Research And Application Of Mark Point Sub-pixel Location Algorithms On Mobile Screen

The huge smart phone market promotes the trend of smart,efficient and high-precision mobile phone production equipment.High-precision positioning based on machine vision is an important part of the research and development of smart devices.Although the mature commercial algorithm library of machine vision has abundant functions,it has limited performance in some specific scenarios and high authorization cost.In this paper,the high-precision location algorithm of cross Mark points on mobile screen is studied,and a method of sub-pixel registration is proposed,which is based on whole-pixel registration and sub-pixel offset calculation.In the integer pixel registration,two registration algorithms based on gray scale and shape are implemented in this paper.Aiming at the inefficiency of the traditional gray-based registration algorithm,this paper proposes the strategy of setting translation step,rotating step and using image pyramid to improve the efficiency of the algorithm.The shape-based registration algorithm uses the improved normalized image gradient product with better robustness as the similarity measure.The algorithm efficiency is improved by early termination of calculation,improved rotation search strategy and image pyramid.Then,in the calculation of sub-pixel offset,the advantages and disadvantages of different sub-pixel registration algorithms are analyzed.Then,in the calculation of sub-pixel offset,the binary quadric surface fitting method with higher accuracy and efficiency is selected to calculate sub-pixel offset,and the solving process is simplified by translating the relevant pixels.In this paper,the robustness,speed and accuracy of the algorithm are compared with the benchmark VisionPro-PatMax algorithm in the commercial algorithm library of machine vision.The test results of multi-group image data sets show that the two image registration algorithms proposed in this paper can achieve high computational efficiency;the gray-level-based registration algorithm has a high registration rate in the case of stable light source,while the shape-based registration algorithm has a stronger robustness in the case of non-linear illumination change,contrast change,low signal-to-noise ratio and partial occlusion.The maximum horizontal and vertical absolute errors of the gray-based sub-pixel registration algorithm are less than 0.1 pixels;the mean horizontal and vertical absolute errors of the shape-based sub-pixel registration algorithm are less than 0.15 pixels,and the maximum absolute errors are less than 0.25 pixels,which verifies the high registration accuracy of the sub-pixel registration algorithm proposed and implemented in this paper.Finally,the proposed and implemented sub-pixel registration algorithm is applied to the mobile phone screen vision alignment platform,and the software system of the alignment platform is developed in a modular way.