Intelligent Vehicle Localization in Urban Environments Using EKF-based Visual Odometry and GPS Fusion

Abstract

AbstractWe present a method for intelligent vehicle localization and 3D mapping in urban environments using integration of stereovision and Real-Time Kinematic GPS (RTK-GPS). In our approach, stereoscopic system is used to recover the scene geometry and to predict the camera motion. Accurate GPS positions are integrated to correct the vehicle positions and orientations using Extended Kalman Filter (EKF), and to rectify the global positions of reconstructed 3D landmarks based on the positions of reference stereo frames. The method was tested using data obtained by a real electrical GEM vehicle equipped with stereoscopic system and RTK-GPS in urban environments. The results show that the trajectory obtained by GPS/Vision integrated method can fit the ground truth better than the vision-only method, and can also avoid the position jumps aroused by GPS signal failures.