A VibV Dataset Integrating Vibration and Vision for Enhanced Safety in Self-Driving Tasks

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Autonomous driving systems rely on sensors like cameras, LiDAR, and radar for environmental perception.
• Current perception systems face limitations in challenging conditions (e.g., extreme weather, special road surfaces), leading to detection failures and safety risks.
• Blind spots and uncontrolled threats remain significant safety challenges in real-world traffic scenarios.

Purpose of the Study:

• To introduce the VibV dataset, integrating vehicle vibration signals into autonomous driving perception systems.
• To enhance the accuracy and robustness of perception systems by using vibration data as supervisory signals.
• To improve the overall safety and reliability of autonomous driving technology.

Main Methods:

• Simultaneously recorded vehicle vibration signals and vision data in diverse road conditions (rumble strips, speed bumps).
• Collected 39 segments of vibration data and 22,677 video frames across 39 experiments over two months.
• Processed vibration signals and manually annotated/classified image data for system training and evaluation.

Main Results:

• Demonstrated the usability and reliability of the VibV dataset through technical evaluations.
• Showcased the potential of vibration signals to augment existing sensor data for improved environmental detection.
• Validated the enhancement of perception accuracy when vibration information is utilized as supervisory signals.

Conclusions:

• The VibV dataset provides a novel approach to enhance autonomous driving perception by incorporating vehicle dynamics.
• Utilizing vibration signals offers a promising method to overcome limitations of traditional sensors in specific scenarios.
• The dataset supports the development of more robust and safer autonomous driving systems for complex real-world applications.