Research On Three-dimensional Reconstruction Of Infrared Multi-camera Vision Based On Opencv

This paper is studied on infrared Multi-Vison which is based on visible binocularvision,its researches are mainly aimed at8~14mband infrared light,theinformations collected on the study object or scene by infrared thermal imagersthrough that band from different angles,and then combined the infrared thermalimages two by two to obtain spatial coordinates of the point clouds,and restored3Dinformations of the object or scene which reconstructed the pont clouds by Matlab.In this paper,an experimental platform is built with infrared thermal imagers tocollecting images,also used Microsoft Visual Studio2009and OpenCV（open sourcecomputer vision library) as development tools,put the collected informations as studyobject,aimed at the calibration of infrared thermal imagers,infrared thermal imagematching and the3D reconstruction of the object or scene.The specific work is asfollows:1.Calibrated the infrared thermal imagers by MATLAB calibration toolbox andcalls to the OpenCV Library,and compared the advantages and disadvantages of thetwo calibration.The calibration plate selection directly affects the internal and externalparameters of infrared thermal imagers in this method.Pointed at8~14mbandinfrared thermal imager,proposed for a calibration plate applied to near infrared stereovision Because the emissivity between black paper and aluminium plates in the sametemperature differs greatly,so we use the calibration plate which constitute by3cmsquare shape black paper and aluminium board.Its advantages are simple manufacture,low cost, obvious effect and meet the extraction of internal and external parameters ofthe infrared thermal imagers.2.The image matching is based on the feature point matching of epipolar lineconstraint.First,corrected the images,and Constrainted the matching points at the sameepipolar line to increase the matching spreed and matching precision.The matchingalgorithm used SIFT operator to extracting feature points and matching featurepoints,then reconstructed the point cloud by matlab...