Pavement skid resistance measurement is a fundamental component of roadway management and maintenance. Most traditional approaches rely on manual operations or heavy devices, which lead to a labor-intensive, inefficient, and vulnerable testing environment. Precise laser scanning technology lays a solid foundation for effective and continuous pavement friction measurement. This paper proposed an automated pavement friction estimation model using 3D point cloud data and a deep neural network. The fine-grained texture data of over 800 pavement sections with various anti-skidding abilities were collected. The impact of the multi-scale textures on pavement friction was separated and analyzed via two-dimensional wavelet decomposition. A multi-input fusion network with deep aggregation modules was designed to fuse the features of sub-images generated by wavelet decomposition. The results show that the average prediction error is 0.0935, outperforming most state-of-the-art models. The impact of different texture scales on friction estimation is then revealed. The proposed method provides a new tool for effective and large-scale pavement friction evaluation.

中文翻译：

---

使用多尺度纹理和深度融合网络的有效路面抗滑性测量

路面防滑测量是道路管理和维护的基本组成部分。大多数传统方法依赖于手动操作或重型设备，这导致了劳动密集型、低效且易受攻击的测试环境。精确的激光扫描技术为有效和连续的路面摩擦测量奠定了坚实的基础。本文提出了一种使用 3D 点云数据和深度神经网络的自动路面摩擦估计模型。收集了800多个不同防滑能力路面断面的细粒度纹理数据。通过二维小波分解分离和分析多尺度纹理对路面摩擦力的影响。设计了一个具有深度聚合模块的多输入融合网络来融合小波分解生成的子图像的特征。结果表明，平均预测误差为 0.0935，优于大多数最先进的模型。然后揭示了不同纹理尺度对摩擦估计的影响。所提出的方法为有效和大规模的路面摩擦评估提供了一种新工具。