Fast Online Coordinate Correction of a Multi-Sensor for Object Identification in Autonomous Vehicles

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Multi-sensor perception systems can suffer from coordinate mismatches between sensors, even after conversion to a common frame.
• These discrepancies, caused by sensor noise or mounting issues, degrade distant object position estimation and identification accuracy.
• Existing correction methods like off-line modeling and real-time estimation have limitations in real-time application or computational complexity.

Purpose of the Study:

• To develop a computationally efficient online method for correcting coordinate mismatches in multi-sensor systems.
• To improve the accuracy of object identification in autonomous driving applications by mitigating sensor coordinate errors.

Main Methods:

• A reduced sensor position error model focusing on dominant parameters was developed.
• Parameters were estimated using rapid mathematical operations for real-time processing.
• The proposed fast online coordinate correction method was applied to multi-sensor data.

Main Results:

• The method significantly reduced computational effort by up to 99.7% compared to previous studies.
• Object identification accuracy, particularly for radar data, improved by 94.8%.
• The correction method operates within acceptable estimation error tolerances.

Conclusions:

• The proposed fast online coordinate correction method effectively addresses multi-sensor coordinate mismatches.
• It offers a computationally efficient solution suitable for real-time autonomous driving applications.
• The method demonstrates substantial improvements in both computational performance and object identification accuracy.