Automatic Real-Time Pose Estimation of Machinery from Images.

Area of Science:

• Robotics and Automation
• Computer Vision
• Geomatics Engineering

Background:

• Traditional machine positioning relies on geodetic sensors like Global Navigation Satellite Systems (GNSS) and robotic total stations.
• High-frequency localization is crucial for advanced automation tasks.
• Existing methods may have limitations in certain environments or require specific infrastructure.

Purpose of the Study:

• To develop and evaluate a stereo camera system for high-frequency, accurate machine localization.
• To provide an image-based alternative to conventional geodetic sensors.
• To assess the achievable accuracies and robustness of the developed system.

Main Methods:

• Development of algorithms for detecting active markers in stereo image pairs.
• Implementation of stereo point correspondence finding and machine pose estimation.
• Utilizing Monte Carlo simulations for theoretical accuracy estimation and system design.
• Field measurements and comparison with laser tracker reference measurements for validation.

Main Results:

• The stereo camera system achieved translation accuracies better than 16 mm and rotation accuracies better than 3 mrad.
• Overall 3D point accuracies for machine localization exceeded 16 mm.
• The prototype demonstrated robustness in field measurements.

Conclusions:

• A novel, powerful image-based machine localization system using stereo cameras has been successfully prototyped.
• This system presents a viable alternative to classical geodetic sensors for machine positioning in automation.
• The developed technology offers high-frequency and accurate localization capabilities.