3D Pose Determination and Non-linear Image Registration of Circular Traffic Sign

摘要

3D pose determination and nonlinear image registration of traffic sign plays an important role in visual guidance of intelligent transportation system. The ellipse parameter matrix of the borderline of a circular traffic sign is firstly optimally estimated. Then, by means of its eigen decomposition, the 3D pose of traffic sign is computed and, in turn, the 3D coordinates of all points on the traffic sign is given out. According to the priori information of building traffic sign, the 3D motion from traffic sign to camera is restored, from which the inverse Euclid transformation recovers its corresponding 3D coordinate in modeling process. At last, the projection of the model’s coordinate gives out the non-linear registration of image pixel. The developed algorithm can not only determine the 3D pose and motion of traffic sign, but also eliminate the non-linear distortion caused by prospective projection. The nonlinear registration of traffic sign image can significantly increase the rate of correct recognition of it. The theoretical analysis and experimental demonstration have proved that the developed algorithm is fast, robust and practical.