Real-time Construction Of Panoramic Images Based On Multi-camera

With the rapid development of economy and technology, it is increasingly more difficult for drivers to park due to the considerably increasing number of vehicles. The inherent problem of blind spot during driving has given rise to formidable obstacles to safe driving. A panoramic image construction system has been presented in this thesis based on multi-camera with four fisheye cameras fixed on the front, rear and sides of a vehicle. The system processes the collected images in real time and constructs a panoramic bird-view image to eliminate the blind spots around the vehicle and improve driving accuracy and security.First of all, this thesis elaborates the method of preprocessing and data extracting before camera calibration. After the process of image denoising, LOG operator and first-order differential expect value algorithm are combined to gain edges of the ellipse rectification points at a sub-pixel level. Thus the center coordinate data has been obtained. The method is proved to be of high accuracy and low calculation amount.Secondly, aiming at analyzing the features of fisheye camera, and on the basis of radial symmetric geometric model, a more accurate full model for fisheye camera has been presented. Meantime, this thesis presented the multi-view calibration algorithm based on fundamental matrix, imitating the process of finding epipole to get distortion center. Then additionally initial values are set to gain all the intrinsic camera parameters by iterative optimal algorithm. Comparing improved RAC two-step algorithm with the innovative method, this thesis carried out simulation experiments and the results indicate that the new algorithm has higher rectification accuracy.Thirdly, image transformation homography is employed to calculate the world coordinate invariants with image coordinate through DLT algorithm. Then image warping is executed for one plane of the target image to get a synthesized view. Meanwhile, this thesis provides the method of getting camera rotation matrix by both the homography matrix and H pattern recognition mark on the ground introducing in vanishing point restraints.Finally, according to the integration of above key technologies, relative experiments are conducted on the basis of real-time construction system of panoramic images based on multi-camera. Backward rectification is also carried out for collected images to improve real-time performance. Then the corrected images are warped into a top view with homography matrix. After that, image stitching is implemented and the system framework is built through the coordinate mapping table. The experimental results indicate that the accuracy and real-time performance of our algorithm is favorable. Furthermore, once the calibration data is obtained, all the system parameters are fixed. Thus, the method in this thesis can eliminate all driving blind spots and finally create panoramic bird-view image for actual demands.