Research Of Subpixel Visual Measurement Technology

For most industrial parts,it is necessary to meet the size requirements in the production process.The traditional size measurement mainly relies on manual inspection.Subject to subjective judgment of the human eye and the influence of measurement tools,manual method is not only inefficient,but also has low detection accuracy.In the size inspection process,machine vision-based measurement systems are more and more popular,which have many advantages over manual measurement,such as high precision,high efficiency,automation,non-contact.On the other hand,the sub-pixel edge detection algorithm is superior to the traditional pixel-level algorithm,for improving the edge positioning accuracy without increasing the resolution of the camera,thereby saving hardware cost.In this paper,the sub-pixel visual measurement technology is studied and implemented,including sub-pixel edge detection algorithm and size measurement method.For the sub-pixel edge detection algorithm,this paper improves the traditional Zernike moment algorithm,including changing the step edge model to the gradient one,adopting an adaptive fitting method to obtain the edge width parameter,and introducing diffusion filter to denoise and preserve the edge.The improved algorithm,achieving 0.03 pixels position accuracy in the simulated image,is better than the traditional one on the anti-noise performance and position accuracy.This paper proposes a cylinder size measurement method,which can map the image edge points to object points and fit these points to obtain the size.Considering the factors such as camera distortion and object thickness,this method has higher position accuracy and is more stable to the field position than the calibration coefficient method.The method achieves measurement accuracy of 0.2 pixels and 0.3~0.7 pixels respectively on coins and buttons.In addition,the author implements the whole algorithm using MATLAB and C++.The algorithm reaches the speed of about 108 fps on 640*480 image while 9 fps on 2448*2048,which meets the industrial real-time requirements.