Flexible lane detection using CNNs
Haixia Li, Li Xizhou
Abstract
To discuss the lane detection have a significant meaning in the autonomous driving research field, particularly, the research direction of path planning. At this stage, to achieve lane line detection, there are two ways, one is based on traditional machine learning methods, and the other is a hot deep learning model in recent years. The method based on traditional machine vision is greatly affected by environmental factors, and the cost is large. In recent years of research, more and more scholars have used deep learning to detect lanes and have achieved good results. The feature of this method is that the labeled data sets are often used to train models. The accuracy of this method is very good, but since the label is fixed, the scene for lane change is difficult to change and is not easy to expand. This paper proposes a method that uses a cascaded convolutional neural network structure for end-to-end lane detection. The whole system adopts the encoder and decoder structure. The lane boundary is instance segmented first, and then the lane classification is classified by the linear classifier.