1.Key Laboratory of Big Data & Artificial Intelligence in Transportation, Ministry of Education, Beijing Jiaotong University, Beijing 100044, China
2.Infrastructure Inspection Research Institute, China Academy of Railway Sciences Corporation Limited, Beijing 100081, China
Show less
文章历史+
Received
Published
2023-05-24
2024-11-01
Issue Date
2026-07-13
PDF (3430K)
摘要
对钢轨廓形的快速准确测量是实现高速铁路线路自动化分析的首要前提。在实际应用中,环境异物飞溅和强反射光等噪声会严重污染钢轨图像,导致钢轨追踪失败和测量精度下降。为此,提出一种钢轨图像激光条纹分割与廓形提取相结合的方法。首先,基于连续采集图像的时空上下文信息,定位钢轨感兴趣区域;然后,利用数据密度比缩放的聚类方法,过滤钢轨感兴趣区域中图像噪声并分割钢轨光带;最后,沿光带截面的法线方向实现钢轨廓形提取及测量。选取典型高铁线路试验数据,将该方法与基于密度聚类和共享近邻密度聚类方法的聚类评价指标进行对比,并将它连同灰度重心法和Steger方法的钢轨廓形提取结果与MiniProf钢轨廓形测量仪的实际测量结果进行精度对比分析。结果表明:相比传统聚类和廓形提取方法,该方法平均值为0.98,廓形测量误差均值为0.08 mm,可使不同形状和大小的钢轨数据聚为同一类,且钢轨廓形动态测量精度满足《高速铁路钢轨打磨管理办法》中0.15 mm的要求,有效克服复杂高铁环境噪声,单幅图像处理时间仅为2.2 ms,适用于最高检测速度350 km · h-1下线路自动化分析的时效性和准确性。
Abstract
The rapid and accurate measurement of rail profile is the premise for achieving automated analysis of high-speed railway lines. In practical applications, noise, such as splashes for foreign objects and strong reflected light in the environment can seriously contaminate the rail image, leading to rail tracking failures and decreased measurement accuracy. Therefore, a method combining laser stripe segmentation with profile extraction for rail images is proposed. Firstly, based on the spatiotemporal contextual information of continuously collected images, the Region of Interest (ROI) of the steel rail is located. Then, using a clustering method with data density ratio scaling, image noise in the rail ROI is filtered out and the rail light bands are segmented. Finally, the rail profile along the normal direction of the light strip cross-section is extracted and measured. Using test data from selected typical high-speed railway lines, this paper compares and verifies the clustering evaluation index F1 with Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Shared Nearest Neighbor Density Clustering (SNN) methods. The accuracy comparison analysis of this method, alongside the grayscale centroid method, the Steger method, and MiniProf measurement results, is carried out. The experimental results show that compared with traditional clustering and profile extraction methods, this method achieves an average F1 value of 0.98, and an average profile measurement error of 0.08 mm. It can cluster rail data of different shapes and sizes into the same category, and the dynamic measurement accuracy of rail profile meets the requirement of 0.15 mm stipulated in the "Management Measures for High-Speed Railway Rail Grinding". It can effectively overcome complex high-speed railway environmental noise, with a processing time of only 2.2 ms per image, making it suitable for the timeliness and accuracy requirements of line automation analysis at the highest detection speed of 350 km · h-1.
由于采集图像中钢轨光带会发生大范围移动,持续准确定位钢轨感兴趣区域(Region of Interest,ROI)是钢轨光带分割首要解决的问题。钢轨ROI定位是指从图像中提取轨头区域,再根据轨头与轨腰的几何结构关系,推导出轨腰区域。已有定位方法通常鲁棒性较差或以牺牲时效性为代价,如:均值漂移[4]方法适应能力不强且会发生严重漂移;帧间差分法[5]不适合连续运动状态下局部定位;基于分类器[6]训练方法的累积误差会造成漂移。
传统聚类方法主要有基于密度聚类 [22]、基于顺序密度聚类(Ordering Points To Identify the Clustering Structure,OPTICS)[23]和共享近邻密度聚类[24]等,该类方法可以发现大小不同和形状各异的聚类,但难以识别具有较大密度变化的聚类结果。利用基于密度比缩放聚类的钢轨光带分割方法可以解决这个问题。首先,基于钢轨ROI直方图进行密度估计;然后,根据每个点η邻域的累积分布函数重新缩放钢轨ROI内数据;最后,基于密度聚类方法利用单一阈值实现钢轨光带分割。流程如图6所示。
缩放数据集相比原始数据集分布更均匀,且不同聚类间的距离更大。利用基于密度聚类(Density-Based Spatial Clustering of Applications with Noise,DBSCAN)方法连接相邻区域得到不同的聚类结果。该方法首先将点的密度定义为数据集中位于该点ε邻域内点的数量,如果密度大于其阈值M,则标记为核心点,如果点v位于核心点的ε邻域内,则点v是从直接密度可达的;然后将所有直接密度可达的点连接在一起形成聚类,直到所有核心点都被分配在聚类中。如果1个点既不是核心点,也不是从核心点密度可达的点,则为噪声;最后选取核心点需满足式(11),利用设定的单一阈值得到不同密度聚类的钢轨光带分割结果。
ZHANDong, YULong, XIAOJian, et al. Study on Dynamic Matching Algorithm in Inspection of Full Cross-Section of Rail Profile [J]. Journal of the China Railway Society, 2015, 37 (5): 72-77. in Chinese
ZHANGHaiyang, WANGPing, QIANLihua. Study on Application of Laser Distance-Meter to Measure Rail Crass-Section of High-Speed Turnout [J]. Railway Engineering, 2009, 49 (4): 102-103. in Chinese
[5]
温泽峰,金学松.钢轨波浪形磨损研究[J].中国铁道科学,2007,28(1):136-138.
[6]
WENZefeng, JINXuesong. Study on Rail Corrugation [J]. China Railway Science, 2007, 28 (1): 136-138. in Chinese
[7]
COMANICIUD, RAMESHV, MEERP. Kernel-Based Object Tracking [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2003, 25 (5): 564-577.
[8]
TSAID M, LAIS C. Independent Component Analysis-Based Background Subtraction for Indoor Surveillance [J]. IEEE Transactions on Image Processing, 2009, 18 (1): 158-167.
[9]
BABENKOB, YANGM H, BELONGIES. Robust Object Tracking with Online Multiple Instance Learning [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33 (8): 1619-1632.
[10]
ZHUY, TINGK M, CARMANM J. Density-Ratio Based Clustering for Discovering Clusters with Varying Densities [J]. Pattern Recognition, 2016, 60: 983-997.
[11]
RODRIGUEZA, LAIOA. Clustering by Fast Search and Find of Density Peaks [J]. Science, 2014, 344 (6191): 1492-1496.
[12]
ZHAOL, THORPEC E. Stereo- and Neural Network-Based Pedestrian Detection [J]. IEEE Transactions on Intelligent Transportation Systems, 2000, 1 (3): 148-154.
WANGShengchun, HANQiang, WANGHao, et al. Laser Stripe Center Extraction Method of Rail Profile in Train-Running Environment [J]. Acta Optica Sinica, 2019, 39 (2): 0212004. in Chinese
[15]
STEGERC. An Unbiased Detector of Curvilinear Structures [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1998, 20 (2): 113-125.
WEIZhenzhong, CAOLijun, ZHANGGuangjun. Fast and Sub-Pixel Extraction Method for Structured Light Stripe Center [J]. Journal of Optoelectronics·Laser, 2009, 20 (12): 1631-1634. in Chinese
ZHOUYuan, MENGXiangqun, JIANGDengbiao. Centerline Extraction of Structured Light Stripe under Complex Interference [J]. Chinese Journal of Lasers, 2020, 47 (12): 1204004. in Chinese
[20]
RIDLERT W, CALVARDS. Picture Thresholding Using an Iterative Selection Method [J]. IEEE Transactions on Systems Man and Cybernetics, 1978, 8 (8): 630-632.
LIYingying, ZHANGZhiyi, YUANLin. Survey on Linear Structured Light Stripe Center Extraction [J]. Laser & Optoelectronics Progress, 2013, 50 (10): 100002. in Chinese
[23]
王昊.基于结构光投影的钢轨廓形动态测量方法研究及应用[D].北京:中国铁道科学研究院,2018.
[24]
WANGHao. Research and Application of Dynamic Measuring Method of Rail Profile Based on Structured Light Projection [D]. Beijing: China Academy of Railway Sciences, 2018. in Chinese
SHIHongmei, ZHANGZhipeng, LIFuqiang. Study on Automatic Registration of Rail Profile during Dynamic Measurement of Rail Wear [J]. Journal of the China Railway Society, 2021, 43 (10): 84-90. in Chinese
[27]
ZHANGK H, ZHANGL, LIUQ S, et al. Fast Visual Tracking via Dense Spatio-Temporal Context Learinng [C]// European Conference on Computer Vision. Zurich, Switzerland. Cham: Springer, 2014: 127-141.
DAIPeng, WANGShengchun, DUXinyu, et al. Image Recognition Method for the Fastener Defect of Ballastless Track Based on Semi-Supervised Deep Learning [J]. China Railway Science, 2018, 39 (4): 43-49. in Chinese
CHUJun, ZHUTao, MIAOJun, et al. Target Tracking Based on Occlusion Detection and Spatio-Temporal Context Information [J]. Pattern Recognition and Artificial Intelligence, 2017, 30 (8): 718-727. in Chinese
ZHAOXinxin, WANGShengchun, WANGHao, et al. Fast Tracking Algorithm of Rail Profile under Vehicle Dynamic Conditions [J]. China Railway Science, 2020, 41 (6): 145-155. in Chinese
[34]
ESTERM, KRIEGELH P, SANDERJ, et al. A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise [C]// Proceedings of the 2nd International Conference on Knowledge Discovery and Data Mining (KDD-96). Portland, USA. Washington, DC: AAAI Press,1996: 226-231.
[35]
ANKERSTM, BREUNIGM M, KRIEGELH P,et al. OPTICS: Ordering Points to Identify the Clustering Structure [J]. Proceedings of the ACM SIGMOD International Conference on Management of Data, 1999, 28 (2): 49-60.
[36]
ERTÖZL, STEINBACHM, KUMARV. Finding Clusters of Different Sizes, Shapes, and Densities in Noisy, High Dimensional Data [C]// Proceedings of the 2003 SIAM International Conference on Data Mining. San Francisco, USA. Philadelphia: Society for Industrial and Applied Mathematics, 2003: 47-58.
WEIShibin, LIYing, ZHAOYanfeng, et al. Design and Manufacture of GJ-6 Track Inspection System [J]. Railway Engineering, 2012, 52 (2): 97-100. in Chinese
WANGHao, WANGShengchun, PANDahai, et al. Research and Application on Train-Borne Non-Contact Rail Wear Measurement Method [J]. Railway Engineering, 2017, 57 (12): 97-101. in Chinese
China State Railway Group Co., Ltd. Management Measures for High-Speed Railway Rail Grinding. China Railway Transportation [2014] No. 357. [2014-12-23]. in Chinese