DEVELOPMENT OF A LANE-KEEPING SYSTEM FOR LANE DEPARTURE AVOIDANCE

The authors are developing a collision avoidance system that will help prevent traffic accidents stemming from human error by generating an alarm or automatically controlling the vehicle, depending on the situation. This paper outlines the MAZDA lane-keeping system, which avoids lane departure when a driver swerves out of lane without his/her intention of lane change. This system has three elements: lane marker recognition, excursion prediction logic, and automatic steering actuator. The quadratic equation of lane marker is estimated with a CCD camera and an image processing unit. The excursion prediction logic calculates emergency situations from vehicle velocity, steering angle, and lane marker information derived from a CCD camera. The automatic steering actuator intervenes in the driver's operation only if either present or future lateral deviation on the estimated ongoing path derived from a CCD camera becomes greater than the given threshold value and if the driver cannot avoid an emergency situation after a warning sound has been generated. In other words, the assumption is that human error is recognized when the estimated lateral deviation is greater than critical. The results of system verification are obtained by using real-time simulator for vehicle control study, driving simulator for human interface study, and field tests.