Research on a Small-Sized Obstacle Detection Method for Railway Intrusion Based on Improved DBSCAN

LI Mengdi; LI Songtao; ZHENG Di; ZOU Xihua

doi:10.3788/gzxb20265507.0714004

Chinese

您当前的位置：

首页 >

文章列表页 >

Research on a Small-Sized Obstacle Detection Method for Railway Intrusion Based on Improved DBSCAN

更新时间：2026-06-12

- Research on a Small-Sized Obstacle Detection Method for Railway Intrusion Based on Improved DBSCAN
- Acta Photonica Sinica Pages: 1-15(2026)
- 作者机构：
  
  1.西南交通大学光电融合集成与通信感知教育部重点实验室，成都 611756
  2.西南交通大学信息科学与技术学院，成都 611756
- 作者简介：
- 基金信息：
  
  Sichuan Science and Technology Program(2024ZHCG0190);National Natural Science Foundation of China(62571456);National Key Research and Development Program of China(2024YFC3015200)
- DOI：10.3788/gzxb20265507.0714004
  CLC： TN958.98
- CSTR：32255.14.gzxb20265507.0714004
- Received：23 March 2026，
  
  Revised：2026-04-27，
  
  Accepted：29 April 2026，
- 稿件说明：
移动端阅览
李梦迪，李松陶，郑狄，等. 基于改进DBSCAN的轨道侵限小尺寸障碍物检测方法研究［J］.光子学报，2026，55（7）：0714004

LI Mengdi， LI Songtao， ZHENG Di， et al. Research on a Small-Sized Obstacle Detection Method for Railway Intrusion Based on Improved DBSCAN［J］. Acta Photonica Sinica， 2026， 55（7）：0714004
李梦迪，李松陶，郑狄，等. 基于改进DBSCAN的轨道侵限小尺寸障碍物检测方法研究［J］.光子学报，2026，55（7）：0714004 DOI： 10.3788/gzxb20265507.0714004. CSTR： 32255.14.gzxb20265507.0714004.

LI Mengdi， LI Songtao， ZHENG Di， et al. Research on a Small-Sized Obstacle Detection Method for Railway Intrusion Based on Improved DBSCAN［J］. Acta Photonica Sinica， 2026， 55（7）：0714004 DOI： 10.3788/gzxb20265507.0714004. CSTR： 32255.14.gzxb20265507.0714004.

摘要

针对轨道侵限检测中背景环境复杂及激光雷达点云远近密度不均，导致小尺寸障碍物检测精度低、漏检率偏高的问题，提出一种轨道侵限小尺寸障碍物检测方法。首先，利用直通滤波初步处理点云，并联合栅格高程差法与随机抽样一致性算法精确划分铁路限界区域。其次，利用轨道背景与障碍物在点云反射强度上的分布差异，提出基于点云强度直方图的自适应强度阈值搜索算法，通过解析点云强度分布，动态锁定轨道背景消亡边界，为后续分割与聚类提供关键自适应强度参数。进而，采用融合反射强度的轨道平面分割算法实现轨道平面与疑似障碍物点的精准分离。最后，提出一种改进的DBSCAN算法，通过引入自适应邻域半径与核心点强度约束，动态适配点云近密远疏分布特性并抑制铁路限界内的非障碍物点云。实验结果表明：相较于传统DBSCAN算法，所提聚类算法正检率提升64%，所提检测方法在35 m范围内对10 cm障碍物检出率达80%以上，70 m范围内对木块、15 cm障碍物检出率分别达80%、95%以上，80 m范围内对20 cm障碍物检出率达97%以上，且150 m处行人检出率达94%。

Abstract

Ensuring railway operational safety is a paramount priority. However， railway perimeters are frequently threatened by unpredictable foreign object intrusions， such as falling rocks and unauthorized pedestrians， which can precipitate severe accidents. While machine vision is cost-effective， it lacks precise depth information and is highly susceptible to adverse weather. Conversely， LiDAR offers exceptional detection accuracy and the capability to acquire precise three-dimensional spatial information. Yet， existing LiDAR research primarily focuses on large targets， leaving small-sized obstacle detection underexplored. Furthermore， railway LiDAR applications face unique challenges： a "dense-near， sparse-far" point cloud distribution and a highly noisy track background comprising ballast and sleepers. These factors make detecting small-sized obstacles distinctively difficult. To address these limitations， this paper proposes a novel LiDAR-based detection framework for small obstacles intruding into the railway clearance， effectively enhancing identification accuracy amidst the intricate backdrops of railway infrastructure.To effectively isolate obstacle es from the complex background and address the challenge of non-uniform point cloud density， this study establishes a systematic detection framework comprising three primary stages： point cloud preprocessing， track plane segmentation， and adaptive obstacle clustering. First， raw data acquired by the Livox Avia LiDAR is filtered via a pass-through filter to constrain the field of view. Subsequently， a grid-based elevation difference method combined with the Random Sample Consensus （RANSAC） algorithm accurately fits rail lines to define the precise railway boundary zone. Crucially， to address point cloud intensity variations caused by fluctuations in ambient background light， an adaptive intensity threshold search based on a point cloud intensity histogram （AITS-PIH） is introduced. By analyzing the histogram， this algorithm dynamically locks the extinction boundary of the track background intensity， providing a robust adaptive threshold for subsequent stages. Second， an track plane segmentation algorithm fusing reflection intensity separates suspected obstacles from the track background. Overcoming traditional RANSAC's reliance on mere geometric information—which often misclassifies small， near-ground obstacles as track bed points—this method utilizes the adaptive threshold to construct a dual-verification criterion. Points exhibiting significant geometric deviation from the fitted plane or possessing a reflection intensity markedly higher than the background threshold are classified as suspected obstacles. Finally， an improved DBSCAN algorithm clusters these suspected points. It optimizes clustering via an adaptive search radius and a core point identification mechanism. Valid core points must simultaneously exceed the adaptive threshold to filter residual non-obstacle point clouds and meet the minimum neighborhood density requirement. Furthermore， the adaptive neighborhood radius dynamically adjusts linearly based on the point’s spatial distance from the LiDAR， ensuring fine-grained segmentation for dense near-field points while preserving connectivity for sparse far-field points.Based on the typical geometric characteristics of the rail cross-section， as well as the standard track gauge and safety clearance requirements， this study first obtains the precise railway boundary zone. Validation under clear and overcast weather conditions demonstrates the AITS-PIH algorithm's ability to dynamically adjust the intensity threshold， ensuring robust discrimination between the track background and obstacles despite changes in ambient illumination. Real-world railway scene tests reveal impressive results for small obstacle detection： the proposed method boosts the positive recognition rate from 34.05% to 98.13% compared to the traditional baseline. Additionally， the detection method achieves excellent detection rates exceeding 80% for both 10 cm obstacles within 35 m and wood blocks within 70 m. For 15 cm and 20 cm obstacles， the detection rates are maintained at 71% and 100% within the 70 m to 80 m range， respectively. The pedestrian detection rate also reaches 94% at 150 m. Comprehensive testing over 5，000 frames yields a low false alarm rate of 4.82% and an average single-frame processing time of 243 ms， further highlighting the method's applicability. These results verify that the proposed method completely avoids severe missed detections and false alarms caused by traditional geometric-only segmentation， significantly improving small obstacle clustering precision.To address the limitations of existing methods in detecting small-sized obstacles intruding into the railway clearance， this paper proposes a novel LiDAR-based detection framework. The primary contributions include the development of the AITS-PIH for dynamic adaptation to ambient background illumination， an intensity-fused track plane segmentation algorithm， and an improved DBSCAN algorithm. The track plane segmentation algorithm leverages intensity differences induced by laser incidence angles to effectively distinguish track background components from true obstacles， thereby significantly reducing the miss rate at the source. Meanwhile， the improved DBSCAN algorithm incorporates an adaptive neighborhood radius model and core point intensity constraints， enabling the precise clustering of obstacles across varying distances and point densities while effectively suppressing non-obstacle point clouds within the railway boundary zone. Experimental results demonstrate that the improved DBSCAN increases the positive recognition rate by approximately 64 percentage points compared to traditional methods. By maintaining high detection rates for multi-scale obstacles at extended ranges and achieving an average processing time of 243 ms， the proposed system satisfies the stringent requirements for high-precision， real-time obstacle detection in railway safety applications.

关键词

Keywords

references

GUI X ， LI Z ， FU X ， et al . Distributed optical fiber sensing and applications based on large-scale fiber bragg grating array： review ［J］. Journal of Lightwave Technology ， 2023 ， 41 （ 13 ）： 4187 - 4200 .

LI Han ， DING Wenchao ， ZHENG Di ， et al . Compact fiber bragg grating inclinometer based on pendulum structure ［J］. Acta Optica Sinica ， 2025 ， 45 （ 3 ）： 0305003 .

李瀚，丁文超，郑狄，等 . 一种基于摆锤结构的紧凑型光纤布拉格光栅倾角传感器［J］. 光学学报， 2025 ， 45 （ 3 ）： 0305003 .

ZHENG D ， YUE Z Z ， ZOU X H ， et al . Multipeak wavelength detection of ultra-short fiber bragg grating array based on arrayed waveguide gratings and convex optimization algorithm ［J］. Journal of Lightwave Technology ， 2025 ， 43 （ 22 ）： 10354 - 10360 .

CAO Z ， QIN Y ， XIE Z ， et al . An effective railway intrusion detection method using dynamic intrusion region and lightweight neural network ［J］. Measurement ， 2022 ， 191 ： 110564 .

YU Houdan ， MI Qiushi ， ZHAO Dong ， et al . Optical fiber perimeter intrusion pattern recognition based on 1D-CNN ［J］. Acta Photonica Sinica ， 2021 ， 50 （ 9 ）： 0906003 .

喻后聃，米秋实，赵栋，等 . 基于一维卷积神经网络的光纤周界入侵模式识别［J］. 光子学报， 2021 ， 50 （ 9 ）： 0906003 . DOI: 10.3788/gzxb20215009.0906003 http://dx.doi.org/10.3788/gzxb20215009.0906003

LIU F ， ZHANG H ， LI X ， et al . Intrusion identification using GMM-HMM for perimeter monitoring based on ultra-weak FBG arrays ［J］. Optics Express ， 2022 ， 30 （ 10 ）： 17307 - 17320 .

RISTIC-DURRANT D ， FRANKE M ， MICHELS K ， et al . A review of vision-based on-board obstacle detection and distance estimation in railways ［J］. Sensors ， 2021 ， 21 （ 10 ）： 3452 .

LIN J ， PENG J . Adaptive inverse perspective mapping transformation method for ballasted railway based on differential edge detection and improved perspective mapping model ［J］. Digital Signal Processing ， 2023 ， 135 ： 103944 .

XU X Y ， WANG S M ， LIU W Q ， et al . Advancements in obstacle intrusion detection methods for rail transit： a comprehensive review ［J］. IEEE Transactions on Instrumentation and Measurement ， 2025 ， 74 ： 1 - 34 .

HELMI W ， BRIDGELALL R ， ASKARZADEH T . Remote sensing and machine learning for safer railways： A review ［J］. Applied Sciences ， 2024 ， 14 （ 9 ）： 3573 .

WANG J ， ZHAI H ， YANG Y ， et al . A review of intrusion detection for railway perimeter using deep learning-based methods ［J］. IEEE Access ， 2024 ， 12 ： 184142 - 184157 .

ZHAO Z ， KANG J ， WU B ， et al . AE-Net： a high accuracy and efficient network for railway obstacle detection based on convolution and transformer ［J］. IEEE Transactions on Instrumentation and Measurement ， 2024 ， 73 ： 1 - 14 .

CHEN C ， QIN H ， QIN Y ， et al . Real-Time railway obstacle detection based on multitask perception learning ［J］. IEEE Transactions on Intelligent Transportation Systems ， 2025 ， 26 （ 5 ）： 7142 - 7155 .

CAO Z ， QIN Y ， JIA L ， et al . Railway intrusion detection based on machine vision： a survey， challenges， and perspectives ［J］. IEEE Transactions on Intelligent Transportation Systems ， 2024 ， 25 （ 7 ）： 6427 - 6448 .

JIANG W ， SONG C ， WANG H ， et al . Obstacle detection by autonomous vehicles： An adaptive neighborhood search radius clustering approach ［J］. Machines ， 2023 ， 11 （ 1 ）： 54 .

ADNAN M ， SLAVIC G ， MARTIN G D ， et al . Systematic and comprehensive review of clustering and multi-target tracking techniques for LiDAR point clouds in autonomous driving applications ［J］. Sensors ， 2023 ， 23 （ 13 ）： 6119 .

ZHANG Changyong ， HAN Liang . Obstacle detection of LiDAR based on optimized DBSCAN ［J］. Laser & Optoelectronics Progress ， 2022 ， 59 （ 12 ）： 1228008 .

张长勇，韩梁 . 基于优化DBSCAN的激光雷达障碍物检测［J］. 激光与光电子学进展， 2022 ， 59 （ 12 ）： 1228008 . DOI: 10.3788/LOP202259.1228008 http://dx.doi.org/10.3788/LOP202259.1228008

NIU Guochen ， WANG Yueyang ， TIAN Yibo . LiDAR obstacle detection based on improved density clustering ［J］. Journal of Beijing University of Aeronautics and Astronautics ， 2023 ， 49 （ 10 ）： 2608 - 2616 .

牛国臣，王月阳，田一博 . 改进密度聚类的激光雷达障碍物检测方法［J］. 北京航空航天大学学报， 2023 ， 49 （ 10 ）： 2608 - 2616 .

QU J ， LI S ， LI Y ， et al . Research on railway obstacle detection method based on developed Euclidean clustering ［J］. Electronics ， 2023 ， 12 （ 5 ）： 1175 .

DAI Honghui ， GENG Chenge ， LIU Dandan . Obstacle detection algorithm of urban rail transit based on lidar ［J］. Journal of Railway Science and Engineering ， 2023 ， 20 （ 6 ）： 2350 - 2360 .

戴鸿辉，耿晨歌，刘丹丹，等 . 基于激光雷达的城轨交通障碍物检测算法研究［J］. 铁道科学与工程学报， 2023 ， 20 （ 06 ）： 2350 - 2360 .

SHEN Tuo ， QIAN Yanzuo ， XIE Lanxin ， et al . Obstacle detection algorithm of fully automatic train considering reflection intensity ［J］. Journal of Tongji University （Natural Science）， 2022 ， 50 （ 1 ）： 6 - 12 .

沈拓，钱沿佐，谢兰欣，等 . 考虑反射强度的全自动运行列车障碍物检测算法研究［J］. 同济大学学报（自然科学版）， 2022 ， 50 （ 01 ）： 6 - 12 .

ZHU G ， NAN Z ， ZHANG X ， et al . High precision rail surface obstacle detection algorithm based on 3D imaging LiDAR ［J］. Optics and Lasers in Engineering ， 2024 ， 178 ： 108206 .

NAN Z ， ZHU G ， ZHANG X ， et al . A novel algorithm for key road sections obstacle detection based on LiDAR ［J］. Robotics and Autonomous Systems ， 2025 ， 194 ： 105136 .

ZHAO Zengxu ， HU Lianqing ， REN Bin ， et al . PointPillars-S 3D object detection algorithm based on Lidar ［J］. Acta Optica Sinica ， 2025 ， 54 （ 6 ）： 0614002 .

赵增旭，胡连庆，任彬，等 . 基于激光雷达的PointPillars-S三维目标检测算法［J］. 光子学报， 2025 ， 54 （ 6 ）： 0614002 .

FISCHLER M A ， BOLLES R C . Random sample consensus： a paradigm for model fitting with applications to image analysis and automated cartography ［J］. Communications of the ACM ， 1981 ， 24 （ 6 ）： 381 - 395 .

XIA Guofang ， HU Chunmei ， CAO Bizheng ， et al . Study on the influence of laser incident angle on the reflection intensity of the point cloud ［J］. Laser Journal ， 2016 ， 37 （ 4 ）： 11 .

夏国芳，胡春梅，曹毕铮，等 . 激光入射角度对点云反射强度的影响研究［J］. 激光杂志， 2016 ， 37 （ 04 ）： 11 - 13 .

TEUFEL S ， VOLK G ， VON BERNUTH A ， et al . Simulating realistic rain， snow， and fog variations for comprehensive performance characterization of lidar perception ［C］// 2022 IEEE 95th Vehicular Technology Conference：（VTC2022-Spring） . IEEE ， 2022 ： 1 - 7 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Machine Learning-based Classification Method for Rayleigh Lidar Data

Detection Performance Verification of Streak Tube Imaging LiDAR Push-broom Scanning Mechanism at Multiple Operating Altitudes

BEV Space 3D Object Detection Algorithm Based on Fusion of Infrared Camera and LiDAR

Nonlinear Frequency Correction Technology of DFB laser with High Frequency Modulation for FMCW LiDAR

Laser Active Fusion Trajectory Measurement Technology in Rocket Take-off Phase

Related Author

CHEN Guxi

ZHONG Kai

WU Tong

ZHANG Xianzhong

LI Xinqi

XU Degang

YAO Jianquan

DONG Chaowei

Related Institution

School of Marine Science and Technology， Tianjin University

Key Laboratory of Optoelectronic Information Science and Technology，Ministry of Education，Tianjin University

School of Precision Instrument and Optoelectronics Engineering，Tianjin University

State Key Laboratory of Laser Spatial Information， Harbin Institute of Technology

Electric Power Research Institute， State Grid Shandong Electric Power Company

AI问答

Postal code：100079
Tel：（010）53879206 Email：tmw@bjxintong.com.cn
Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09082226号-64 京公网安备11010602201714号
It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.

⁰