Research And Application Of Sub-pixel Displacement Measurement Algorithm

Digital image correlation method(DSCM)is widely used in experimental mechanics and other fields of science research and engineering practice,it is a common technique of optical mechanics.Common application scenarios include displacement measurement of objects,electronic device vibration and deformation measurement,biomechanics and so on.The key of the DSCM is the accuracy and stability of sub-pixel displacement registration algorithm.Therefore,this research is of great significance.In this paper,two sub-pixel displacement measurement algorithms are proposed,and an object displacement monitoring system is designed to meet the requirements of real-time processing.First of all,this paper introduces the significance of the research and the domestic and overseas research progress.Then it introduces the basic principle of DSCM and the correlation function commonly used;summarizes the characteristics of sub-pixel displacement registration algorithms,with emphasis on image interpolation algorithm,curve surface fitting method,gradient-based algorithm,these three kinds of algorithms have been used most recently.Secondly,based on the analysis of the advantages and disadvantages of the two methods,image interpolation algorithm and curve surface fitting method,a hybrid algorithm is proposed.The algorithm can obtain higher measurement accuracy than the curve surface fitting method without greatly increasing the computational complexity,and has better anti noise performance.Numerical experiments are carried out to discuss the influence of different interpolation times and different surface fitting windows on the hybrid method.Finally,the anti noise performance of the algorithm is analyzed.Thirdly,according to the characteristics of machine learning algorithm which is good at capturing the potential rules,a sub-pixel displacement registration algorithm based on GBDT is proposed.The basic principle and important parameters of the algorithm are introduced,a numerical experiments were carried out to discuss the influence of the number of iterations,the maximum depth,maximum decision tree leaves and characteristic length of the algorithm,and then gives the optimal parameters of the current experimental environment.The experimental results show that the algorithm has the same performance as the curve surface fitting method.Finally,a set of displacement monitoring system is designed and built.The system can accurately measure the displacement of the target in real time,and the system test shows thatthe system can achieve sub-pixel accuracy.