Research On Key Technologies Of Micro Vision In Micro Assembly

In the assembly process of micro part,the micro vision system is the key section of the assembly positioning information feedback.The camera calibration is the basis of measurement,and its accuracy directly affects the assembly accuracy.Clear image is the precondition for the precise acquisition of positioning features.Real-time focusing can be finished by zoom mirco vision system,which can overcome the contradiction between fixed magnification and measuring target scale change,and the contradiction between fixed focal length and measuring the defocus of the object.In addition,the feature edge of parts as assembly measurement information is very important for the whole assembly process.Precise feature recognition and positioning can ensure assembly accuracy.When the edge of the feature has defects and occlusion,the assembly measurement information can not be aligned with the assembly datum.By studying a matching method of part edge and model,the model is used to align the assembly datum to ensure the assembly process to proceed smoothly.In summary,the study of camera calibration,autofocus,part feature recognition and positioning,and model matching techniques in micro vision will help improve the overall performance of micro-assembly systems.The main research work of this article is as follows:(1)The calibration technology of zoom micro vision system in micro-assembly is studied.Firstly,the principal point of image is determined by method of convertible magnification.Based on traditional Zhang's calibration method,the linear calibration under fixed magnification is firstly performed by homography matrix decomposition of single view,then the distortion model and the Quantum-behaved Particle Swarm Optimization(QPSO)are employed sequentially to do nonlinear optimization for the linear calibration result.After nonlinear optimization,the maximum re-projection error is 0.13 pix and the average re-projection error is about 0.1pix.Furthermore,the calibration for magnification at arbitrary working condition is completed by Gaussian curve fitting,and the experiment verified that the rotation parameters in the external parameters are basically unchanged under different magnifications.(2)For real-time automatic focusing,the method of maximum gradient threshold of eight-neighborhood and gradient threshold are used,for the traditional gray gradient function only considered the fixed gradient direction and is susceptible to noise.The auto-focusing algorithm used in this paper is compared with the computational efficiency and anti-noise performance of several other gray-scale gradient focusing algorithms,and the focusing positioning experiments under different magnifications are completed.(3)Several edge feature extraction algorithms and least-squares iterative fitting are used to complete feature recognition and location of parts.For the partial occlusion of the part positioning features,preliminarily photographing the complete part image without occlusion and the partial occlusion image of the part are matched by SURF algorithm.Using the locally matched feature points in the two images,the complete part image is translated and rotated to replace the occluded part image.Aiming at the problem of maching defects in part positioning features,a method of matching the edge of a part with a corresponding CAD model is studied.Minimum bounding rectangle and principal axes centroid method are comprehensively considered.A reasonable method is chosen to achieve rough matching according to the different conditions of the edge of the part feature,then the precise matching process is finished by ICP algorithm.Finally,we can use the corresponding model edge instead of the positioning feature edge of part.