Reasearch On Non-contact Measurement Technology Of Stereo Vision Based On Polarization Information

With the rapid development of modern industry,it requires faster and more accurate realtime detection of products and as well higher measurement accuracy.More emphasize has been put on the research of non-contact measurement with higher measurement accuracy and quicker speed in machine vision field.The traditional measurement hasn't met the variety and surface complexity of products any more,while the non-contact measurement technology based on the binocular stereo vision system can achieve a better measurement accuracy under the condition of no-contact and no-damage.It utilizes CCD cameras to capture the object from various angles and the image features extraction and match algorithms are applied to process the operation of features point detection.And then the detected points are transformed to the three-dimensional space for a better information reconstruction.A higher measurement accuracy of products and better security in production are guaranteed based on non-contact binocular stereo vision.Under the influence of normal illumination,highlight existing in the image acquisition makes a big difference in the binocular stereo vision system.Research on the technology of image highlight detection and removal is essential because it covers the internal texture and color features of object.Non-contact measurement of binocular stereo vision system based on the polarization is proposed in this paper,and the corresponding research achievements are explained as below.(1)Image highlight detection and removal system is established in this paper.Images under different polarization angles are acquisited for operating the algorithms of dark-channel and mean-shift in order to extract the highlight pixels.Then highlight pixels in different images of various polarization angles are introduced to BP Neural Network,and the Gaussian is applied to low the highlight pixels effectively.As illustrated in the experimental results,the method of image highlight detection and removal is verified its feasibility and better performance.As for the processed image,the corresponding pixels have a better distribution and highlight pixels are effectively limited.(2)Camera calibration based on polarization information is constructed in this paper.As for the chessboard images captured by CCD camera,the polarization Stokes components is utilized for computing the optimal polarization angles with the preconditions of image acquisition of 0° 45° and 90° polarization angles.Then the chessboard images under optimal polarization angles are applied to process the camera calibration.As shown in the results of corner extraction,the chessboard images with polarization information has a better performance of corner extraction,and it applies to the real space distribution of chessboard image corners.(3)Image features extraction and match system based on polarization system is proposed in this paper.SIFT and SURF features are chose as the description of image features extraction.And the RANSAC algorithm is introduced to achieve a better performance of image features match.As for the image acquisition under normal illumination and optimal polarization angles,the number of latter image features is many more than the former while the polarization images can obtain a better performance of image match.(4)Image depth information is studied in this paper while the algorithms of stereo match are tested on the images with optimal polarization angles.The 3D points cloud and texture image are described on the basement of PCL software.(5)Finally,measurement accuracy of binocular stereo vision system based on polarization proposed in this paper is tested on the experiment of object measurement.As clearly shown in the results,the measurement accuracy of the method proposed in this paper is 87.3167%,which shows the feasibility and higher performance of the non-contact binocular stereo vision system based on polarization information.