Bruedigam 1990: The paper describes a system for road vehicle guidance that is capable of handling obstacles using information provided by monocular vision only. The general approach is to detect obstacles as an irregularity in the image of the road, to distinguish between real obstacles and false alarms and then to determine the state variables of the obstacle, such as its velocity components relative to the road and its distance to the test vehicle. These data are sent periodically to the longitudinal control process, which is capable of reacting in different ways, depending on the situation. Up to now, the vehicle will stop in front of static objects, follow moving objects in a speed dependent distance or it will behave in a ‘Stop-And-Go’ manner, like in a traffic jam. In the future it will also be able to change lane for collision avoidance. Some results are shown.


1 Introduction

As traffic on our roads is becoming more and more dense, also the work load on the human driver is increasing. For promoting efficient driver support as well as for more safety on our roads, the PROMETHEUS-project was founded.
   Our institute is developing an approach, which processes conventional video signals and which allows the computer to interpret 3D scenes in real time based on a ‘Kalman filter’ approach. In the beginning, work concentrated on road following, where speeds of nearly 100 km/h on well structured roads and about 60 km/h on minor roads without lane markings have been reached [1,2]. For the use of such a system in normal environments, the reaction to obstacles is essential. This includes the detection and analysis of obstacles as well as the dynamics of the vehicle.