A three-dimensional (3D) model of asphalt pavement macro-texture is essential for assessing pavement performance. However, the existing methods of 3D macro-texture reconstruction are unstable in various lighting conditions. This paper proposes a method of 3D reconstruction of asphalt pavement macrotexture using an event camera and evolved recurrent convolution network. In this method, an event camera collects macrotexture information on asphalt pavement as an event tensor sequence. The sequence is then enhanced by a character accumulation block and fed into an evolved recurrent convolution network. Finally, the network outputs the 3D reconstruction results of the macrotexture as a depth image. An experiment with 13,000 pavement section samples demonstrates that the proposed method reconstructs 3D macro-texture under various lighting and materials conditions with an absolute relative error of 2.47 % and a threshold accuracy of 93.65 %, which exceeds the other state-of-the-art image- and event-based methods on the task.

中文翻译：

---

使用事件相机和演化的循环卷积网络对沥青路面宏观纹理进行三维重建


沥青路面宏观纹理的三维 （3D） 模型对于评估路面性能至关重要。然而，现有的 3D 宏观纹理重建方法在各种光照条件下不稳定。本文提出了一种使用事件相机和进化的循环卷积网络对沥青路面宏观纹理进行 3D 重建的方法。在这种方法中，事件相机将沥青路面上的宏纹理信息作为事件张量序列收集。然后，该序列通过字符累积块进行增强，并馈送到演化的递归卷积网络中。最后，网络将宏观纹理的 3D 重建结果输出为深度图像。对 13,000 个路面截面样本的实验表明，所提出的方法在各种照明和材料条件下重建 3D 宏观纹理，绝对相对误差为 2.47 %，阈值精度为 93.65 %，这超过了其他最先进的基于图像和事件的方法该任务。