Promoting excellence in mobility engineering

  1. FISITA Store
  2. Technical Papers

Image Based Visual Servoing for Lane Keeping Assistance
FISITA2010/F2010E030

Authors

Kracht, Manuel - Institute of Control Theory and Systems Engineering, TU Dortmund, Germany
Malzahn, Jörn - Institute of Control Theory and Systems Engineering, TU Dortmund, Germany
Hoffmann, Frank - Institute of Control Theory and Systems Engineering, TU Dortmund, Germany
Bertram, Torsten - Institute of Control Theory and Systems Engineering, TU Dortmund, Germany

Abstract

State-of-the-art lane keeping systems utilize vision systems mounted behind the windshield to estimate lane parameters such as curvature from the acquired images. Together with sensor readings from gyros the visual information is fused with a vehicle dynamics model to allow the computation and prediction of the vehicle pose in the lane. The vehicle lateral displacement and its relative yaw angle with respect to the lane center line are controlled. This approach, known as position based visual servoing, is sensitive to model uncertainties, signal noise and calibration errors. From robotics it is known that image based visual servoing concepts, which minimize a control error in the image plane rather than performing full pose estimation, are robust against noise and calibration errors. This observation motivates the novel lane keeping control scheme proposed in this paper. A spline-based lane center line detection algorithm, which is robust against noise, shadows, absence of lane markers and illumination variations, is augmented to simultaneously plan the approach trajectory onto the lane center line in the image plane. Therefore, the proposed technique only assumes the lane to have parallel boundaries forming strong edges in the image. The trajectory is warped onto a virtual horizontal ground plane and constitutes a smooth yaw rate command sequence to approach the lane center line. The lane keeping controller compensates both errors in lane coordinates, namely the vehicles lateral displacement, as well as its orientation error by minimizing the angles θi between the ego motion path and the planned trajectory in the virtual road plane. In this way the control scheme accounts for the non-holonomic nature of the plant. The controller is derived without the need of any dynamic model and requires a calibrated camera setup only. The work is outlined as follows: After a brief description of the basic lane detection algorithm the augmented smooth control sequence generation strategy is presented. Next, the design of the proposed control scheme is introduced and lane keeping results are presented for a scenario comprising a clothoid curved road section with initial lateral displacement followed by a straight road section.

KEYWORDS - lane keeping, visual servoing, driver assistance, lane detection, spline

Add to basket