基于图神经网络的手机扫描点云路面坑洞检测方法

张庭瑞; 张学全; 杨子川; 马文硕; 刘兵

doi:10.3963/j.jssn.1674-4861.2025.02.007

基于图神经网络的手机扫描点云路面坑洞检测方法

doi: 10.3963/j.jssn.1674-4861.2025.02.007

张庭瑞^1,,
张学全^{1, 2, ,},
杨子川³,
马文硕¹,
刘兵¹

1.
武汉理工大学交通与物流工程学院武汉 430063
2.
武汉理工大学交通信息与安全教育部工程研究中心武汉 430063
3.
伦敦大学学院英国伦敦 E201EJ

基金项目:

湖北省自然科学基金项目 2025AFD764

详细信息

作者简介:
张庭瑞（2002—），硕士研究生. 研究方向：智能交通. E-mail：312108@whut.edu.cn

通讯作者:
张学全（1989—），博士，讲师. 研究方向：交通地理建模，智能交通等. E-mail：xqzhang@whut.edu.cn

中图分类号: U418.6+2
计量
- 文章访问数: 43
- HTML全文浏览量: 29
- PDF下载量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2024-09-01
- 网络出版日期: 2025-09-29

A Detection Method for Road Surface Pothole Based on Mobile-scanned Point Cloud Using Graph Neural Networks

1.
School of Transportation and Logistics Engineering, Wuhan University of Technology, Wuhan 430063, China
2.
Engineering Research Center of Transportation Information and Safety of Ministry of Education, Wuhan University of Technology, Wuhan 430063, China
3.
University College London, London E201EJ, UK

摘要

摘要: 路面坑洞快速检测与评估对确保道路交通安全至关重要。针对目前基于采集车或无人机的检测方法成本高、部分场景受限，而基于智能手机的检测方法量化精度低的缺陷，研究了基于图神经网络（graph-based attention neural network，GANN）的手机扫描点云路面坑洞提取量化方法。通过搭载激光雷达的智能手机环扫采集路面坑洞点云数据，利用平面拟合和聚类算法对点云数据进行预处理；针对坑洞点云数据中隐含的局部几何特征，在传统图神经网络点云模型的基础上，研究了基于图注意力机制的点云深度学习模型。该模型设计了注意力邻居卷积层，在更大的感知域内通过注意力机制寻找重要节点作为邻居，改善了当前算法动态图构建不佳的缺陷；同时构建了几何特征提取器，通过引入伞曲面准确表示点的局部几何特征，改善了当前算法忽略几何特征的缺陷，对预处理后的点云数据实现高精度分类和量化评估。实验使用智能手机iPhone14 Pro对武汉市武昌区武汉理工大学余家头校区周边路网中的路面坑洞进行扫描测量，构建城市道路路面坑洞点云数据集，进行坑洞检测和评估，结果表明：GANN模型的深度量化误差和体积量化误差分别为4.58%和5.57%，能够准确提取点云数据中的路面坑洞。得益于GANN模型的信息保留和几何特征挖掘，与最新模型PointNeXt和PointMLP的检测结果相比，GANN模型的深度量化误差和体积量化误差分别降低了2.41%和0.11%，对路面坑洞的量化精度更高。
- 交通安全 /
- 路面坑洞检测 /
- 激光点云 /
- GANN模型 /
- 图神经网络
Abstract: Rapid detection and assessment of road surface potholes are essential for traffic safety. To address the high cost and limited applicability of current detection methods based on survey vehicles or drones, as well as the low quantitative accuracy of smartphone-based approaches, this study proposes a novel method for pothole extraction and quantification from smartphone-scanned point clouds using a graph-based attention neural network (GANN). Road surface point cloud data are collected using a LiDAR-equipped smartphone via circular scanning and are preprocessed through planar fitting and clustering algorithms. To effectively capture the local geometric features characteristic of pothole structures, a deep learning model is developed based on graph attention mechanisms, extending traditional graph neural network (GNN) models. The proposed GANN model introduces an Attention Neighbor Convolution Layer, which identifies key neighboring nodes within an expanded receptive field using attention mechanisms, addressing limitations associated with dynamic graph construction present in existing approaches. Additionally, a Geometric Feature Extractor is designed by incorporating an umbrella surface representation to accurately characterize local geometric structures that are often overlooked by prior methodologies. These architectural enhancements enable high-precision classification and quantitative analysis of the preprocessed point cloud data. Experiments were conducted using an iPhone 14 Pro to scan road surface potholes around the Yujiatou Campus of Wuhan University of Technology in Wuchang District, Wuhan, resulting in a real-world urban road pothole point cloud dataset. Results show that the proposed GANN model achieves a depth quantification error of 4.58% and a volume quantification error of 5.57%, demonstrating its effectiveness in extracting potholes from point cloud data. Compared with state-of-the-art models such as PointNeXt and PointMLP, GANN reduces depth and volume quantification errors by 2.41% and 0.11%, respectively, offering superior accuracy in pothole quantification through improved information retention and geometric feature extraction.
- Traffic safety /
- road pothole detection /
- LiDAR point cloud /
- GANN /
- GNN

HTML全文

图 1 路面坑洞检测流程

Figure 1. Road surface pothole detection process

下载: 全尺寸图片幻灯片

图 2 GANN模型结构图

Figure 2. GANN architecture diagram

下载: 全尺寸图片幻灯片

图 3 伞曲率示意图

Figure 3. Umbrella curvature diagram

下载: 全尺寸图片幻灯片

图 4 邻居探寻器示意图

Figure 4. Neighbor finder diagram

下载: 全尺寸图片幻灯片

图 5 邻居探寻器流程图

Figure 5. Neighbor finder flowchart

下载: 全尺寸图片幻灯片

图 6 信息聚合示意图

Figure 6. Information aggregation diagram

下载: 全尺寸图片幻灯片

图 7 数据采集与预处理

Figure 7. Data collection and preprocessing

下载: 全尺寸图片幻灯片

图 8 领域半径选定

Figure 8. Select the suitable epsilon

下载: 全尺寸图片幻灯片

图 9 数据集的典型数据

Figure 9. Typical data of the dataset

下载: 全尺寸图片幻灯片

图 10 模型训练过程

Figure 10. Model training process

下载: 全尺寸图片幻灯片

图 11 不同邻域结果

Figure 11. Results from different neighborhoods

下载: 全尺寸图片幻灯片

图 12 道路坑洞示意

Figure 12. Road pothole diagram

下载: 全尺寸图片幻灯片

表 1 实验对比结果

Table 1. Experimental comparison results

坑洞	实际		PointNet++		DGCNN		PointNeXt		PointMLP		GANN
坑洞	深度/ cm	体积/ $ \mathrm{cm}^{3} $	深度/ cm	体积/ $ \mathrm{cm}^{3} $	深度/ cm	体积/ $ \mathrm{cm}^{3} $	深度/ cm	体积/ $ \mathrm{cm}^{3} $	深度/ cm	体积/ $ \mathrm{cm}^{3} $	深度/ cm	体积/ $ \mathrm{cm}^{3} $
a	1.80	37.11	1.84	36.75	1.67	34.24	1.72	36.68	1.72	36.41	1.77	35.33
b	2.10	20.46	1.96	20.30	2.14	19.39	2.01	19.65	1.97	19.67	2.09	19.72
c	1.40	13.57	1.34	12.97	1.39	12.76	1.42	13.4	1.39	13.42	1.41	13.14
d	1.80	27.44	1.35	25.66	1.84	25.71	1.56	25.92	1.42	26.42	1.74	26.72
e	1.75	253.03	1.61	241.01	1.81	237.34	1.66	244.48	1.69	239.8	1.73	246.81
f	2.15	31.98	2.08	30.31	2.22	30.19	2.06	30.74	2.05	30.62	2.18	30.97

下载: 导出CSV

表 2 对比结果误差

Table 2. Comparison result error

坑洞	深度/cm					体积$ / \mathrm{cm}^{3} $
坑洞	PointNet	DGCNN	PointNeXt	PointMLP	GANN	Pointnet++	DGCNN	PointNeXt	PointMLP	GANN
a	0.1532	0.0199	-0.042 6	-0.046 5	-0.016 3	-0.021 3	-0.051 9	-0.0117	-0.018 9	-0.048 1
b	0.1651	0.0341	-0.045 5	-0.063 2	-0.003 9	-0.051 9	-0.056 1	-0.039 7	-0.038 4	-0.036 1
c	0.1807	-0.005 3	0.0137	-0.009 7	0.0045	-0.013 2	-0.059 9	-0.012 8	-0.010 9	-0.032 0
d	0.1528	0.0352	-0.132 3	-0.213 1	-0.034 5	-0.1015	-0.062 0	-0.055 4	-0.037 2	-0.026 1
e	-0.001 9	-0.070 5	-0.052 3	-0.032 5	-0.010 2	-0.049 9	-0.077 2	-0.033 8	-0.052 3	-0.024 6
f	0.1476	0.0202	-0.039 9	-0.048 5	0.0142	-0.058 4	-0.062 9	-0.038 7	-0.042 6	-0.031 6

下载: 导出CSV

表 3 回归结果

Table 3. Regression results

模型名称	深度误差/%	体积误差/%
PointNet	9.35	9.46
PointNet++	7.87	7.33
DGCNN	6.98	7.45
PointNeXt	6.99	5.68
PointMLP	7.92	5.72
GANN（Ours）	4.58	5.57

下载: 导出CSV

参考文献(28)

[1]	OLIVEIRA H, CORREIA P L. Automatic road crack detection and characterization[J]. IEEE Transactions on Intelligent Transportation Systems, 2012, 14(1): 155-168.
[2]	李海莲, 高雅丽, 江晶晶, 等. 基于NAR-ARIMA组合模型的高速公路沥青路面破损状况预测[J]. 大连理工大学学报, 2024, 64(3): 307-313. LI H L, GAO Y L, JIANG J J, et al. Prediction of expressway asphalt pavement damage based on NAR-ARIMA combined model[J]. Journal of Dalian University of Technology, 2024, 64(3): 307-313. (in Chinese)
[3]	中华人民共和国交通运输部. 公路技术状况评定标准: JTG 5210—2018[S]. 北京: 人民交通出版社, 2018. Ministry of Transport, People's Republic of China. Highway performance assessment standards: JTG 5210—2018[S]. Beijing: China Commnications Press, 2018. (in Chinese)
[4]	KARUKAYIL A, QUAIL C, CHEEIN F A. Deep learning enhanced feature extraction of potholes using vision and lidar data for road maintenance[J]. IEEE Access, 2024, 12: 184541-184549. doi: 10.1109/ACCESS.2024.3512783
[5]	马建, 赵祥模, 贺拴海, 等. 路面检测技术综述[J]. 交通运输工程学报, 2017, 17(5): 121-137. MA J, ZHAO X M, HE S H, et al. Review of pavement detection techonology[J]. Journal of Traffic and Transportation Engineering, 2017, 17(5): 121-137. (in Chinese)
[6]	DOSHI K, YILMAZ Y. Road damage detection using deep ensemble learning[C]. 2020 IEEE International Conference on Big Data, Amsterdam, Netherlands: IEEE, 2020.
[7]	惠冰, 李远见. 基于改进U型神经网络的路面裂缝检测方法[J]. 交通信息与安全, 2023, 41(1): 105-114, 131. doi: 10.3963/j.jssn.1674-4861.2023.01.011 HUI B, LI Y J. A detection method for pavement cracks based on an improved u-shaped network. [J]. Journal of Transport Information and Safety, 2023, 41(1): 105-114, 131. (in Chinese) doi: 10.3963/j.jssn.1674-4861.2023.01.011
[8]	SHIM S, KIM J, LEE S W, et al. Road damage detection using super-resolution and semi-supervised learning with generative adversarial network[J]. Automation in Construction, 2022, 135: 104139. doi: 10.1016/j.autcon.2022.104139
[9]	MAEDA H, KASHIYAMA T, SEKIMOTO Y, et al. Generative adversarial network for road damage detection[J]. Computer-Aided Civil and Infrastructure Engineering, 2021, 36 (1): 47-60. doi: 10.1111/mice.12561
[10]	CHEN M, LIU R, YANG J, et al. Pavement damage identification method based on point cloud multi-source feature enhancement[J]. International Journal of Pavement Research and Technology, 2022, 15(2): 257-268. doi: 10.1007/s42947-021-00116-z
[11]	HUYAN J, LI W, TIGHE S, et al. CrackU-net: a novel deep convolutional neural network for pixelwise pavement crack detection[J]. Structural Control and Health Monitoring, 2020, 27(8): e2551.
[12]	张大伟, 田抑阳, 徐培娟, 等. 基于E-HRNet的路面破损区域识别方法[J]. 北京交通大学学报, 2023, 47(4): 110-119. ZHANG D W, TIAN Y Y, XU P J, et al. Road surface damage area identification method based on E-HRNet[J]. Journal of Beijing Jiaotong University, 2023, 47(4): 110-119. (in Chinese)
[13]	张瑢, 彭勃, 朱宇, 等. 现代沥青路面抗滑检测方法综述[J]. 工程机械, 2024, 55(1): 159-164. ZHANG R, PENG B, ZHU Y, et al. The summary of modern asphalt pavement skid resistance testing methods[J]. Construction Machinery and Equipment, 2024, 55(1): 159-164. (in Chinese)
[14]	LUETZENBURG G, KROON A, BJØRK A A. Evaluation of the Apple iPhone 12 Pro LiDAR for an application in geosciences[J]. Scientific Reports, 2021, 11(1): 1-9. doi: 10.1038/s41598-020-79139-8
[15]	MAEDA H, SEKIMOTO Y, SETO T, et al. Road damage detection and classification using deep neural networks with smartphone images[J]. Computer-Aided Civil and Infrastructure Engineering, 2018, 33(12): 1127-1141. doi: 10.1111/mice.12387
[16]	乌日娜, 白云, 韩建峰. 基于深度学习的公路路面破损检测识别方法[J]. 计算机仿真, 2023, 40(1): 208-212. WU R N, BAI Y, HAN J F. Detection and identification method of road surface damage based on deep learning[J]. Computer Simulation, 2023, 40(1): 208-212. (in Chinese)
[17]	李娇娇, 孙红岩, 董雨, 等. 基于深度学习的3维点云处理综述[J]. 计算机研究与发展, 2022, 59(5): 1160-1179. LI J J, SUN H Y, DONG Yu, et al. Survey of 3-dimensional point cloud processing based on deep learning[J]. Journal of Computer Research and Development, 2022, 59(5): 1160-1179. (in Chinese)
[18]	QI C R, SU H, MO K, et al. Pointnet: deep learning on point sets for 3D classification and segmentation[C]. 2017 IEEE/ CVF Conference on Computer Vision and Pattern Recognition(CVPR), Hawaii, USA: IEEE, 2017.
[19]	CHAITHAVEE S, CHAYAKUL T. Classification of 3D point cloud data from mobile mapping system for detecting road surfaces and potholes using convolution neural networks[J]. International Journal of Geoinformatics, 2022, 18 (6): 11-23.
[20]	CHARLES R, LI Y, HAO S, et al. Pointnet++: deep hierarchical feature learning on point sets in a metric space[C]. Advances in Neural Information Processing Systems, Long Beach, USA: NIPS, 2017.
[21]	QIAN G, LI Y, PENG H, et al. Pointnext: revisiting pointnet++ with improved training and scaling strategies[J]. Advances in Neural Information Processing Systems, 2022, 35: 23192-23204.
[22]	MA X, QIN C, YOU H, et al. Rethinking network design and local geometry in point cloud: a simple residual MLP framework[C]. International Conference on Learning Representations, Virtual: ICLR, 2022.
[23]	WANG Y, SUN Y, LIU Z, et al. Dynamic graph cnn for learning on point clouds[J]. ACM Transactions on Graphics, 2019, 38(5): 1-12.
[24]	PHAN A V, LE NGUYEN M, NGUYEN Y L H, et al. Dgcnn: a convolutional neural network over large-scale labeled graphs[J]. Neural Networks, 2018, 108: 533-543. doi: 10.1016/j.neunet.2018.09.001
[25]	SUN J, ZHANG Q, KAILKHURA B, et al. Modelnet40-c: a robustness benchmark for 3D point cloud recognition under corruption[C]. International Conference on Learning Representations, Virtual: ICLR, 2022.
[26]	ZHANG K, HAO M, WANG J, et al. Linked dynamic graph cnn: learning through point cloud by linking hierarchical features[C]. 27th International Conference on Mechatronics and Machine Vision in Practice(CVPR), Shanghai, China: IEEE, 2021.
[27]	LEI H, AKHTAR N, MIAN A. Spherical kernel for efficient graph convolution on 3D point clouds[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 43(10): 3664-3680.
[28]	WU Z, PAN S, CHEN F, et al. A comprehensive survey on graph neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 32(1): 4-24.