Motion estimation in image sequences for traffic applications

This paper describes a digital rear-view mirror that helps the driver to analyze situations on the road behind him. A single camera is mounted opt the back of the car. From its observations the optical motion in the image plane can be estimated. Provided with this information we have to derive the real-world motion of the vehicles behind our car. We present a method to interpret the projected motion of object points on the image plane. A method is described how to retrieve the real-world motion parameters of interest: the time-to-contact and lane-shifts. The estimations of these parameters are largely influenced by (small) vibrations of the camera. Our technique estimates these vibrations on the fly. The complete method will be illustrated by means of simulation experiments. Results based on practical data are presented to indicate its practical validity. The experiments show how the time-to-contact and lane-shifts can be detected with our method in a robust way