A Vision-Based Single-Sensor Approach for Identification and Localization of Unloading Hoppers

Area of Science:

• Robotics and Automation
• Computer Vision
• Artificial Intelligence

Background:

• Accurate identification and localization of bulk cargo unloading hoppers are critical for rail freight automation.
• Industry 4.0 and AI integration are transforming manual operations to intelligent control in bulk cargo unloading.

Purpose of the Study:

• To develop a vision-based 3D localization system for unloading hoppers.
• To address the technical challenges in accurate identification and localization for intelligent rail freight.

Main Methods:

• A single visual sensor architecture integrating object detection, corner extraction, and 3D localization.
• Utilizing a lightweight hybrid attention mechanism in YOLOv5 for enhanced hopper detection.
• Employing depth consistency constraint (DCC) and geometric constraints for sub-pixel corner extraction.
• Implementing a real-time 3D localization via corner-based initialization and RGB-D SLAM.

Main Results:

• The proposed system achieved an average localization accuracy of 97.07% in complex industrial scenarios.
• Demonstrated high precision and robustness under challenging working conditions.
• Successfully integrated object detection, corner extraction, and 3D localization modules.

Conclusions:

• The developed system meets the demands for automation, intelligence, and high precision in railway bulk cargo unloading.
• The system shows strong engineering practicality and significant application potential for intelligent rail freight.
• The approach enhances the efficiency and safety of bulk cargo handling processes.