Research On Camera Calibration Method In3D Reconstruction

3-D reconstruction is one of the key points of computer vision research, andcamera calibration is an indispensable step in3-D reconstruction. The relation betweenthe3-D geometric positions of the points on the surface of object and the correspondingpoint of2-D image is decided by camera’s imaging model, and these geometricparameters is called camera parameters. Camera parameters include the internalstructure of camera parameters such as the coordinate of main point and focal length,and the external parameters such as the rotation and translation of the camera. As the3-D reconstruction technology is widely applicated, camera calibration has become oneof the research hotspots. The research of camera calibration pursues simpler, faster andhigher precision of calibration process.This thesis summarizes the camera calibration technique comprehensively,including introduction of background, present situation of calibration research,categories and rudiments of camera model. This thesis also studies image preprocessingtechnology of camera calibration process, and applies it in image process of calibration.A geometrical constraint based corner detection method was presented, which iseffective to improve the classic camera calibration method. After several evaluationcriterions are summarized systematically, a new evaluation method, weighted averagemagnitude of error, is proposed. The new ideas in this thesis are summed up as follows:1) A new improved adaptive median offset filtering based on the statistic of theneighborhood pixels is put forwarded, and is applied in image preprocessing ofcalibration. The restored calibration plate images can obtain superior peak signal tonoise ratio, which improves the subsequent image processing steps and achieves goodeffect in the process of camera calibration.2) A geometrical constraint based corner detection method was proposed. Firstly, atraditional corner detection algorithm was used to extract corners from the image, andthen the edge points of the tessellated pattern was found by using the detected corners asinitial seeds．The edge points were then fitted with a set of cubic polynomial curves andthe intersections of the curves were used as real corners in the image, Finally, theextracted corners were used to perform the task of camera calibration．The experimentalresults showed that the improved method can greatly improve the performance offeature detection and camera calibration. 3) After several evaluation criterions are summarized systematically, the method ofweighted average magnitude of error was proposed, which can evaluate the result ofcamera calibration.