恶劣天气下基于锚点的多尺度融合车道线检测及优化方法

Multi-scale fusion lane detection and optimization method based on anchor points in adverse weather conditions

针对黑夜和大雾天气下车道线检测的问题,在数据预处理阶段采用改进的自适应伽马变换对过暗或漂白的图片进行增强,并利用暗通道先验法对大雾场景下的图像进行数据增强,从而降低干扰.在特征提取阶段,采用改进的rotation forest block(RFB)网络提取车道线的特征信息,并通过基于锚点的分类方法实现了快速而准确的车道线检测功能.

The challenges of lane detection in night time and foggy weather scenarios were addressed and researched in this paper.In data preprocessing stage,an enhanced adaptive gamma transform was applied to enhance images that were too dark or overexposed.Additionally,the dark channel prior method was employed to augment image data in foggy conditions,reducing interference from heavy fog and low-light nighttime scenes on lane recognition.For feature extraction,an improved rotation forest block(RFB)network was utilized to capture lane features effectively.Furthermore,a rapid and accurate lane detection effect was achieved through an anchor-based classification approach.