A Practical Fusion Localization Method for Autonomous Vehicles in Poor GPS Areas

Most of the existing intelligent vehicle navigation methods depend significantly on an accurate Global Positioning System (GPS). However, when the GPS signal drifts or loses, the navigation task will fail. To get rid of the influence of low precision positioning, we propose a novel fusion localization method for autonomous vehicles in low positioning accuracy areas. Specifically, key points with obvious features in the previous frame of the point cloud are extracted, converted to the next frame of the point cloud through the GPS data, and detect whether these features still exist to determine the GPS state at this time. Moreover, a detachable fusion positioning module based on LIDAR odometry is proposed and used for localization when the GPS signal is drifting or invalid. Furthermore, through real-world experiments, the proposed method was proven effective in the scene of GPS signal drift and failure.