halcon機(jī)器視覺8三維定位方法

1、HALCON三定位段德山總監(jiān)恒（集團(tuán)）圖像視覺技術(shù)分公司© 2003-2012 MVTec Software GmbHTypical ApplicationsPick-and-Place Micro electronics AlignmentPrint inspectionRoboticsHALCON offers matching tools for any requirementWhich matching method suits your application?Perspectivedeformations3D surface3D contours3DStartLocal

2、deformationsRigid transformationComponentsColor(Aniso-)Scaling© 2003-2012 MVTec Software GmbHDescriptor-based MatchingHALCON supports descriptor-based matchingChoose betweencalibrated and uncalibrated matchingUncalibrated returns perspective mapCalibrated returns poseThe 3D pose of the mmight d

3、iffer from the reference poseImage planeMROIwith moriginWorld plane defined by reference poseMposeOffsetReferencePosefind_calib_descriptor_m returns a posexczcxmycymzmPoseDescriptor-based matching is done in four steps1. Train interest points2. Detect points3. Match points4. Calculate mapOfflineOnli

4、neDescriptor-based matching is best for text or printsUse textured imagesRobustscaling, tilt and clutterTraining takes a long time with many interest points6Do notuse periodic patternsMetallic objects are badOnly perspective distortions can be found© 2003-2012 MVTec Software GmbHDeformable Matc

5、hingHALCON finds objectswith perspective distortionsHALCON finds objectswith perspective distortionsFirst, create a planar deformable mCreate mofflineonlineClear mFind mCreate a mfrom a planar part of the objectCalculate a reference pose for the mThe 3D pose of the mmight differ from the reference p

6、oseImage planeMROIwith moriginWorld plane defined by reference poseMposeOffsetReferencePoseSecond, find the monlineCreate mofflineonlineClear mFind mA coarse-to-fine strategy speeds up detectionThe shape changes only a little after deformationOn each pyramid levelonly small corrections are neededfin

7、d_planar_calib_deformable_m returns a posexczcxmycymzmPoseFinallythe memoryCreate mofflineonlineClear mFind mPerspective, deformable matching is best for objects with clear planar edgesOUse objectsGood for metallic objects with clear edgesPlanar surface needednly perspective distortionsPolarity chan

8、gesDo not mix up can be foundslow down searchdifferent object planes6© 2003-2012 MVTec Software GmbHShape-based 3D matchingShape-based 3D matching§For a flexible production, the picking of objects with robots is crucialTo realize this task, the 3D position of theobjects have to be determin

9、ed firstIn addition, the objects are typically not planar, but have a 3D shapeTo solve this taskHALCON offersshape-based 3D matching§§§Pose estimation with shape-based 3D matchingnnnnnnCan handle general objectsDoes not require manual preprocessing or segmentation No approximate pose

10、must be knownThe pose can be determined without ambiguityRobustness to occlusions, clutter, and contrast changesA CAD mof the object must be providedHow it worksBased on a 3D CAD m, a 3D shape mis created.The 3D shape mconsists of 2D views of the object.For this, the object is assumed to be at the c

11、enter of a sphere.The virtual cameras are placed on the sphere and point to its center.Shape-based 3D matching© 2003-2012 MVTec Software GmbHSurface-based 3D MatchingHALCON offers surface-based 3D matchingSurface-based 3D matching workflowConfigurationCreate mcreate_surface_mEvaluationPerform m

12、atchingImprove accuracyGet resultsget_surface_matching_resultrefine_surface_m_posefind_surface_mHALCON offers surface-based 3D-Matchingcreate_surface_mCreate mget_surface_m_paramfind_surface_mDetermine mposerefine_surface_m_poseget_surface_matching_infoGet additional informationwrite_surface_mMIO_su

13、rface_mclear_surface_mclear_surface_matching_resultCleanup memoryclear_all_surface_msSurface-based 3D matching is easy to trainCAD-DataSurfacemTrainingOfflineOnlineSearchExamples for 3D ms from CAD dataSurface-based 3D matching is easy to trainObjectTraining imageStereoSurfacemSheet of lightTraining

14、SearchSurface-based 3D matching is robustSurface-based 3D matching is robust even with erroneous dataSurface-based 3D matching is robust even with erroneous dataSurface-based 3D matching is robust even with low resolutionDepth image from TOF camera (176 x 144)Object size about 70 x 45Engine partSurf

15、ace-based 3D matching is robust even with low resolution (TOF data)Surface-based 3D matching is robust even with low resolutionSurface-based 3D matching is robust also with occlusionSurface-based 3D matching is robust even with occlusionSurface-based 3D matching is robust even with occlusionSurface-

16、based 3D matching is robust even at (outside) the image borderSurface-based 3D matching is fast and offers the full pose rangeDifferences between surface-based 3D matching and shape-based 3D matchingShape-based 3D matchingSurface-based 3D matchingclear geometric edges2D image CAD datatypically restr

17、ictedhigharbitrary shape 3D point cloudCAD data or 3D point cloudfullmediumObject Search image Training Pose rangeMemory© 2003-2012 MVTec Software GmbHHand-Eye CalibrationImproved hand-eye calibrationThere are two scenarios for hand-eye calibrationneededknownfrom hand-eye calibrationtooltoolcam

18、eracamerabasebaseobjectobjectMoving cameraStationary cameraHand-eye calibration has become easierknownfrom hand-eye calibrationcameratoolbaseMoving cameraHand-eye calibration has become easierknownfrom hand-eye calibrationtoolcamerabaseStationary cameraMoving CameraChain of transformations for a mov

19、ing camera systembase-1 baseHtool ×(H)×Htoolcalcam H=cam calHand-eye calibration is done in two steps1. Prepare calibration data m2. Calibrate hand-eye systemStationary CameraChain of transformation for a stationary camera systemcam H=cam calHbase ×H×HbasetooltoolcalHand-eye cali

20、bration is done in two steps1. Prepare calibration data m2. Calibrate hand-eye systemObject pose in base coordinates (moving)neededknownpose_invert( ToolInCamPose, CamInToolPose)𝑐𝑎𝑚H𝑡𝑜𝑜𝑙𝑏𝑎𝑠𝑒H𝑡𝑜𝑜𝑙pose_compose( ToolInBasePose, CamInBasePose)CamInToolPose,𝑐𝑎Ү