To see and to be seen is an important rule for safety in particular during twilight or night time on remote country roads. "During dusk and dawn, the likelihood of getting involved into an accident with pedestrians, bicyclists or game animals rises exponentially," explained KIT researcher Marko Hoerter who works at the KIT institute for measuring and control technology. While systems are available that make such objects visible on a screen, they are not very practical, finds Hoerter, since they require that the driver actively watches out for them during the ride which is less than optimal from the perspective of safety.
Hoerter has developed an infrared camera-based system which analyzes the image content, identifies any objects in the course of the car and illuminates them shortly with a highly focused LED-based spotlight. This directs the driver's attention to the respective object without requiring that drivers deviates their eyes from the road.
Centerpiece of the system is a camera and a powerful computer to process the video data stream. The challenge was the huge amount of data to be processed in real-time, Hoerter explains. The image processing computer used in the system can tell animals and humans from other objects within 40 milliseconds. In addition, it can transform and determine the real 3D position, direction and speed of these objects from the data contained in the 2D infrared images and the speed at which they change. Based on the results of these calculations, the system also computes the probability of an accident. If there is a significant probability of an impact, a motor-driven apparatus directs the spotlight to the object in question. The system does not generate a continuous light but instead illuminates the object only for a short period in time to avoid bedazzling the person in question.
The approach has been tested during a field trial in a rural environment with 33 participants. The results were encouraging: The