Research On Approach For Vision-Based Lane Departure Warning System

This dissertation has researched on vision-based lane departure warning system. A new lane departure detection approach is proposed, and proved in the HONGQI prototype.A driver-vehicle-road closed loop system is researched. The simulation turned out that the driver's state can be deduced reversely from the road-vehicle relationship.Lane departure warning system is introduced systematically in the dissertation, including performance demand, system component module, and system evaluate standard. A widely used lane departure detection approach TLC is also analyzed. How the TLC curve varies according to different driver states is simulated.A new criterion, angle between lane markings (ALM), is proposed to detect lane departure. The ALM method can work based on the perspective image. The expression of ALM and its relation to TLC is also presented to prove the feasibility of ALM method.The ALM method has been verified in HONGQI Autonomous Land Vehicle. Results under different conditions are introduced, including straight road, curved road, and different driving customs.