Back to ASAM International Conference 2019
Dr. Karsten Schwalbe
FusionSystems GmbH
For autonomous driving, the detection of relevant objects like cars, traffic signs, lane markings or pedestrians is an important task. For this purpose, the camera images need to be interpreted on a pixel level. The information obtained provides a precise and comprehensive understanding of the observed traffic scene, which is required for the complex decision making related to autonomous driving. The mapping of semantic information to each pixel is called semantic segmentation and can be achieved by using convolutional neural networks.
Therefore, in this talk, a brief overview on convolutional neural networks and the corresponding mathematical operations is given. Based on this, the fundamental mechanisms of deep learning as well as some challenges arising from the complexity of traffic scenes are discussed.