FAST Pre-Filtering-Based Real Time Road Sign Detection for Low-Cost Vehicle Localization

Area of Science:

• Computer Vision
• Robotics
• Automotive Engineering

Background:

• Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) limitations hinder precise vehicle localization.
• Low-cost sensors are crucial for mass-produced vehicles but often lack accuracy.
• Existing perception-based localization systems require accurate landmark detection.

Purpose of the Study:

• To develop a precise and affordable vehicle localization system.
• To propose a robust road sign detection method using a monocular camera for landmark identification.
• To fuse data from low-cost GNSS/INS and perception sensors for enhanced localization.

Main Methods:

• A part-based approach for road sign detection using a monocular camera.
• Detecting road sign corners and combining them to identify the sign.
• Fusing estimated vehicle positions from low-cost GNSS/INS with landmark data.

Main Results:

• The proposed method achieves high recall (97.48%) and precision (98.78%) in road sign detection.
• Processing time is drastically reduced to 66.7 ms, compared to 4802 ms for a state-of-the-art detector.
• Detection performance is comparable to state-of-the-art methods.

Conclusions:

• The developed road sign detection method is effective and efficient for vehicle localization.
• The part-based approach offers a viable solution for detecting varied road signs.
• This system provides a cost-effective and precise localization solution for mass-produced vehicles.