提示
基于时序信息融合的车道线检测
首发时间：2025-03-14
摘要：针对连续复杂场景下车道线检测精度不佳的问题，本文提出一种基于时序信息特征融合的残差车道线检测网络，首先，在主干网络引入基于行列方向的双向注意力机制，在不影响实时性的条件下显著提升网络的特征提取能力，然后在不同通道中融合连续帧的特征，强化当前帧的车道线预测能力，以针对遮挡等复杂环境下车道线检测精度不佳的问题。实验结果表明，本文模型在两个公开数据集上均取得了良好效果，其中在具有复杂场景的culane数据集上相较于原来的模型准确率提升了2个百分点。
关键词： 注意力机制
for information in english, please click here
lane detection based on temporal information fusion
abstract：to address the issue of poor lane detection accuracy in continuous complex scenarios, this paper proposes a residual lane detection network based on temporal feature fusion. first, a bidirectional attention mechanism along row and column directions is introduced into the backbone network, significantly enhancing its feature extraction capability without compromising real-time performance. then, features from consecutive frames are fused across different channels to strengthen the lane prediction capability of the current frame, effectively tackling challenges such as occlusion in complex environments. experimental results demonstrate that the proposed model achieves favorable performance on two public datasets, with a 2% accuracy improvement on the culane dataset, which includes complex scenarios, compared to the original model.
keywords： attention mechanism
基金：
论文图表：
引用
导出参考文献
no.****
动态公开评议
共计0人参与
勘误表
基于时序信息融合的车道线检测
评论
全部评论