Free Space Detection: A Corner Stone of Automated Driving

Increasingly complex automated driving functions require ever more accurate environment models. If relevant information is missing, automated vehicle control may induce safety risks. In this paper, we describe a joint development of BMW and Continental used to ensure the safety of advanced driver assistance systems. We demonstrate free space detection (FSD) carried out on a stereo vision system and show how it contributes to a number of use cases in high-level driver assistance functionality. One major challenge is the diversity of requirements by various functions and we show how confidence estimation is key to serving them in parallel. Moreover, we demonstrate a novel semi-automatic validation procedure for the generation of ground truth data in order to assess FSD performance in real-world driving scenarios.