Fast Detection of Lane Based on Convolutional Neural Networks and Connected Components Constraints

Lane detection is an important technical basis for the driverless system. The accuracy of lane detection using traditional algorithms could be affected by the complex factors including the unclear lane, the ground traffic signs, and vehicle occlusion. Lane detection based on Deep Learning has the higher accuracy, but its poor real-time performance makes it difficult to apply to in-vehicle computing devices due to the complexity of the deep neural network. Inspired by LaneNet, this paper proposes a new method that simplifies the LaneNet network structure and uses the connected components constraints to classify the lanes. The experimental results show that the detection algorithm can work in real time with 97.43% accuracy on various datasets of highway, and can also effectively overcome the influence of the complex environmental factors in some scenarios.