A road lane detection approach based on reformer model

Intelligent driving has now become the hot research topics in the field of intelligent transportation system (ITS), and its maturity has a significant impact on road traffic safety in information environment. As one of the key technologies of intelligent driving, lane detection is an important prerequisite for identifying driving environment and driving scenarios, and providing auxiliary decision-making for driving. At present, road lane detection methods based on Transformer model are currently considered to be most effective and accurate; however, Transformer model-based road lane detection methods still have their own drawbacks, like high computational complexity of attention mechanisms and defects in activation function and loss functions. Thus, in this paper, a road lane detection method based on Reformer model, which is in essence an improved version of Transformer model has been proposed. By utilizing local sensitive hashing (LSH) attention mechanism, reversible Transformer structure and partitioning mechanism introduced in Reformer model, the high complexity of Transformer model can be overcome; and by configuring the Mish activation function and Huber loss function, the difficulties in network training and parameter optimization in Transformer model can also be solved. Via numerical analysis and real vehicle scenario experiment on Shanghai-Jiaxing expressway in China, the effectiveness of the proposed Reformer model and its superiority over Transformer models has been demonstrated.