A Novel Framework for Image Matching and Stitching for Moving Car Inspection under Illumination Challenges.

Area of Science:

• Computer Vision
• Robotics
• Automotive Engineering

Background:

• Vehicle exterior inspection is crucial for safety but faces challenges with moving objects and reflections.
• Current deep learning methods struggle with accurate defect localization due to image orientation issues.
• Manual inspections are labor-intensive, costly, and error-prone.

Purpose of the Study:

• To develop an efficient end-to-end pipeline for image matching and stitching of moving vehicles.
• To address challenges in feature extraction and correspondence for highly reflective moving objects.
• To improve the accuracy and reduce false positives in automated vehicle defect detection.

Main Methods:

• An innovative pipeline for efficient image matching and stitching was developed.
• A novel filtering scheme was introduced to exclude background points during feature extraction.
• Sequential stereo-rectified pairs were used to generate a high-quality image mosaic.

Main Results:

• The proposed method effectively handles feature extraction from moving objects with strong reflections.
• Accurate localization of vehicle damages within a 3D reference system was achieved.
• A high-quality image mosaic was generated, improving defect identification.

Conclusions:

• The developed pipeline offers a robust solution for automated vehicle exterior inspection.
• The novel filtering scheme significantly enhances the accuracy of feature matching for reflective surfaces.
• This approach promises to reduce labor costs and increase the reliability of vehicle defect detection systems.