Research On Methods Of Three-Dimensional Motion Stereo Vision Measurement

As a non-contact measurement technology,computer vision measurement technology which integrate close photography and computer measurement technology has shown its importance and wide development prospects.Especially the three-dimensional measurement technology based on computer vision can accomplish big space,multi-objective precision,and online measurement.In the three-dimensional motion measurement based on stereo vision motion measurement technology was more practical by using two or more high-speed cameras with a higher measurement accuracy and stability.Based on the existing research of computer vision motion measurement,a sequence of images of encoding marker motion measurement method was proposed to improve the accuracy of measurement.For this system,we must solve some key problems,such as high precision identification and positioning for encoding marker,stereo cameras calibration and three-dimensional visual pose estimation,etc.Firstly,in order to accurately locate and identify the coded marks,a morphological mark recognition algorithm based on feature is proposed to improve the practicability of the algorithm and the recognition rate of the complex environment.The mathematical morphology operator identification code applied to the identification point,the clustering of signs point to eliminate background noise,that signs point to do several image acquisition after the morphology operation to extract a region where each point mark border,the last point feature vectors accurately identify the position of each landmark and the code number according to flag raised.Secondly,for vision measurement round landmarks centering problem,based on the camera's perspective projection transform and spatial analytic geometry theory,an improved Zernike moments of the circular center precision positioning algorithm is proposed.First,round space perspective projection transformation errors generated are analyzed and concluded the basic principles related to,next use the constructed Zernike moments to solve the model,and then analyzed the error of the principle and ideal model deviation moment templates generated,propose amendments formula to compensate and using improved edge criterion for sub-pixel positioning of the image edge.Finally,the least-square ellipse fitting algorithm is used for fitting center.Reversing edge points to filter the residuals larger point,and then the center of the ellipse for precise positioning.This method has good stability,high positioning accuracy.Then,the calibration of stereo vision measurement system is studied.A new algorithm for stereo vision based on cross target is proposed.Multi view geometry constraint and optimization of the beam adjustment are used to obtain the internal and external parameters of the two cameras.Introduce inter camera relatively fixed constraints,for the traditional bundle adjustment method of structural equation process are improved,so that each iteration only the external parameters of a camera shooting of two images is optimized.The introduction of relatively fixed constraints can effectively increase the number of redundant observation system adjustment and the optimization accuracy higher,also the dimension of the matrix method of than the traditional algorithm reduced 1/2.Uing the error between the reprojection coordinates and original target coordinates to construct the error compensation matrix which compensates the stereo camera attitude.The method improved the calibration precision of the stereo camera.In the following,the position and pose estimation method based on feature points is studied.In the actual measurement process,the noise error is often anisotropic and non independent identically distributed,the current method does not fully take into account the uncertainty of the feature point.A kind of generalized orthogonal iteration algorithm based on uncertainty-weighted error of the feature points is proposed.Covariance matrix describes the feature point direction of uncertainty and establishing objective function of the actual application,a new uncertainty into the objective function of the manipulation method can be adapted to different levels of feature points with direction uncertain situation.Extended to stereo vision,multiple cameras can be combined to improve the system's field of view,but also can take advantage of the redundant measurement information to enhance system robustness to noise.Finally,in order to provide theoretical guidance for the design of the Three-dimensional motion stereo vision measurement system,the needs for impact motion stereo vision measurement system uncertainty of various factors is analyzed and gives the error calculation formula of the system.According to the analysis,we design and test the Three-dimensional motion stereo vision measurement system to prove the correctness of the proposed algorithm and reliability of the system.The test results show that the stereoscopic motion measurement system is capable of high-precision measurement of moving targets.