Abstract
In recent years, driver assistance systems have become a strong trend in automotive engineering. Such systems increase safety and comfort by supporting the driver in critical or stressful traffic situations. A great variety of surround sensors with different fields of view include radar, ultrasonic, laser, and vision systems. These sensors are based on different technologies and measurement principles. They all have their specific advantages and disadvantages and range from low-cost to high-end systems. They also differ in size, mounting position, maintenance, and weather compatibility. Hence, such sensors are used in various configurations to explore the surroundings ahead, sideways, and behind a vehicle. In addition, vehicle dynamics information from speed, steering angle, yaw rate, and acceleration sensors is available. Data fusion algorithms on raw data, feature, or object levels are used to collect all this information and set up vehicle surround models. An important issue in this context is the question of data accuracy and reliability. Situation interpretation of the traffic scene is based on these surround models. Any situation interpretation has to be performed in real-time, independent of the situation complexity. Typically, the prediction horizon is a couple of seconds. Depending on the results of the driving environment analysis critical situations can be identified. In consequence, the driver can be informed or warned. Some driver assistance systems already perform driving tasks like following, lane changing, or parking autonomously. The art of designing new, valuable driver assistance systems includes many factors and aspects and is still an engineering challenge in automotive research.
TopSurround Sensors
Machine Perception
Surround sensors are among the key innovations that have made driver assistance systems possible at all. Most assistance systems need information not only about the vehicle itself but also about the vehicle environment. Surround sensors are no traditional automotive sensors. They have been developed intensively in the last two decades. Many different physical measurement principles including electromagnetic waves, ultrasonic waves, image perception, and laser have been considered and tested. The general key questions in the context of machine perception are:
At first glance, these questions look quite general and simple. Especially the second question sounds superfluous. Emergency evading systems need reliable information of sufficient free evading space. This issue is addressed by the second question. Although much research on automotive surround sensors has been carried out giving answers to these questions completely and precisely still is an engineering challenge. Assistance systems requirements can be conflicting. Safety systems need reliable and stable data with little preprocessing whereas comfort systems prefer current data that is prepared in an appropriate way. Model assumptions on the dynamics of a detected object can be generated by classifying the object and by tracking the object over several system cycles. Objects in standstill cannot achieve high velocity in the following time cycle.