Multi-Sensor Fusion for Navigation of Ground Vehicles.

In this paper, a navigation solution for a ground vehicle is developed by fusing navigation data from Inertial Navigation System (INS), Visual Odometry (VO), and Global Positioning System (GPS) using a Dual Extended Kalman Filter (DEKF) algorithm. The research contributions are divided in two stages. The first stage presents VO navigation system termed as Modified Stereo Visual Odometry (ModSVO) which modifies the pose estimation and pose optimization segments of the traditional stereo vision pipelines to provide an algorithm which is shown to improve accuracy when compared with the tradition Stereo Visual Odometry (SVO) approach. The second stage presents the development of INS/VO/GPS integrated navigation system using DEKF. The developed navigation solution is shown to outperform the INS/VO integrated system in case of VO failure and outperform the INS/GPS integrated system in case of GPS failure. The experimental evaluation is conducted on the well-known KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) real-world dataset.