Unet-based Image Restoration for Enhancing Object Detection in Projector-Camera Systems

Key Points

• The study aims to resolve color distortion in projector-camera systems to enhance object recognition accuracy.
• Developed a U-Net based model to restore distorted colors from projector images.
• Input images with color distortion to evaluate similarity and object recognition accuracy.
• Tested the model's performance under varying lighting without requiring re-training.
• The proposed model significantly restored colors closer to the original images.
• Improved accuracy in object recognition was validated through experiments.
• Demonstrated independence from the object detection system, adaptable to lighting changes.

Abstract