자율 주행을 위한 강인한 차선 파라미터 추정에 관한 연구

Abstract

In order for a vehicle to drive autonomously, it must be able to perceive complex driving conditions including vehicle position, obstacles, and of special importance, lane marking. In order to control a vehicle along lane markings, lateral offset and the angle between the vehicle direction and lane direction. However, signal noises from road surface and vehicle dynamic motion are often an obstacle to accurate measurement. In this study, we proposed a robust lane parameter estimation algorithm to overcome these noises for vision systems. In order to reduce the influence of the shadows and the cracked road surfaces for the extraction of the lane marking, morphological image processing was used as a preprocessor. From the extracted lane marking data, the lane model, which has a form of second order polynomial equation, was derived to obtain the lane parameters. A Kalman filter that considers the vehicle dynamics was used to reduce the noise as well. A validation gate is a threshold which is associated with the acceptability of the measurement data. In the measurement update of the Kalman filter, the validation gate was used in one of the methods to filter the measurement data when the camera sensor was in temporary failure. To steer the vehicle along the lane marking, we used the lateral controller which considers the kinematic and dynamic characteristics of the vehicle. The proposed lane keeping algorithm was verified via repetitive simulation. The simulation results illustrate the robustness of the lane parameter estimation algorithm under various noise conditions.