Robust Pose Estimation and Size Classification for Unknown Dump Truck Using Normal Distribution Transform

Area of Science:

• Construction Automation
• Robotics
• Computer Vision

Background:

• The Japanese construction industry faces labor shortages, increasing demand for automation.
• Automating soil loading onto dump trucks requires accurate dump truck pose and size estimation.
• Existing methods struggle with unknown dump truck specifications.

Purpose of the Study:

• To develop a robust method for estimating the pose and classifying the size of dump trucks with unknown specifications.
• To enable automated soil loading operations in construction.

Main Methods:

• A two-stage approach combining pose estimation and size classification.
• Utilizing Normal Distribution Transform (NDT) with specialized dump truck templates for pose estimation.
• NDT templates designed to differentiate size categories while accommodating intra-category variations.

Main Results:

• The method accurately estimates dump truck pose across various positions and orientations in real-world environments.
• Successfully classifies dump trucks into three predefined size categories.
• Achieves a computation time of approximately 0.13 seconds, suitable for practical applications.

Conclusions:

• The proposed method effectively addresses the challenge of dump truck pose and size estimation for unknown specifications.
• It significantly contributes to the automation of soil loading processes in construction.
• The approach shows promise for enhancing efficiency and addressing labor shortages in the construction sector.