面向轨道安全巡检的全断面表观高速感知方法

丁建隆; 金辉; 李兆新; 程志全; 宋天浩; 徐浩轩

doi:10.3963/j.jssn.1674-4861.2025.03.003

面向轨道安全巡检的全断面表观高速感知方法

doi: 10.3963/j.jssn.1674-4861.2025.03.003

丁建隆^{1, 2,},
金辉²,
李兆新²,
程志全^2, ,,
宋天浩³,
徐浩轩⁴

1.
城市轨道交通系统安全与运维保障国家工程研究中心广州 510330
2.
广州地铁集团有限公司广州 510330
3.
铁科院(北京)工程咨询有限公司北京 100081
4.
武汉大学遥感信息工程学院武汉 430079

基金项目:

国家重点研发计划项目 2023YFB2603702

详细信息

作者简介:
丁建隆(1963—)，博士，正高级工程师. 研究方向：轨道交通安全. E-mail: dingjianlong@gzmtr.com

通讯作者:
程志全(1989—)，本科，工程师. 研究方向：轨道交通安全. E-mail: chengzhiquan@gzmtr.com

中图分类号: U216.3
计量
- 文章访问数: 23
- HTML全文浏览量: 12
- PDF下载量: 3
- 被引次数: 0
出版历程
- 收稿日期: 2024-08-27
- 网络出版日期: 2025-10-11

A Method of Full Section Surface High-speed Sensing for Track Safety Inspection

1.
National Research Center for Rail Transit System Safety and Maintenance, Guangzhou 510330, China
2.
Guangzhou Metro Group Co., Ltd, Guangzhou 510330, China
3.
China Academy of Railway Sciences Engineering Consulting Co., Ltd (Beijing), Beijing 100081, China
4.
School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China

摘要

摘要: 轨道安全巡检要素多，基于单张影像的巡检方法垂直拍摄轨道局部影像，无法快速感知轨道全断面表观细节信息，多视角影像融合受轨检车转弯与晃车影响大，且不同视角内物体的反射强度差异导致影像存在明显亮度偏差，降低病害识别准确度。针对上述问题，研究了基于多线阵影像融合的轨道安全巡检全断面表观感知方法。使用棋盘格标定板建立各相机刚性连接关系，调整横向分辨率和纵向分辨率相等。分析不同类型轨道中的相对不变量，建立钢轨轨面和轨底边缘不变特征，使用改进的YOLO v10算法自动提取钢轨不变特征，利用不变特征计算直线段和弯道段、晃车段的像素偏差，实现影像高精度拼接。根据轨检车前进方向影像的像素统计特征，以亮度差异不大的两侧钢轨为约束，使用改进直方图匹配方法恢复图像整体亮度，实现接缝处均匀过渡。在武汉地铁、武汉局某工务段进行数据采集，轨检车以10、40、60、80、120 km/h的速度运行，使用本文提出基于YOLO v10的影像偏移恢复方法修正图像，查准率95.20%，召回率93.58%，平均精度84.03%，其中召回率较现有最优方法提升了0.51%；对图像关系恢复的平均误差为11.21 px，80 km图像修正合格率为99.43 %；经过亮度调整的整体影像与约束影像像素均值差异为0.682，标准差差异仅0.344，较现有方法分别提升了61.38%和83.38%。实验结果表明：所提面向轨道安全巡检的表观细节感知方法，能够在轨检车以高速运行的条件下，获取位置误差小于4.5 mm、3σ置信度下灰度误差小于1.71的轨道全断面高分辨率影像，极大地提高轨道安全巡检效率。
- 轨道交通 /
- 轨道安全巡检 /
- 线阵相机 /
- 表观病害 /
- 钢轨不变特征
Abstract: Inspection of rail transit safety includes many elements. Methods based on a single image cannot perceive the details of the entire track section quickly. Multi-image stitching is greatly affected by the turning and swinging of the inspection vehicle, and the exposure conditions of different cameras result in significant brightness deviations, which reduces the accuracy of the identification of diseases. A full section surface sensing method for track inspection based on multiple line array images is proposed to address the above issues. A checkerboard calibration board is used to establish a rigid connection relationship between each camera, and the horizontal and vertical resolutions are adjusted to be equal. An invariant feature of tracks is established based on the edges of track surface and bottom to calculate the deviation of straight, curved, and swinging sections. Improved YOLO v10 algorithm is used to automatically extract the above feature. An improved histogram matching method is used to restore the overall brightness and achieve uniform transition at the joint of images based on the statistical characteristics of the forward direction and constrained by the two sides of the tracks with little brightness difference. Data collection is conducted at a certain section of Wuhan Metro and Wuhan Railway Bureau, with inspection vehicle running at speeds of 10, 40, 60, 80, and 120 km/h. The restoration method for image offset based on improved YOLO v10 reaches an average error of 11.21 pixels, with a precision rate of 95.20%, a recall rate of 93.58%, and an average accuracy of 84.03%, in which the recall rate has an increasing of 0.51%, and a pass rate of 99.43 % for 80 km. The mean difference between the overall image and the constrained image after brightness adjustment is 0.682, and the standard deviation is only 0.344, which are improved by 61.38% and 83.38% respectively compared to existing methods. The experimental results show that the proposed surface detail perception method for track safety inspection can obtain high-resolution images of the entire track section with a position error of less than 4.5 mm and a grayscale error of less than 1.71 under high-speed condition, greatly improving the efficiency of track safety inspection.
- rail transit /
- track safety inspection /
- line array camera /
- surface diseases /
- invariant characteristics of track

HTML全文

图 1 轨道巡检线阵影像获取

Figure 1. Process of obtaining line array images

下载: 全尺寸图片幻灯片

图 2 影像纵向被压缩的棋盘格标定板

Figure 2. Checkerboard with longitudinal distortion

下载: 全尺寸图片幻灯片

图 3 不同类型轨道的不变特征示意图

Figure 3. Schematic of invariant features of different types of tracks

下载: 全尺寸图片幻灯片

图 4 改进的bottle neck结构图

Figure 4. Structure of improved bottle neck

下载: 全尺寸图片幻灯片

图 5 车载式轨道巡检系统实验图

Figure 5. Experimental diagram of vehicle mounted track inspection system

下载: 全尺寸图片幻灯片

图 6 影响轨道图像质量的问题

Figure 6. Issues affecting the quality of track images

下载: 全尺寸图片幻灯片

图 7 不同类型钢轨不变特征识别结果

Figure 7. Identification results of invariant features for different types of tracks

下载: 全尺寸图片幻灯片

图 8 不同方法处理后的直方图

Figure 8. Histograms of original image pairs and after different algorithms

下载: 全尺寸图片幻灯片

图 9 拼接影像重点区域

Figure 9. Key areas of stitching images

下载: 全尺寸图片幻灯片

表 1 PA2KGV-80KM线阵相机参数

Table 1. Parameters of PA2KGV-80KM linear array camera

参数名称	参数指标
传感器类型	Global Shutter CMOS
图像模式	黑白
分辨率/px²	2 048×1
光学尺寸/mm	14.3
像素大小/μm²	7×7
最大线速度/kHz	55（80 SAcclTM模式）
像素位宽/bit	8
动态范围/dB	55
数据率（/MB/s）	115

下载: 导出CSV

表 2 基于YOLO v10的偏差修正统计结果

Table 2. Statistical results of correction based on YOLO v10

序号	Δ_M/px	Δ_A/px
Fold1	12	9.76
Fold2	19	10.07
Fold3	17	13.22
Fold4	14	9.53
Fold5	20	11.81
Fold6	22	16.03
Fold7	19	10.39
Fold8	15	10.16
Fold9	18	9.30
Fold10	20	11.87

下载: 导出CSV

表 3 主流目标检测网络对钢轨不变特征的识别结果

Table 3. Results of recognition of invariant features for tracks in mainstream object detection networks

模型	Ω_P/%	Ω_R/%	Ω_mAP/%	Δ_A/px
YOLO v5	93.72	88.46	79.21	11.25
YOLO v6	93.96	89.28	81.05	12.73
YOLO v7	95.07	91.44	82.82	11.94
YOLO v8	94.15	89.96	80.79	11.57
Faster R-CNN	89.24	81.29	77.58	13.82
Cascade R-CNN	91.30	82.77	79.63	12.08
YOLO v10-S	94.57	91.38	83.26	12.69
YOLO v10-M	95.72	93.07	84.15	11.92
Ours	95.20	93.58	84.03	11.21

下载: 导出CSV

表 4 不同拼接方法的统计结果

Table 4. Statistical results of different stitching methods

拼接方法	μ	σ	D
原始图像	55.327	28.038	[27.289, 83.365]
直接拼接	58.280	33.740	[24.540, 92.020]
反距离加权	57.093	31.586	[25.507, 88.679]
Wallis	79.435	30.108	[49.327, 109.543]
本文方法	56.009	27.694	[28.315, 83.703]

下载: 导出CSV

表 5 不同速度下的典型轨道板病害

Table 5. Typical diseases of track board at different speeds

速度/（km/h）	异物	渗水	裂缝
10
40
60
80
120

下载: 导出CSV

表 6 不同速度下的典型扣件和轨面病害

Table 6. Typical diseases of fastener and track at different speeds

速度/（km/h）	扣件病害	轨面波磨	其他病害
10
40
60
80
120

下载: 导出CSV

参考文献(24)

[1]	王广琦, 毛庆洲, 夏梦璇, 等. 基于轨道结构高差特征值的CRTSⅡ型板离缝定量检测方法[J]. 铁道学报, 2024, 46(3): 50-59. WANG G Q, MAO Q Z, XIA M X, et al., Quantitative detection method for CRTS Ⅱ slab gap based on characteristic values of elevation difference of track structure[J]. Journal of the China Railway Society, 2024, 46(3): 50-59. (in Chinese)
[2]	孙旭, 王平. 高速铁路扣件失效对车辆-轨道耦合系统动态响应的影响[J]. 铁道学报, 2022, 44(8): 108-116. SUN X, WANG P. Effect of fastener failure of high-speed railway on dynamic response of vehicle-track coupling system[J]. Journal of the China Railway Society, 2022, 44(8): 108-116. (in Chinese)
[3]	张翕然, 李正中, 张馨, 等. 城市轨道交通系统韧性研究现状及展望[J]. 交通信息与安全, 2024, 42(4): 1-11. doi: 10.3963/j.jssn.1674-4861.2024.04.001 ZHANG X R, LI Z Z, ZHANG X, et al. Status and prospects of studies on urban rail transit resilience[J]. Journal of Transport Information and Safety, 2024, 42(4): 1-11. doi: 10.3963/j.jssn.1674-4861.2024.04.001
[4]	JING G, QIN X, WANG H, et al. Developments, challenges, and perspectives of railway inspection robots[J]. Automation in Construction, 2022, 138: 104242. doi: 10.1016/j.autcon.2022.104242
[5]	张志华, 邓砚学, 张新秀. 基于改进SegNet的沥青路面病害提取与分类方法[J]. 交通信息与安全, 2022, 40(3): 127-135. doi: 10.3963/j.jssn.1674-4861.2022.03.013 ZHANG Z H, DENG Y X, ZHANG X X, A method for detecting and differentiating asphalt pavement distress based on an improved SegNet algorithm[J]. Journal of Transport Information and Safety, 2022, 40(3): 127-135. (in Chinese) doi: 10.3963/j.jssn.1674-4861.2022.03.013
[6]	杜昊天, 陈丰, 李琛, 等. 低能见度下人行横道发光标线视认及预警效果研究[J]. 交通信息与安全, 2024, 42(2): 59-66. doi: 10.3963/j.jssn.1674-4861.2024.02.006 DU H T, CHEN F, LI C, et al. An Evaluation for Impacts of Illuminating Crosswalk Markings on Driving Safety under Low Visibility Conditions[J]. Journal of Transport Information and Safety, 2024, 42(2): 59-66. doi: 10.3963/j.jssn.1674-4861.2024.02.006
[7]	马子骥, 沈伦旺, 蒋志文, 等. 面向复杂轨形的钢轨断面轮廓实时识别方法[J]. 铁道学报, 2023, 45(4): 92-101. MA Z J, SHEN L W, JIANG Z W, et al., Method of real-time recognition of effective rail profiles from complex track structures[J]. Journal of the China Railway Society, 2023, 45(4): 92-101. (in Chinese)
[8]	MAO Q, CUI H, HU Q, et al. A rigorous fastener inspection approach for high-speed railway from structured light sensors[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 143: 249-267. doi: 10.1016/j.isprsjprs.2017.11.007
[9]	ASHISH J, WANG J, YANG X L, et al. Tracknet-a deep learning based fault detection for railway track inspection[C]. 2018 International Conference on Intelligent Rail Transportation(ICIRT), Singapore: IEEE, 2018.
[10]	李杉杉. 基于深度学习的轨道巡检小车轨面检测算法研究[D]. 兰州: 兰州交通大学, 2023. LI S S. Research on the detection algorithm of track surface for track inspection trolley based on deep learning[D]. Lanzhou: Lanzhou Jiaotong University, 2023. (in Chinese)
[11]	LI Y, TRINH H, HAAS N, et al. Rail component detection, optimization, and assessment for automatic rail track inspection[J]. IEEE Transactions on Intelligent Transportation Systems, 2013, 15(2): 760-770.
[12]	GAN J, WANG J, YU H, et al. Online rail surface inspection utilizing spatial consistency and continuity[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2018, 50 (7): 2741-2751.
[13]	YANG F, ZHAO J, YU B, et al. Research on automatic dodging method for road surface line array images[J]. Advances in Civil Engineering, 2022(1): 2447359.
[14]	FENG H, JIANG Z, XIE F, et al. Automatic fastener classification and defect detection in vision-based railway inspection systems[J]. IEEE Transactions on Instrumentation and Measurement, 2013, 63(4): 877-888.
[15]	YANG H, WANG Y, HU J, et al. Deep learning and machine vision-based inspection of rail surface defects[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 71: 1-14.
[16]	REN Y, LU P, AI C, et al. Review of emerging technologies and issues in rail and track inspection for local lines in the United States[J]. Journal of Transportation Engineering, Part A: Systems, 2021, 147(10): 04021062. doi: 10.1061/JTEPBS.0000567
[17]	钱永军, 姜仕军, 臧勐佳, 等. 基于深度学习的轨道车辆线阵相机图像畸变校正方法[J]. 智慧轨道交通, 2021, 58(6): 5-10. QIAN Y J, JIANG S J, ZANG M J, et al. A method for correcting train linear ray camera image distortion based on deep learning[J]. Intelligent Rail Transit, 2021, 58(6): 5-10. (in Chinese)
[18]	李一凡. 基于机器视觉的轨道缺陷智能巡检小车设计研究[D]. 兰州: 兰州交通大学, 2022. LI Y. Research on intelligent inspection trolley for track defects based on machine vision[D]. Lanzhou: Lanzhou Jiaotong University, 2022. (in Chinese)
[19]	程雨, 杜馨瑜, 顾子晨, 等. 基于FPGA和DSP的高速实时轨道巡检图像采集处理系统[J]. 中国铁道科学, 2021, 42(1): 32-42. CHENG Y, DU X Y, GU Z C, et al. High-speed real-time acquisition and processing system of track inspection images based on FPGA and DSP[J]. China Railway Science, 2021, 42 (1): 32-42. (in Chinese)
[20]	房磊, 史泽林, 刘云鹏, 等. 几何联合分段亮度的线阵图像配准[J]. 中国图象图形学报, 2024, 29(1): 80-94. FANG L, SHI Z L, LIU Y P, et al. Joint geometric and piecewise photometric line-scan image registration[J]. Journal of Image and Graphics, 2024, 29(1): 80-94. (in Chinese)
[21]	李奇, 戴宝锐, 杨飞, 等. 轨道平顺性检测方法现状及发展综述[J]. 铁道学报, 2024, 46(7): 101-116. LI Q, DAI B R, YANG F, et al. Overview on existing and developing methods for track irregularity detection[J]. Journal of the China Railway Society, 2024, 46(7): 101-116. (in Chinese)
[22]	余建, 周旺保, 蒋丽忠, 等. 高速铁路桥上设计震致轨道几何不平顺等效方法[J]. 交通运输工程学报, 2024, 24(3): 110-123. YU J, ZHOU W B, JIANG L Z, et al. Equivalent method for designed earthquake-induced track geometric irregularities on high-speed railway bridges[J]. Journal of Traffic and Transportation Engineering, 2024, 24(3): 110-123. (in Chinese)
[23]	陈仁祥, 潘升, 杨黎霞, 等. 基于注意力引导多尺度降噪卷积神经网络的钢轨表面缺陷图像降噪[J]. 铁道学报, 2024, 46 (5): 123-131. CHEN R X, PAN S, YANG L X, et al. Noise reduction of rail surface defect images based on attention-guided poly-scale denoising convolutional neural networks[J]. Journal of the China Railway Society, 2024, 46(5): 123-131. (in Chinese)
[24]	李湘宁, 贾宏志, 张荣福, 等. 工程光学(第三版)[M]. 北京: 科学出版社, 2022. LI X N, JIA H Z, ZHANG R F, et al. Engineering optics[M]. 3rd ed. Beijing: China Science Publishing & Media Ltd., 2022.