| Inspection of dimension and shape of automotive panel is considered as one ofimportant measures by which the industry could evaluate manufacturing quality of theautomotive panel. And manufacturing quality of panels directly affect accuracy ofassembly, safety of motor vehicle and visual aesthetic feeling. In industrial plants, inspectionis usually performed in two different ways:100%inspection and batch inspection. A CMMas a traditional inspection system is designed only for off-line sampling inspection. Wheninspecting, though the accuracy of the CMM can reach0.0001, the probe must be used totouch the surface of the panel. The point-to-point measurement process cost too much timeand makes a CMM not feasible for100%inspection of automotive panels.The inspection technology based on coded structured light applied in this thesis is akind of machine vision, which has the strengths of non-contact, wide range and FOV (fieldof view), high precision, real-time inspection, active control and so on. So machine visionhas been successfully implemented in various applications. And then because spatialneighborhood strategy based on M-arrays which is one of structured light coding strategiescould be used to inspect dynamic objects, so it has a large development potential in real-timeinspection of automotive panel. The method applied in this thesis is designed for theinspection of the smoothness, shape and dimension of panels. First vision inspection schemefor dimensional inspection is proposed in this thesis, and then intensive researches onstructured light technology were made. At last3D Point-cloud could be calculated by stereotriangulation.For structured light system in this thesis, plane-based calibration strategy is applied todetermine the intrinsic and extrinsic parameters of the camera and projector and geometryrelationship between the camera and projector. And the results show that plane-based calibration strategy could simplify the calibration tasks and satisfy the required precision.Taking the requirement of real-time inspection into consideration, the projecting patternbased on spatial neighbors is designed for one-shot projection. Primitive values are assignedas0,1,2and3in accordance with X-point primitives’ orientation. Because when leastHamming distance is3constructing M-arrays is so difficult and cost too much time, soaccording to the characteristic of automotive door’s surface M-array is designed byconnecting two sub-arrays whose least Hamming distances are1and3respectively. The lastprojecting pattern based on M-array and X-point is created.In the process of inspection experiment, to build the relationship between projectingpattern in projector coordinate frame and projected image in camera coordinate frame,decoding of image taken in experiment is necessary. So image pre-processing method, edgeextracting algorithm, primitive segmentation and marking algorithm, corner extractingalgorithm based on centroid and codeword determination algorithm based on orientation ofprincipal axis and base axis are successively developed and implemented to finish thedecoding task. And then one-to-one matching procedure of characteristic points is performedby a recursive search algorithm until all corresponding primitives are found.After the corresponding pixel matching has been performed, the parameters acquired incalibration procedure are used to reconstruct the3D points cloud. After that ICP-based pointclouds registration method is applied to register the measured point clouds with point cloudmodel acquired by CMM. And then errors between measured points clouds and points cloudsmodel are calculated and quality of the measured point clouds can be optimized for errormap generation. The developed structured light system show good results for dimensionaland shape inspection of automotive panel. But the technologies need to be improved further. |
PDF Full Text Request