Temporal local route modeling using the recognized lane for autonomous driving comfort

Abstract

Recently, the intelligent Driver Assistant System (DAS) and an autonomous driving system have been widely studied. For those systems, a local route that represents the road shape is essential information for controlling a vehicle's behavior. When the local route is recognized by the perception sensors attached to the vehicle, the discontinuous information caused by the noise and detection failure worsens the driving comfort and stability. Since filtering methods in previous studies have caused time delays, the reaction of the vehicle control may be late when the curvature of the road changes. In this paper, the local route is temporary modeled into a mathematical form with several nodes to smooth the discontinuous information without delay problems. The node location of the temporal roadway geometry model is probabilistically updated by a Bayesian filtering scheme using the recognized local route. The proposed method was evaluated with a Mobileye camera and a real road. This method not only provided road shape information without a time delay but also interpolated the road shape information during the misdetection of sensor information and updating period.