A Hardware-Software Integrated PCB Image Registration Method Based on Local Adaptive KNN and SIFT

Area of Science:

• Robotics and Automation
• Computer Vision
• Manufacturing Technology

Background:

• Solder-joint detection on complex printed circuit boards (PCBs) is challenging due to image variations like uneven features, non-uniform lighting, and geometric distortion.
• Conventional SIFT-based registration methods struggle with varying component densities due to fixed matching parameters.

Purpose of the Study:

• To develop a robust hardware-software integrated framework for PCB image registration.
• To improve the accuracy and adaptability of SIFT-based registration for complex PCBs.

Main Methods:

• Integration of a robotic end effector with a lifting mechanism and ring-light illumination for stable image acquisition.
• Implementation of a locally adaptive K-nearest neighbor (LAKNN) strategy combined with SIFT descriptors.
• Software module dynamically adjusts neighbor-search space based on local feature density and matching confidence.

Main Results:

• The proposed LAKNN method achieved 90.8% inlier-match accuracy in ablation experiments.
• Demonstrated improved registration robustness against variations in height, region, and viewpoint.
• The framework provides a practical solution for automated PCB inspection and robotic soldering alignment.

Conclusions:

• The proposed hardware-software integrated framework significantly enhances PCB image registration.
• The LAKNN strategy offers adaptive matching crucial for complex industrial applications.
• This approach lays a foundation for advanced automated PCB inspection and robotic soldering processes.