GPS/INS is considered the most popular methodology for land vehicle navigation because this system fills the GPS data gap in urban environment. However, any type of GPS/INS system still suffers from frequent occurrence of GPS outages since INS can only provide short-time and high accuracy navigation solution. Hence, alternative aiding sources or techniques can be utilized to meet the superior navigation solution. Among various vision sensors, the omnidriectional vision sensor is the most popular in sensor fusion field since it can provide much more information from the environment owing to wider field of view. In this paper, we propose a novel system that integrates GPS, INS and Omnidirectional vision sensor to overcome time-dependent accumulated position error from single INS solution. Specifically, it first automatically extracts and tracks vanishing points in Omnidirectional images, which can compute the rotation information of the vehicle. After rotations are determined, translations or motion estimation can be determined by the intersection of two epipolar planes obtained from two pairs of point correspondences. The use of motion based measurements can play a crucial role of aiding source in GPS-denied environment to improve the single INS performance. Extensive experiment demonstrated that the proposed omnidirectional aided approach reduce the accumulated position errors by 30% over conventional odometer aided approach during 100 second GPS outages.

Keywords: omnidirectional vision sensor, GPS/INS, epipolar plane