An End-to-End Practical System for Road Marking Detection

摘要

Road marking is a special kind of symbol on the road surface, used to regulate the behavior of traffic participants. According to our survey, it seems that no papers has yet proposed a mature, highly practical method to detect and classify these important fine-grained markings. Deep learning techniques, especially deep neural networks, have proven to be effective in coping with a variety of computer vision tasks. Using deep neural networks to construct road marking detection systems is a practical solution. In this paper, we present an accurate and effective road marking detection system to handle seven common road markings. Our model is based on the R-FCN network framework, with the ResNet-18 model as backbone. SE blocks and data balancing strategies are also used to further improve the accuracy of the detection model. Our model has made a good trade-off between accuracy and speed, and achieved quite good results in our self-built road marking dataset.