Vision-based motion measurement by directly extracting image features for vehicular steering control

Abstract

An image sensing technique is developed using image plane analysis to measure the vehicular motion variables through an in-vehicle camera. A new measurement model is derived to directly connect the image features of vanishing point and base point with the desired motion variables comprising heading angle, lateral deviation, yaw rate and sideslip angle of the vehicle. These are useful in vehicular steering control applications such as automatic and four-wheel steering systems for providing feedback motion information in a convenient way. In order to test the proposed vision-based measuring scheme, a computerized road scene is simulated as the test sample using a newly proposed model for curvature-based road generation. Finally, experimental works are performed using the real road scene to verify the image sensing method. Consistent results are obtained by comparing with other measurements from a yaw rate gyro and from vehicular traces left on the road.