Spatial-Semantic Object Relation Graph Networks for Vehicle Attachment Detection in Automatic Car Wash System.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Accurate object detection is crucial for preventing damage to vehicle attachments during automated car washes.
• Existing methods struggle with diverse shapes and visual ambiguity, leading to low accuracy and false detections.

Purpose of the Study:

• To develop a novel framework for precise object detection of vehicle attachments in automatic car wash environments.
• To improve the accuracy and reliability of object detection, thereby reducing damage to vehicle parts.

Main Methods:

• Integration of a YOLOv11-based object detector with a graph neural network (GNN).
• Introduction of a spatial graph module to refine object localization using spatial constraints.
• Incorporation of a class graph module to model inter-class semantic correlations for improved classification.

Main Results:

• The proposed method achieved a mean Average Precision at 50% IoU (mAP50) of 97.9% on a real-world dataset.
• Outperformed state-of-the-art models like D-FINE (96.5%) and RT-DETR (96.1%).
• Demonstrated robustness across varying viewpoints and background conditions.

Conclusions:

• The novel framework significantly enhances the precision and reliability of object detection for vehicle attachments.
• This advancement contributes to improved safety and reduced damage in automated car wash systems.
• The integration of GNNs offers a promising direction for addressing challenges in complex visual detection tasks.