Research And Applications On Vision-based Structural Motion Extraction Algorithms

As a fundamental technology of modern industry,precision and efficiency of the measurement techniques play important roles in national science.In recent years,the requirement of measuring precision and speed become more and more strict because of the development of industrial manufacturing and processing technology.With the rapid advance of digital imaging and graphics processing unit,the resolution and frame rate of image sensor also get higher.More attention has been paid into the vision-based measurement technique for advantages,such as non-contact,no added mass,full field,high precision and intelligence.Vision-based techniques have been widely applied in dimension measurement,object identification and tracking,structure motion analysis and photoplethysmograph imaging.The invention of high-speed imaging sensor enables record and analysis of motion that happens instantaneously with high frequency.This thesis takes the high-speed vision measurement system as the research object,also studied the hardware construction and related image processing algorithms of the system,and then proposed several measuring algorithms for object motion in different situation.Our thesis firstly introduced the present situation and developing process of the vision-based measurement and gave an analysis of the hardware selection.With proper selection of the hardware in the system,vision-based technique can realize non-contact measurement in large length scale and different distances.This thesis also derived the coordinate relationship in geometrical imaging principle and specified the error source in camera calibration that using simple calibration board.Convectional means of image processing are commonly used to analyse video during high-speed measurement.However,these methods expose many shortcomings in practical applications,such as multi-parameters adjustment,less automation and less robustness.To overcome these shortages,this thesis studied the information extraction algorithms for various structure's motion such as translation,rotation and subtle vibration.For the translation issue,the simplified optical flow and the quadratic surface fitting approaches were employed to improve the integer pixel level image matching algorithms to sub-pixel accuracy.The modified algorithms executed much faster than the ones with pixel interpolation and achieved acceptable sub-pixel accuracy.The normalized root-mean square error between grating rule displacement sensor data and displacement extracted by using the proposed algorithms are less than 1%.The time consuming of extraction between frames are less than 1ms so that real-time measurement around 1000fps can be realized.For structural rotation measurement,the Lucas-Kanade and the enhanced correlation coefficient(ECC)maximization image alignment algorithms and their fast inverse compositional(IC)algorithms were used to extract rotation parameter from axis and radius direction.A nonlinear warping function was proposed to describe the relationship between template image and object image and the iteration mode of rotation parameter was optimized for higher computational efficiency.Last but not least,vision-based system was applied to extract subtle intensity or motion changes in image sequences in this thesis.In video magnification algorithms,subtle intensity or motion vibrations should be precisely extracted before magnified and reconstructed into the original video.Concerning about the inefficiency of the signal extraction process,this thesis introduced a novel efficient algorithm based on SVD decomposition that is able to extract subtle changes in image sequences.All the proposed algorithms were integrated into the software of the constructed vision-based high-speed measurement system and achieved good results in practical applications.The measurement system was used to extract vibration signals of forklift's steering wheel system and sound barrier that installed on the railway bridge,the rotation parameters of rolling bearing and sound induced subtle motions.The test results implied that the visual measurement is a reliable,effective way of measuring,and has important practical significance.