Research On 3D Testing Technique Based On Computer Vision

Computer vision is a new cross science. It's unique theory frame andtechnique have been formed during the past 20 years . It's essence is therealization of one or more algorithms which are based on the computer. So,machine vision is also computer vision. There are several kinds of system special for 3D measurement to get target's whole geometry information ,such as digital 3D probe measurement system ,3D imagery system based on contactless " structured light" or on disparity. This paper aims to obtain target's geometry features to show the coordinateof any point. The author analyzes 3D measurement theory and technique based onthe traditional principle of disparity imagery. In addition, optics, projectivegeometry, signal processing, or morphology science and error theory science areadapted to make it. This paper abstracts the whole research work. So, hardware design of imageacquisition sub-system, optical imagery sub-system, synchronous controlsub-system and some software design about algorithms are omitted for the lengthof this paper.The author supplements and improves followed theory and technique which isbased on optical disparity principle in 3D measurement system. calibration forcamera's parameters, optical undistortion of image ,acquisition of image'scorrelation coefficient from different vision ,information acquisition of themaxim flow rate and minimum distance, morphology filter and geometryreconstruction of target.The research is around calibration and corresponding points matching baseddisparity theory, which are difficult in 3D imagery system. This is shown asbelow:When solving the nonlinear equation about intrinsic parameters'calibration, maxim likelihood method is introduced as the standard ofconvergence. It is with a weakness that pre-evaluating the distance value isnecessary.When solving extrinsic parameter, the author presents a z-coordinate axis(without rotation) calibration method which simplifies calculation but can notkeep chessboard freely.In stereo imagery system, a kind of parameter trimmer method under someconstraint conditions is proposed. In addition, Gauss-Newton algorithm fornonlinear equation is improved when LM method can't work under some constraintcondition. The popular algorithm is parameters resolving first, correspondingpoints adjustment following. It1s with an advantage that makes algorithm simpleand it has no effect on every couple points. However, its weakness is also obvious.Computer has to operate n times for N couple points. The author puts forward:parameters adjustment first, calculation following. It is proved that thecalculation using popular algorithm is 0.1million times as more as the one usingthe algorithm the author presented.Parameter trimmer algorithm makes the followed 3D stereo algorithmdifferent. So, the author remodels the 3D stereo system. It is clear that thisalgorithm is not used for under-rectification system.The author uses "multi-resolution" technique to realize "coarse andfine" matching. Moreover, an accurate solution for equation with one unknownquantity is described.When solving the equation with unknown matching points, the author showsa algorithm that is: correlation coefficient tomogram to maxim flow rate andshortest path other than popular algorithm that is correlation coefficientanalysis to subinterval to intervals integrated to correlation coefficientreanalysis to acquisition for the minimum path. They have features that mediumspeed and filter free;high speed and efficiency with more noise and signalneeding filter respectively. Then, the author deals with filter process bymorphology method and discusses "open" algorithm to remove spike and littlebridge.