随着我国经济的快速发展，城市化进程不断加快，同时道路建设水平不断提高，全国公路总里程数不断增加。与此同时，随着不同交通载荷和天气因素对道路的影响，公路路面病害的出现更加普遍并且病害更加严重，从而导致道路养护任务不断加重。集通信、感知与计算于一体的智能交通系统（Intelligent Transportation Systems, ITS）的出现，极大地推进了道路智能化建设，为解决智能道路健康监测任务提供了一个可行的解决思路。其中，路面状况检测是ITS中道路健康监测的重要组成部分，路面损伤检测和路面不平度识别是路面状况检测的重要环节。及时对路面损伤状况开展病害研究能为道路养护决策提供不可或缺的道路状态信息，而路面不平度直接影响车辆行驶的安全性和道路的养护决策，因此，准确对路面损伤进行检测并对路面不平度进行识别可以为交通部门提供重要的数据支撑，对道路的养护决策具有重要意义。本文旨在提出一种依托物联网（Internet of Things, IoT）技术的路面损伤检测方法，并提出多车融合路面不平度识别方法，从而助力ITS中道路健康监测的建设。针对上述内容，本文主要完成的工作如下：

首先，不同于现有研究中基于道路振动信号进行损伤检测时，直接使用原始时域信号或频域信号检测造成运算量大、数据存储和传输消耗IoT节点大量能量的问题，本文开发了一种基于道路振动响应的路面损伤检测方法，从而实现对路面损伤状况的实时检测。具体来说，首先利用路侧部署的物联网传感器内部的振动采集器收集道路振动信号，通过数据预处理算法实现车辆行驶特征的提取，提出了利用状态机的车辆行驶振动波形提取算法，路侧物联网传感器只需把车辆行驶引起的振动信号上传到云服务器。随后，基于车辆引起的振动信号进行特征提取，利用机器学习算法实现了路面损伤检测。最后，通过在不同场景的实际测试，验证了所提出的路面损伤检测算法能实现对横向裂缝、坑洼损伤的检测，并且与现有的检测方法相比，所提出的方法能够实现相同的检测准确率且计算时间最低。

其次，考虑到车辆振动响应采集会叠加传感器背景噪声，以及车辆模型参数的不确定性、车辆异质性的因素，提出了车辆振动响应路面不平度识别方法。具体来讲，首先提出一种结合经验小波变换（Empirical Wavelet Transform, EWT）和编码器-解码器网络架构的单车路面不平度识别算法，该算法利用EWT对车辆振动响应信号进行自适应分解，并通过门控递归单元（Gate Recurrent Unit, GRU）、注意力模型对车辆振动响应进行特征分析，从而实现单车路面不平度识别。在此基础上，为充分利用道路上行驶的多辆车的识别结果，进一步提出多车融合路面不平度识别算法。在考虑对车辆上传的路面不平度识别结果筛选的情况下，利用高斯过程回归（Gaussian Process Regression, GPR）融合多辆车的识别结果。最后，仿真结果表明，路面不平度识别精度随着车辆数目增加而逐渐升高，提出的方法可以有效识别路面不平度，能够为道路养护决策提供实时可靠的路面状况信息。

With the ongoing progress of the economy, the pace of urbanization has been quickening, and the standard of road construction has been consistently advancing. As a result, the overall length of national highways has been steadily expanding. At the same time, with the impact of different traffic loads and weather factors on roads, road pavement diseases are more common and more serious, which leads to increasing road maintenance tasks. The emergence of intelligent transportation systems (ITS), which integrates communication, sensing and computing, has greatly promoted the intelligent construction of roads and provided a feasible solution for solving the task of intelligent road health monitoring. Among them, road surface condition detection is an important part of road health monitoring in ITS, and road surface damage detection and road profile estimation are important links in road surface condition detection. Timely disease research on road surface damage can provide indispensable road state information for road maintenance decisions, and road surface profile directly affects vehicle driving safety and road maintenance decisions. Therefore, accurate detection of road surface damage and road profile estimation can provide important data support for the traffic department, and are of great significance to road maintenance decisions. This thesis aims to propose a road surface damage detection method based on Internet of Things (IoT) technology, and propose a multi-vehicle fusion road profile estimation method, so as to help the construction of road health monitoring in ITS. Based on the content discussed above, the primary tasks accomplished in this thesis are:

First of all, unlike the existing research based on road vibration signals for damage detection, directly using the original time-domain signal or frequency-domain signal detection causes a large amount of computation, and data storage and transmission consume a lot of energy in IoT nodes. This thesis develops a road surface damage detection method based on road vibration response can realize real-time detection of road surface damage. Specifically, firstly, the vibration collector inside the Internet of Things sensor deployed on the roadside is used to collect road vibration signals, and the vehicle driving characteristics are extracted through the data preprocessing algorithm, and a vehicle driving vibration waveform extraction algorithm using a state machine is proposed. The IoT sensor only needs to upload the vibration signal caused by the driving of the vehicle to the cloud server. Subsequently, feature extraction is performed based on the vibration signal caused by the vehicle, and the road surface damage detection is realized by using machine learning algorithm. Finally, through actual tests in different scenarios, it is verified that the proposed road damage detection algorithm can detect transverse cracks and pothole damage, and the suggested method can achieve identical detection accuracy and the shortest computation time when compared to existing detection techniques.

Secondly, considering that the vehicle vibration response collection will superimpose the background noise of the sensor, as well as the uncertainty of the vehicle model parameters and the vehicle heterogeneity, a road profile estimation method for the vehicle vibration response is proposed. Specifically, a single-vehicle road profile estimation algorithm combining Empirical Wavelet Transform (EWT) and encoder-decoder network architecture is proposed first. This algorithm uses EWT to adaptively decompose the vehicle vibration response signal, and the characteristic analysis of the vehicle vibration response is carried out through the Gate Recurrent Unit (GRU) and the attention model, so as to realize the single-vehicle road profile estimation. On this basis, in order to make full use of the estimation results of multiple vehicles driving on the road, a multi-vehicle fusion road profile estimation algorithm is further proposed. Considering the screening of road profile estimation results uploaded by vehicles, Gaussian Process Regression (GPR) is used to fuse the estimation results of multiple vehicles. Finally, the simulation results show that the estimation accuracy of road profiles gradually increases with the increase of the number of vehicles. The proposed method can effectively identify road profiles and provide real-time and reliable road surface condition information for road maintenance decisions.