Tightly Coupled Integration of GNSS/INS/Stereo Vision/Map Matching System for Land Vehicle Navigation
This thesis proposes a tight integration architecture with an Inertial Measurement Unit (IMU) as the core sensor for land vehicle navigation. Kalman filter is used to fuse different types of measurements. The IMU provides the dynamic information of the land vehicle and it is used to predict the measurements of other sensors. When the measurements become available, the differences between the predicted and actual measurements are applied to estimate the error states in the Kalman filter. There are four types of integration implemented in this thesis. First, the IMU is integrated with the Stereo Visual Odometry (VO). In this tight integration, the camera attitude and perspective center position can be predicted by the IMU mechanization and the 3-dimensional (3D) coordinates of the features can be obtained by the triangulation of the stereo images at the previous epoch. Therefore, the pixel coordinates of the features can be predicted based on the feature position, perspective center position and camera attitude. The difference between the predicted and actual pixel coordinates can be used to estimate the errors. In this way, the accumulated errors of inertial sensors can be largely reduced by the tightly coupled integration. The integration can reduce the accumulated errors of an individual system. Second, the integration system can be easily expanded with GNSS measurements when GNSS measurements are available. The state vector in the Kalman filter needs to be extended with the receiver clock error and the clock drift to adopt the GNSS code, carrier-phase and Doppler measurements. In this thesis, Precise Point Positioning (PPP) using one receiver is integrated with the Inertial Navigation System (INS) and Stereo VO. Third, fuzzy logic Map Matching (MM) is introduced to be integrated into the system. When there is GNSS outage, the INS/ Stereo VO solution can be projected on the digital map by fuzzy logic MM. The projected point provides the position information for the system, which effectively limits the drift of DR systems. Fourth, when the land vehicle is moving in open sky environment, PPP dominates the system accuracy. A fast PPP ambiguity resolution (AR) method is proposed to fix the ambiguities in the kinematic mode with the aid of MM. Generally, the float PPP solution is projected on the digital map. The projected point and the road link azimuth is applied as additional measurements to accelerate the convergence of ambiguities which reduces the integer search space, resulting in fast ambiguity resolution. Filed test datasets collected by Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) and datasets collected in Calgary are used to verify the effectiveness of the proposed methods.
GNSS, INS, Stereo VO, Tight integration
Fei, L. (2018). Tightly Coupled Integration of GNSS/INS/Stereo Vision/Map Matching System for Land Vehicle Navigation (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/31740