Multi-Sensor Fusion of Automotive Radar and Onboard Motion Sensors for Seamless Land Vehicle Positioning in Challenging Environments
Positioning services for land vehicle navigation have long relied on global navigation satellite systems (GNSS) including GPS; however, GNSS cannot maintain an accurate position while travelling under bridges, around tall building and under tree canopies due to signal blockage or multipath. Integration with the onboard motion sensors or perception systems can not maintain adequate level of accuracy in all environment during long GNSS outages. The proposed research overcomes the limitations of current positioning technologies to expand their capabilities in degraded environments, providing an ‘uninterrupted everywhere’ positioning. The proposed system integrates a GNSS receiver, onboard motion sensors, and an electronic scanning radar (ESR) presently used in land vehicles for adaptive cruise control. This research develops a new method for ESR-based static object detection using Median Absolute Deviation (MAD) approach to support accurate computation of the vehicle’s forward speed. A radar odometry method is then developed to obtain the vehicle position based on ESR. Integration with onboard motion sensors based on Extended Kalman filtering (EKF) is designed and realized to achieve accurate positioning in degraded vision and challenging GNSS environments. An intelligent switching mechanism is designed to choose between GNSS and ESR to be integrated in different scenarios with the onboard motion sensors in order to maintain reliable and continuous positioning in all enviroments. The proposed method sustained reliable positioning accuracy with errors less than 1% of the traveled distance. During GNSS outages, the fusion of ESR with the onboard motion sensors has resulted in at least 50% improvement in positioning accuracy when compared to the onboard motion sensors operating in standalone mode. This thesis discusses the merits and limitations of the proposed method and gives recommendations for future research.