Degenerate Near-Planar 3D Reconstruction from Two Overlapped Images for Road Defects Detection.

Area of Science:

• Computer Vision
• Robotics
• Geospatial Analysis

Background:

• Road defects such as potholes possess 3D geometric characteristics.
• Accurate detection of road surface anomalies requires precise three-dimensional (3D) surface reconstruction.
• Existing methods may face challenges with degenerate road surface reconstruction.

Purpose of the Study:

• To present a novel technique for 3D road surface reconstruction using two overlapped images.
• To enhance the accuracy of road defect detection through improved 3D reconstruction.
• To address the degenerate road surface reconstruction problem by utilizing near-planar characteristics.

Main Methods:

• Reconstruction of 3D road surfaces from overlapped images captured by a downward-facing camera.
• Exploitation of near-planar road surface properties to solve reconstruction degeneracy.
• Parametric studies in simulation and accuracy analysis using real road images.

Main Results:

• Simulated studies confirmed minimal impact of camera noise, orientation, and vertical movement on reconstruction accuracy.
• Real-world image analysis demonstrated mean and standard deviation errors below 0.6 mm and 1 mm, respectively.
• On-road tests achieved over 94% precision, accuracy, and recall for road defect identification.

Conclusions:

• The proposed technique effectively reconstructs 3D road surfaces with enhanced accuracy.
• The method enables reliable detection of road surface defects, including potholes.
• The approach demonstrates high performance in real-world road conditions.