Research Of Registration Algorithm In PCB AOI

PCB (Printed Circuit Board) is the important part of electronic equipment. At present, with the rapid development of MEMS (Micro_Electro_Mechanical Systems), electronic device is going to be miniaturization. It gradually demands the PCB towards the direction of high density board. So it needs efficient and high-precision detection methods. AOI (Automatic Optic Inspection) is a new and efficient technology for detecting online, PCB AOI uses AOI technology, it achieves high precision and high efficiency detection to the quality of PCB products.PCB AOI is a new type of detection technology. It based on machine vision, and other subjects, such as image processing, pattern recognition, artificial intelligence and so on. PCB AOI can be divided into two parts which are software and hardware. Image processing belongs to software and is the important part in the system. Image registration is one of the core algorithms of image processing. The precision of the algorithm and time complexity have the decisive effect in the accuracy and real-time of the system.This topic uses edge points as registration features for PCB conductive district, and put forward a method which based on global and local registration, nonlinear least square fitting method. Defined the edge gradient vector as registration similarity measure function accordingly. Firstly used global registration to the whole image, eliminated global translation error. Secondly corrected nonlinear geometric distortion and rotation error through the block local registration. And finally completed the registration through global edge fitting by least-square method which used to solve the fracture and dislocation of the edge between blocks.Using Visual C++6.0programming to achieve the algorithm, and it is embedded in the AOI software system to do the experiment. The results indicate that it corrects nonlinear distortion, like the nonlinear distortion of camera, warped PCB, and fix-errors of camera, etc. it takes about25seconds on average to complete registration of the5mega-pixel PCB image in the dual-core CPU of PC. The registration results showed that4accuracy could be achieved. It can satisfy the requirements of real-time and precision in production process.