The Research And Implement Of Fisheye Image Distortion Correction

Fisheye lens has a large view of180°~270°, for the particularity of its optical properties,which is widely used in different kinds of monitoring field, like transportation, bank, school,etc. It also can be applied to varieties of panorama system, with the superiority of simplefusion, full covering and no blind area. However, the large view image captured by fisheyelens is introduced into serious distortion, and the deformation image is hard to meet people’snormal visual requirements. In order to enhance the visual effect of image, the distortedfisheye images need to be recovered to the undistorted normal visual images.The distortion correction of fisheye image is based on the parameters of the lens, theseparameters can be obtained through the calibration of camera. Among all kinds of calibrationalgorithm, the method based on the camera imaging model is the most accurate one. Whenusing this calibration method, for the convenience of calculation, we extract the feature pointsin the fisheye image to replace the whole image, corresponding to the real world coordinate.This research about the fisheye distortion correction mainly includes the following threeaspects contents: the feature point detection, fisheye camera calibration and image distortioncorrection.Firstly, several current algorithms in the field of corner detection are researched, Harris,SUSAN, SV, etc. After analyzing and comparing the advantages and disadvantages of thesimulation results of these algorithms, I studied the algorithm deeply, which is put forward byMartin Rufli et, based on closed contour extraction, and then improved some inadequacies ofthis algorithm. Finally we realized a high precision and high accuracy automatic cornerdetection algorithm.Second, the camera calibration is the precondition of image correction, which is also acrucial step. The accuracy of the calibrate results directly reflects the accuracy of the camera imaging model. Among the existing camera calibration algorithms, the algorithm based oncamera imaging model can achieve more accurate results, its basic idea is to build acorresponding relationship between real world coordinate space and the fisheye image planecoordinate system to solve the distortion parameters. The representative algorithm is proposedby Davide Scaramuzza. Aiming at the limitations of the calibration method, some measuresare improved to achieve low constraints, high precision of calibration results.Finally, the paper analyzes several correction algorithms, including plane correction,cylindrical correction and longitude and latitude unfold correction, etc. These algorithms arebelong to global projection correction, mapping the distortion image to the target un-distortionimage directly. There are inevitably some certain degree of deformation in the image aftercorrection. Carroll puts forward the correction method based on the content. This methodminimizes the distortion by the changing projection model. The correction results can satisfythe demands of normal vision. Aiming at the limitations of its user interaction, we propose acheckerboard image correction algorithm based on Carroll.This research results can be applied to panorama splicing systems，which use DM6437development board as the core hardware platform. The experiment results show that thecorrection algorithm is well.