Research on lane detection algorithm based on multi-scale feature fusion

Lane detection faces the problems of illumination change, complex lane shape, background interference, and noise blur. At the same time, it needs to respond quickly to meet the real-time requirements. Therefore, this paper proposes a lane detection method based on multi-scale feature fusion to improve the robustness and generalization ability of the model. Firstly, the backbone network is improved, and the original baseline only performs simple concatenation of multi-scale features. Instead, the Mult-Scale module is proposed to realize the fusion and interaction of multi-scale features. Secondly, after feature interaction, the FFN module is used to enhance the fusion features. Finally, the multi-scale features are spliced by the CCA module to obtain accurate lane detection results. The experimental results show that the proposed method can effectively improve the accuracy and robustness of lane detection, and adapt to the complex road environment and real-time application requirements.