Fault Diagnosis of the Autonomous Driving Perception System Based on Information Fusion

Area of Science:

• Autonomous Driving Systems
• Sensor Fusion
• Fault Diagnosis

Background:

• Reliability of autonomous driving sensing systems is critical for safety.
• Perception system fault diagnosis is an under-researched area with limited solutions.

Purpose of the Study:

• To present an information-fusion-based fault diagnosis method for autonomous driving perception systems.
• To develop a method for diagnosing camera sensor defects using millimeter wave radar data.

Main Methods:

• Built an autonomous driving simulation scenario using PreScan software.
• Collected data from millimeter wave (MMW) radar and camera sensors.
• Used convolutional neural networks (CNN) for image identification and labeling.
• Fused MMW radar and camera sensor inputs in space and time.
• Mapped MMW radar points onto camera images to obtain regions of interest (ROI).

Main Results:

• The method effectively detects various camera sensor faults, including missing pixels, pixel shifts, and color loss.
• Fault detection deviations ranged from 0.26% to 99.84% depending on the fault type.
• Response times for fault alerts ranged from 0 s to 1.6 s, demonstrating real-time capabilities.

Conclusions:

• The developed fault diagnosis technology is effective for detecting sensor faults and issuing real-time alerts.
• This approach provides a foundation for simpler, more user-friendly autonomous driving systems.
• The information fusion principles established can support the development of more complex autonomous driving systems.