Application of a YOLOv8-Based Monocular Ground-Plane Intersection Method to Pothole Detection

Area of Science:

• Computer Vision
• Robotics
• Automotive Engineering

Background:

• Road potholes pose significant risks to vehicle safety and infrastructure integrity.
• Accurate detection and distance estimation of potholes are crucial for autonomous driving systems and road maintenance.
• Existing methods often struggle with real-time performance and varying environmental conditions.

Purpose of the Study:

• To develop and validate a monocular vision-based system for pothole detection and distance estimation.
• To integrate camera calibration with object detection and geometric methods for precise pothole localization.
• To assess the feasibility of the proposed method in a controlled experimental environment.

Main Methods:

• Camera calibration using Zhang Zhengyou's method to determine intrinsic and extrinsic parameters.
• Pothole identification leveraging the YOLOv8 object detection model.
• Distance estimation employing the ground-plane intersection method.
• System validation on a smart-car platform with Raspberry Pi 5B in a sandbox environment.

Main Results:

• The integrated system successfully identified potholes within the controlled sandbox.
• The ground-plane intersection method provided preliminary distance estimations for detected potholes.
• Experimental results confirmed the initial feasibility of the proposed approach for short-range pothole detection and ranging.

Conclusions:

• The proposed method demonstrates preliminary effectiveness for pothole detection and distance estimation in controlled environments.
• Further research and validation are necessary under diverse real-world road conditions.
• This work contributes to the advancement of intelligent transportation systems and road safety technologies.