Image Super-Resolution for Improved 6D Pose Estimation in Industrial Robotic Systems

Accurate six-dimensional (6D) pose estimation is critical for improving the precision and efficiency of robotic systems in industrial automation. However, existing methods struggle in complex environments with custom objects, variable lighting, and occlusions. This thesis explores the use of super-resolution techniques to enhance image quality and improve pose estimation accuracy. The research integrates GDR-Net, a state-of-the-art 6D pose estimation network, with DRLN, an advanced super-resolution model. It hypothesizes that higher-resolution images can significantly boost accuracy. A synthetic dataset simulating real industrial conditions was developed to test this approach, demonstrating improved accuracy for small and medium size objects. The findings highlight the potential of super-resolution to enhance 6D pose estimation robustness and pave the way for more resilient robotic vision systems in challenging industrial settings.