Keynote lecture: A camera on wheels - the challenging way to operate vision algorithms in cars
Oliver Sbanski · 2016
Autonomous driving will be the next big thing in automotive industry and in mobility in general. Commodity functions like adaptive cruise control, lane keeping support, road sign recognition, parking and maneuvering, as well as automatic emergency braking and evasion have already been fused to higher level functions with more autonomy to do piloted driving functions. A key sensor to fulfill the growing demands of such functionalities is the video camera. No other sensor has such a broad base of information about the world the car is driving in. Powerful and complex algorithms are needed to gather all this information from the environment and select it as input for the functions. Of course these algorithms have to work as precise and robust as possible. But, the camera is mounted in a car, where it is moving all the time, mounted behind a distorting windshield, exposed to unpredictable scenarios under very different, fast changing illumination and visibility conditions. The presentation describes how a camera is “installed” in a car algorithmically, to compensate most of the named influences. This measures have to be taken to ensure best algorithmic performance of the machine videos algorithms for object detection, etc.. There are challenges to face and problems to solve, but the technology has already proven that it is ready to use and opens a wide range of functionalities - ending up in the dream of autonomous driving.