ABSTRACT
This chapter provides an introduction to Scene Labeling techniques and their application in Advanced Driver Assistance Systems. It explains the concepts of Convolutional Neural Networks (CNN) and Deep Learning. The chapter presents an exemplary CNN. It describes different hardware platforms for Scene Labeling. Scene Labeling is a technique to classify images on different levels of detail. Scene Labeling can also be combined with other detection methods in order to increase reliability and thereby increase the integrity level of safety functions. The bias is another value summed up along with the weighted inputs. Supervised learning for neural networks is performed by measuring the neural net’s estimated output against the expected output with a so called cost function. A successful training converges against the minimum value of the cost function. It is important to choose the cost function suitable for the task that the neural network needs to perform.
