• Vision-based lane detection effectively corrects inertial sensor drift.
• Lane keeping and changing behaviors extracted from vision data further enhance lateral positioning accuracy.

Conclusions:

• The novel integrated vehicular positioning system demonstrates high accuracy and robustness in urban canyons.
• Sensor fusion, incorporating 3D maps and vision-based lane detection, significantly overcomes GNSS limitations.
• The system provides lane-level positioning performance, crucial for advanced driver-assistance systems and autonomous driving.