A lane tracking system for intelligent vehicle applications

Abstract

An image-based lane tracking system for use in intelligent vehicles is developed. For each frame, the algorithm develops estimates of the geometry and width of the current lane ahead of the vehicle and also the position and orientation of the vehicle with respect to the center-line of the lane. Basic image processing techniques are used to extract a candidate set of lane marker locations from the image. These are used to generate a pool of center-line candidates with properties dependent on the lane markers. A method of elimination based on dynamic programming is used to isolate a final set of center-line candidates that constitute the actual geometry of the road. The road geometry is modeled using a clothoid curve, which stipulates that the curvature of the road varies as a linear function of arc length. The clothoid center-line representation also aids in determining the offset of the vehicle from the center-line and the heading angle of the vehicle with respect to the road. Finally, a Kalman filter is applied to the estimated parameters to preserve smoothness and to predict lane parameters for the next image frame. A set of confidence measures for the estimated data is calculated for use by a higher level data fusion algorithm The system gives an estimate of parameters under normal traffic and driving conditions and runs in real-time on off-the-shelf hardware.