Authors:
Jurij Kuzmic
and
Günter Rudolph
Affiliation:
Department of Computer Science, TU Dortmund University, Otto-Hahn-Str. 14, Dortmund, Germany
Keyword(s):
Lane Detection, Convolutional Neural Network (ConvNet), Filtered Canny Edge Algorithm, Autonomous Driving, Simulator in Unity 3D, Sim-to-Real Transfer, Training Data Generation, Computational Intelligence.
Abstract:
This paper presents two methods for lane detection in a 2D image. Additionally, we implemented filtered Canny edge detection and convolutional neural network (ConvNet) to compare these for lane detection in a Unity 3D simulator. In the beginning, related work of this paper is discussed. Furthermore, we extended the Canny edge detection algorithm with a filter especially designed for lane detection. Additionally, an optimal configuration of the parameters for the convolutional neural network is found. The network structure of the ConvNet is also shown and explained layer by layer. As well known, a lot of annotated training data for supervised learning of ConvNet is necessary. These annotated training data are generated with the Unity 3D environment. The procedure for generation of annotated training data is also presented in this paper. Additionally, these two developed systems are compared to find a better and faster system for lane detection in a simulator. Through the experiments d
escribed in this paper the comparison of the run time of the algorithms and the run time depending on the image size is presented. Finally, further research and work in this area are discussed.
(More)