A novel approach to the enhancement of lane estimator by using Kalman filter

Recently vision sensor based lane detection technology is being magnified as automated driving service get into the acceleration step. However, lane detection is not used lateral risk assessment or lateral control alone but also used front target vehicle selection for the longitudinal control system. Ego lane detected from vision sensor is a good assumption as an ego vehicle's driving path and would be useful to front target vehicle decision. In this sense, the importance of curvature and curvature-rate which have been relatively devaluated then lateral offset and heading angle is getting higher. In this study, a novel approach of lane estimator for more robust curvature information is proposed with Kalman filter. The lane estimator is designed with clothoid road model and an assumption based on lane width. To verify the performance of proposed lane estimator, real-car test in the public road is carried out. The test result shows that robustness of the proposed lane estimator even in failure situation of front camera. Furthermore, the result also showed that accuracy and usefulness of lane estimation based target selection. Through the test result analysis, filtered only two coefficients which related clothoid curvature parameters could make lane information more accurately.