Vehicle Ego-Localization in Autonomous Lane-Keeping Evasive Maneuvers

This paper presents a method for precise and reliable vehicle ego-localization required for autonomous vehicle guidance in highly dynamic lane-keeping evasive maneuvers. It combines the measurements of a low-cost mono-camera used for lane detection with the information of standard vehicle sensors. The proposed data-fusion concept uses an advanced vehicle model based Extended Kalman Filter that takes into account the nonlinear tire characteristics at the limits of driving physics. Extensions of the fusion concept are derived which compensate the application-related limitations of the camera. The experimental results of a pedestrian collision avoidance maneuver demonstrate the accuracy and robustness of the method.