Extraction And Decoding Technology Of Circular Coded Markers

Circular coded markers are widely used in the field of three-dimensional(3D)measurement for their easy identification,high positioning accuracy and good stability,especially in camera calibration and photogrammetry.The decoding of the marker is related to the matching of images from different perspectives and the fusion of depth data,and the center extraction of the marker directly affects the accuracy of measurement system,which are two key factors of 3D reconstruction.There are some problems in the current center extraction and decoding methods.Camera resolution,shooting angle,shooting environment and other factors will affect the decoding effect and extraction accuracy.Therefore,how to further improve the decoding accuracy and center extraction accuracy of circular coded marker is a research direction worth exploring.This dissertation mainly focuses on the extraction and decoding technology of circular coded marker in camera calibration process.The main work of this dissertation is as follows.Aiming at the decoding of the marker,the Scale Invariant Feature Transform(SIFT)and the global Hu moment are extracted as the feature description of coded marker,and the SIFT features are quantified by Bag of Features(BOF).Then,the fusion features are extracted,and the classifiers are trained by Support Vector Machine(SVM)respectively for markers with different number of connected domains.Finally,the coding value of the marker is predicted by corresponding classifier.After that,the perspective projection model is used to make datasets,and the optimal number of BOF clustering centers and feature fusion weight are selected.In the test set with imaging angle less than 70 ~o,the average decoding accuracy of this method is 97%.And this method has good stability for the change of noise.Aiming at the center extraction of the marker,a center location method based on radial line fitting of coding band is proposed.Firstly,the radial straight line edges in the coding band of marker are segmented after edge extraction,and the sub-pixel positions of edge points are extracted.Then,the extracted sub-pixel edge points set is mapped to the parameter space into curves,and the intersection points between these curves are obtained.Finally,the Random Sample Consensus(RANSAC)algorithm is used for fitting the above points into a curve,and the actual center is obtained by mapping the fitted curve to original coordinate space.The simulation results show that compared with the ellipse fitting method,the center extraction error of this method is reduced from 1 pixel to 0.05 pixel with the increase of marker size and shooting angle.This method has higher extraction accuracy and stronger stability to the change of marker size and shooting angle.Finally,a 3D measurement system based on fringe projection is built,and a plane target composed of circular coded markers is made to verify the effectiveness of proposed methods.After taking target images from different positions and angles,the decoding accuracy of the proposed decoding method is over 96%with the target imaging angle less than 70 ~o.This method has a significant improvement compared with the traditional method.When the target is used for camera calibration,the reprojection error of the proposed center extraction method is reduced by 20%compared with the ellipse fitting method.After camera calibration,the measurement error of 3D measurement system is less than 50 um,and the measurement accuracy is increased by 10%compared with the ellipse fitting method.It is proved that this center extraction method can improve the camera calibration accuracy and system measurement accuracy to a certain extent.