Image Three-dimensional Displacement Measurement Method Based On Particle Swarm Optimization

In the modern industrial society,the deformation measurement for large civil engineering structures and mechanical structures is an important monitoring,which means to ensure its use in the process of use.Traditional contact mechanical measurement can't meet the requirement.The DIC measurement method based on binocular stereo vision is a non-contact,high-precision measurement method.In practical applications,distortion,zero drift,corner detection inaccuracy and matching will result in the problem of low measurement accuracy.In order to solve this problem,this thesis' proposes an image threedimensional displacement measurement method based on improved PSO algorithm.The main research contents include:Matching algorithm of digital image correlation method and its theory are studied in detail.A DIC method combining integer pixel matching algorithm and sub-pixel positioning algorithm is proposed to search the position of the measured point on the deformed image.The Matlab software is used to simulate the speckle pattern for comparing several common algorithms included in the two algorithms.The experimental results show that the hillclimbing search method is an integer pixel matching algorithm combined with the inverse combination method as the DIC method of the sub-pixel localization algorithm,and its matching precision and speed are optimal.Based on the three-dimensional displacement measurement method of binocular stereo vision is studied.A three-dimensional reconstruction mathematical model with the left camera optical center as the origin of the world coordinate system is constructed,and the spatial transformation matrix of the left camera coordinate system to the right camera coordinate system is calculated.The coordinate formula is used to calculate the threedimensional coordinates of the point to be measured in the left camera coordinate system,that is,the three-dimensional coordinates of the point in the world coordinate system.According to the DIC method,the position of the measured point on the image before and after the deformation is calculated,and the three-dimensional coordinate of the position before and after the deformation in the world coordinate system is calculated for the threedimensional displacement.An adaptive inverted S-variant PSO-BP neural network is proposed to calibrate measurement data.The general formula of sigmoid function is introduced into PSO algorithm to propose inverted S-type weight decrement strategy.The influence of its parameters on inertia weight and algorithm performance is discussed and the optimal parameters are finally selected.At the same time,a kind of selectivity based on population average granular distance is also proposed.The variation factor of variation is introduced into the PSO algorithm to expand the population search space and increase the possibility of the algorithm searching for the optimal value.Compared with the standard PSO algorithm,the improved PSO algorithm respectively improves the speed of the three test function by 84%,87% and 80.2%.The improved PSO algorithm improves the prediction accuracy of the network by finding the best initial weight threshold of the BP neural network.The equipment needed for the experiment was set up,and the three-dimensional displacement measurement experiment based on the image measurement method longitude verification were completed.The result of the three-dimensional displacement measurement was selected as the data set of the network fitting calibration test.The accuracy of the method described in thesis is verified by comparing the error between data which is from before and after the network calibration and the actual data of the electronically controlled stage.The relative error range of the X/Y/Z direction after calibration is respectively reduced by 65%,37.5% and 66.7%.The results show that the measurement accuracy of the adaptive threedimensional displacement measurement method optimized by the inverted S-type PSO-BP neural network is obtained is higher than before optimization.