A global and local condensation for lane tracking

Abstract

In this paper, we present a new condensation filtering algorithm that combines local and global information for tracking of lane markings. Local model is used for condensation particle filtering while global one for resampling of local filters and vehicle control command. The detection and tracking process is done based on bird's-eye images in real-time. Unlike most of the state-of-art systems, in our system, with the help of a 3D lidar mounted on the vehicle, the computation of bird's-eye image and the dynamic pitch estimation could be readily and accurately accomplished through joint calibration of camera and 3D lidar. In addition, multiple image features are fused to weight particles. The experimental results on various road environments demonstrate the robustness of our approach.