Dynamic detection and warning device and method for track obstacles

Abstract

The application provides a device and method for dynamic detection and early warning of track obstacles, and relates to the technical field of obstacle detection. The device comprises a data retrieval module for retrieving original surveying and mapping data and line design information; a multi-level risk zoning module for multi-level risk zoning to obtain a plurality of track zones; a monitoring strategy matching module for monitoring strategy matching to output a plurality of reference monitoring strategies, including monitoring reference frequency and monitoring sensor configuration; a sensing monitoring module for dynamic sensing monitoring of obstacle intrusion; an adaptive frequency increase triggering module for triggering adaptive frequency increase; and an early warning instruction generation module for generating directional and hierarchical early warning instructions when real-time obstacles are identified. The application solves the technical problem that the track obstacle detection in the prior art mostly relies on a single or a small number of sensors, and the detection result is difficult to meet the requirements of high precision and high frequency, thereby affecting the safety of rail transit.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of obstacle detection, in particular to a dynamic detection and early warning device and method for track obstacles. BACKGROUND

[0002] As a vital public transportation mode in modern society, rail transit system undertakes a large number of passenger and freight transportation tasks. With the continuous expansion of railway network and the increase of traffic volume, rail safety issues have increasingly become the focus of attention. In railway operation, obstacles on the track often seriously affect the safe operation of trains and may even cause major accidents. Existing track obstacle detection mostly relies on single or a small number of sensors. The detection range and accuracy of these sensors are often insufficient to fully cover the entire track area. Especially in complex terrain and environmental conditions, the detection results are difficult to meet the requirements of high precision and high frequency, and obstacles on the track cannot be timely and accurately detected, which may cause train collision with obstacles and affect the safety of rail transit, even causing serious traffic accidents. SUMMARY

[0003] The present application provides a dynamic detection and early warning device and method for track obstacles, aiming to solve the technical problem that the existing track obstacle detection mostly relies on single or a small number of sensors, and the detection results are difficult to meet the requirements of high precision and high frequency, thereby affecting the safety of rail transit.

[0004] The first aspect of the present application provides a dynamic detection and early warning device for track obstacles, comprising: a data retrieval module for retrieving original surveying and mapping data and line design information of a target track; a multi-level risk zoning module for performing multi-level risk zoning of the target track according to the original surveying and mapping data and line design information, obtaining a plurality of track partitions; a monitoring strategy matching module for performing monitoring strategy matching according to a plurality of fusion feature vectors of the plurality of track partitions, outputting a plurality of baseline monitoring strategies, wherein the baseline monitoring strategy includes monitoring baseline frequency and monitoring sensor configuration; a sensor monitoring module for performing dynamic sensor monitoring of obstacle intrusion in the plurality of track partitions according to a plurality of monitoring sensor configurations after a plurality of integrated sensor array layouts of the plurality of track partitions; an adaptive frequency increasing trigger module for triggering adaptive frequency increasing of the plurality of monitoring baseline frequencies according to real-time train positioning of a real-time track train and spatial topological relations of the plurality of track partitions; and an early warning instruction generation module for generating directional hierarchical early warning instructions according to real-time train positioning and spatial topological relations of the Kth track partition when a real-time obstacle is identified in the Kth track partition.

[0005] The second aspect of this application discloses a method for dynamic detection and early warning of track obstacles. This method is implemented using the aforementioned dynamic detection and early warning device for track obstacles. The method includes: retrieving original mapping data and route design information of the target track; performing multi-level risk zoning of the target track based on the original mapping data and route design information to obtain multiple track zones; performing monitoring strategy matching based on multiple fused feature vectors of the multiple track zones to output multiple benchmark monitoring strategies, wherein the benchmark monitoring strategies include monitoring benchmark frequencies and monitoring sensor configurations; deploying multiple integrated sensor arrays of the multiple track zones according to the multiple monitoring sensor configurations, and performing dynamic sensing monitoring of obstacle intrusion in the multiple track zones based on the multiple monitoring benchmark frequencies; triggering adaptive frequency upsampling of the multiple monitoring benchmark frequencies based on the real-time train positioning of the real-time track train and the spatial topology relationship of the multiple track zones; and generating a directional, hierarchical early warning command based on the real-time train positioning and the spatial topology relationship of the Kth track zone when a real-time obstacle is identified in the Kth track zone.

[0006] One or more technical solutions provided in this application have at least the following beneficial effects:

[0007] By retrieving the original survey data and route design information of the target track, precise track geometry and design parameters can be obtained, providing accurate foundational data for subsequent risk analysis and monitoring strategy development. Based on the original survey data and route design information, the track is divided into multiple risk zones. This hierarchical zone management allows for customized monitoring strategies to be developed according to the characteristics of different areas, achieving more efficient and accurate risk warnings. By matching monitoring strategies to the fused feature vectors of multiple track zones, a baseline monitoring strategy that meets actual needs can be automatically generated. This strategy matching can set corresponding monitoring frequencies and sensor configurations based on the specific characteristics of each track zone, thereby achieving personalized monitoring and warnings and improving the system's flexibility and accuracy. Based on the configuration of multiple monitoring sensors, integrated sensor arrays are deployed in each track zone to comprehensively monitor obstacles within the track area. The system dynamically detects obstruction intrusions on the track based on a baseline monitoring frequency. This dynamic monitoring provides real-time feedback and adjusts the monitoring frequency according to the needs of different risk areas, thereby improving monitoring efficiency and response speed. By combining real-time train positioning with the spatial topology of track zones, the system triggers adaptive frequency upsampling of the baseline monitoring frequency. This allows for dynamic adjustment of the monitoring frequency based on train location and the risk status of the track area. This frequency upsampling mechanism ensures efficient monitoring while avoiding unnecessary monitoring burdens, improving system responsiveness and resource utilization efficiency. When an obstruction is identified within any track zone, a directional, graded early warning command is generated based on the real-time train positioning and the spatial topology of the obstruction. This graded early warning adjusts the warning level according to the severity of the threat, improving the accuracy of the warning and the timeliness of the response, thus ensuring effective protection of rail transit safety.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the structure of a dynamic detection and early warning device for track obstacles provided in an embodiment of this application.

[0010] Figure 2 This is a schematic flowchart of a dynamic detection and early warning method for track obstacles provided in an embodiment of this application.

[0011] Figure labeling: Data retrieval module 10, multi-level risk zoning module 20, monitoring strategy matching module 30, sensor monitoring module 40, adaptive frequency increase triggering module 50, early warning instruction generation module 60. Detailed Implementation

[0012] This application provides a dynamic detection and early warning device and method for track obstacles, which solves the technical problem that the detection of track obstacles in the prior art mostly relies on a single or a small number of sensors, and the detection results are difficult to meet the requirements of high precision and high frequency, thus affecting the safety of rail transit.

[0013] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0014] Example 1, as Figure 1 As shown in the figure, this application embodiment provides a dynamic detection and early warning device for track obstacles, the device comprising:

[0015] The data retrieval module 10 is used to retrieve the original surveying data and route design information of the target track; the multi-level risk zoning module 20 is used to perform multi-level risk zoning of the target track based on the original surveying data and route design information to obtain multiple track zones; the monitoring strategy matching module 30 is used to perform monitoring strategy matching based on multiple fused feature vectors of the multiple track zones and output multiple benchmark monitoring strategies, wherein the benchmark monitoring strategies include benchmark monitoring frequencies and monitoring sensor configurations; the sensing monitoring module 40 is used to deploy multiple integrated sensor arrays of the multiple track zones based on the multiple monitoring sensor configurations and then perform dynamic sensing monitoring of obstacle intrusions in the multiple track zones based on multiple benchmark monitoring frequencies; the adaptive frequency amplification triggering module 50 is used to trigger adaptive frequency amplification of the multiple benchmark monitoring frequencies based on the real-time train positioning of the real-time track train and the spatial topology relationship of the multiple track zones; and the early warning command generation module 60 is used to generate directional and hierarchical early warning commands based on the real-time train positioning and the spatial topology relationship of the Kth track zone when the Kth track zone identifies a real-time obstacle.

[0016] Furthermore, the adaptive upsampling trigger module 50 includes:

[0017] The system comprises the following components: a partition positioning unit for locating multiple partition boundary beacons in the multiple track partitions based on the real-time track train's operating line topology; a dynamic calculation unit for dynamically calculating multiple spatial topology parameters based on the real-time train positioning and the multiple partition boundary beacons, wherein the spatial topology parameters include dynamic Euclidean distance, relative azimuth angle, and remaining arrival time; a frequency upscaling partition positioning unit for verifying whether the multiple spatial topology parameters meet the frequency upscaling conditions based on multi-dimensional trigger thresholds, and selecting and locating M frequency upscaling partitions; a template matching unit for performing frequency upscaling strategy template matching based on the M spatial topology parameters and M partition risk levels of the M frequency upscaling partitions to obtain M real-time frequency upscaling strategies; and an adaptive frequency upscaling unit for adaptively upscaling the M monitoring reference frequencies of the M frequency upscaling partitions based on the M real-time frequency upscaling strategies, wherein the frequency upscaling process dynamically executes load balancing scheduling based on a sliding monitoring window.

[0018] Furthermore, the multi-level risk zoning module 20 includes:

[0019] The model building unit is used to build a track BIM model based on the original surveying data and route design information; the model segmentation unit is used to segment the track BIM model based on a preset grid scale to obtain multiple grid-level track models; the vector extraction unit is used to extract multiple terrain feature vectors from the multiple grid-level track models; and the clustering and merging unit is used to perform track grid clustering and merging of the target track based on the multiple terrain feature vectors to obtain the multiple track partitions.

[0020] Furthermore, the multi-level risk zoning module 20 includes:

[0021] The weight configuration unit is used to configure terrain feature risk weights according to track safety requirements, and perform weighted normalization processing on the multiple terrain feature vectors according to the terrain feature risk weight configuration to obtain multiple standardized feature spaces; the homogeneity verification unit is used to traverse the multiple standardized feature spaces based on multi-feature risk thresholds, perform risk homogeneity verification of adjacent grids, and output a risk homogeneity identifier array; the similarity calculation unit is used to calculate the terrain similarity of adjacent grids based on the multiple standardized feature spaces, and output a terrain similarity quantification array; the topology clustering unit is used to superimpose the risk homogeneity identifier array and the terrain similarity quantification array, perform spatial continuity topology clustering, and output the multiple track partitions.

[0022] Furthermore, the monitoring strategy matching module 30 includes:

[0023] The feature vector extraction unit is used to extract a first set of terrain feature vectors based on the grid structure of the first track partition; the dimensionality reduction unit is used to perform PCA dimensionality reduction on the first set of terrain feature vectors based on the variance contribution rate constraint to obtain a first principal component feature subset; the risk level mapping unit is used to input the first principal component feature subset into the fuzzy rule decision engine, perform risk level mapping according to the partition type of the first track partition, and output the risk level of the first partition; the policy matching unit is used to match the first benchmark monitoring policy of the risk level of the first partition with the dynamic policy mapping matrix.

[0024] Furthermore, the monitoring strategy matching module 30 includes:

[0025] A dynamic feature weighting unit is used to perform dynamic feature weighting on the first principal component feature subset based on the partition type of the first track partition, generating a first weighted feature vector; a function transformation unit is used to perform fuzzy membership function transformation on the first weighted feature vector to construct a first fuzzy input vector; an inference calculation unit is used to load the first fuzzy input vector into a pre-built hierarchical fuzzy rule base, perform cross-level risk inference calculation, and output a first fuzzy risk value, wherein the hierarchical structure design of the hierarchical fuzzy rule base includes basic risk factor rules, composite risk coupling rules, and special scenario correction rules; and a defuzzification processing unit is used to defuzzify the first fuzzy risk value and output the risk level of the first partition.

[0026] Furthermore, the terrain feature vector consists of elevation gradient, surface curvature, vegetation cover density, water body distribution distance, and proximity of artificial structures.

[0027] Furthermore, the sensing and monitoring module 40 includes:

[0028] A node deployment unit is used to pre-deploy multiple edge obstacle intrusion detection nodes in the multiple track partitions, wherein the multiple integrated sensor arrays and the multiple edge obstacle intrusion detection nodes are connected by a low-latency communication link; a sensor monitoring unit is used to send time-series sensor data streams to the multiple edge obstacle intrusion detection nodes in a polling manner, constrained by the multiple monitoring reference frequencies, to perform dynamic sensor monitoring of obstacle intrusions in the multiple track partitions.

[0029] Furthermore, the early warning instruction generation module 60 includes:

[0030] The system includes a coordinate extraction unit for extracting multi-dimensional features and spatial coordinates of the real-time obstacle based on multi-modal sensing data; a scoring and quantization unit for weighted normalization and fusion of the multi-dimensional features of the obstacle to quantify and output a real-time threat score, wherein the multi-dimensional features of the obstacle include geometric dimensions, material type, and motion vector features; a collision time calculation unit for interactively obtaining the train speed of the real-time track train and calculating the collision time of the track geometric constraints based on the train speed and the spatial coordinates of the obstacle; and a graded warning matching unit for performing graded warning matching based on the coupling relationship between the real-time threat score and the collision time, and outputting the directional graded warning command.

[0031] Through the detailed description of the dynamic detection and early warning method for track obstacles that follows, those skilled in the art will clearly understand the dynamic detection and early warning device for track obstacles in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section.

[0032] Example 2, based on the same inventive concept as the dynamic detection and early warning device for track obstacles in the foregoing examples, such as... Figure 2 As shown in the figure, this application provides a dynamic detection and early warning method for track obstacles, the method comprising:

[0033] Retrieve the original survey data and route design information of the target track.

[0034] The original surveying data consists of topographical data acquired through technologies such as laser scanning, drone aerial photography, and satellite imagery. This data primarily includes information on the track's geographical location, ground elevation, curves, and gradients. The route design information includes track design parameters, such as track geometry (curves, gradients, straight sections, etc.), track laying specifications (such as gauge and track type), and other relevant design elements, such as the signaling system and environmental factors affecting the track's surroundings. This information provides essential data support for subsequent risk assessment and monitoring.

[0035] Based on the original survey data and route design information, the target track is divided into multiple risk zones, resulting in multiple track zones.

[0036] Using original survey data and route design information, a BIM (Building Information Model) of the track is constructed. This model includes information such as track structure, surrounding environment, and facility layout. By rasterizing the BIM model, the track space is divided into multiple grids, each representing a small area of ​​the track. Topographic features are extracted from the rasterized model, and each grid has a topographic feature vector. These vectors are the basis for subsequent risk assessment. Based on the extracted topographic feature vectors, the track grids are clustered and merged, and grids with similar topographic features are divided into a track partition. Since different topographic features have different impacts on track safety, risk weights are assigned to each topographic feature according to the track's safety requirements. Through weighted normalization, a standardized feature space is obtained. Through these steps, the track is finally divided into multiple track partitions with different risk levels.

[0037] Based on multiple fused feature vectors of the multiple orbital partitions, a monitoring strategy is matched to output multiple benchmark monitoring strategies, wherein the benchmark monitoring strategy includes a benchmark monitoring frequency and a monitoring sensor configuration.

[0038] Each track partition has multiple feature vectors. These feature vectors are fused to obtain a complete and comprehensive fused feature vector. Based on the fused feature vector of each track partition, an appropriate monitoring strategy is matched. Each track partition is configured with different monitoring baseline frequencies and monitoring sensor configurations. For example, the monitoring frequency is higher for high-risk areas and lower for low-risk areas. Through strategy matching, multiple baseline monitoring strategies are finally obtained. These baseline monitoring strategies indicate how to monitor the track partitions through the sensor network, thereby tracking the track safety status in real time.

[0039] After deploying multiple integrated sensor arrays for the multiple track zones based on multiple monitoring sensor configurations, dynamic sensing and monitoring of obstacle intrusion in the multiple track zones are performed based on multiple monitoring reference frequencies.

[0040] Based on the determined monitoring sensor configuration, an integrated sensor array is deployed in each track section. Sensors may include radar, cameras, infrared sensors, acoustic sensors, etc. An integrated sensor array refers to multiple sensors working together in the same area, collecting multi-dimensional monitoring data using different sensor types. The deployed integrated sensor array then begins continuous dynamic monitoring of the corresponding track section according to the corresponding monitoring baseline frequency. Dynamic monitoring means that the sensors not only collect data at specific times but can also adjust to changes in real-time conditions. Through real-time feedback, the sensors can adapt to environmental changes and dynamically respond to potential obstacle intrusion events.

[0041] Based on the real-time train positioning of the real-time rail train and the spatial topology relationship of the multiple track zones, the adaptive upsampling of the multiple monitoring reference frequencies is triggered.

[0042] Real-time train positioning is typically achieved through GPS, track location calibration, and onboard positioning systems to obtain the train's accurate location on the track. There is a clear spatial topological relationship between the train and multiple track sections, namely, geographic space and location relationship. Based on the train's current position and the topological structure of the track sections, it is determined whether the train is approaching a high-risk area. When the train enters a high-risk area, the monitoring system increases the monitoring baseline frequency of the corresponding area to more accurately detect potential obstacles. Increasing the frequency not only improves monitoring accuracy but also ensures that sensors can detect any potential obstacles in real time when the train passes through a dangerous area.

[0043] When the Kth track zone identifies a real-time obstacle, a directional and hierarchical early warning command is generated based on the real-time train positioning and the spatial topology relationship of the Kth track zone.

[0044] The Kth track zone is any one of multiple track zones, where K is a positive integer. When the Kth track zone identifies a real-time obstacle, it locates and classifies the obstacle, obtaining its type, size, and position. The train's real-time position is then acquired, and combined with the spatial topology of the track zone, information such as the spatial distance and relative speed between the train and the obstacle is calculated to generate directional, graded early warning commands. For example, a low-level warning is issued when the obstacle is far from the train and poses a minor threat; a high-level warning is issued if the obstacle is about to collide with the train. The generated directional, graded early warning information is promptly transmitted to relevant personnel or systems so that timely action can be taken, such as slowing down, stopping operations, or activating other emergency plans.

[0045] Furthermore, based on the real-time train positioning of the real-time rail train and the spatial topology relationship of the multiple track zones, the method triggers adaptive frequency upsampling of the multiple monitoring reference frequencies, including:

[0046] Based on the real-time track train's operating route topology, multiple zone boundary positioning beacons are located in the multiple track zones. Multiple spatial topology parameters are dynamically calculated based on the real-time train positioning and the multiple zone boundary positioning beacons, including dynamic Euclidean distance, relative azimuth angle, and remaining arrival time. The multiple spatial topology parameters are verified based on multi-dimensional trigger thresholds to determine if they meet the frequency upscaling conditions, and M frequency upscaling zones are selected and located. Frequency upscaling strategy templates are matched based on the M spatial topology parameters and M zone risk levels of the M frequency upscaling zones to obtain M real-time frequency upscaling strategies. Adaptive frequency upscaling is performed on the M monitoring reference frequencies of the M frequency upscaling zones according to the M real-time frequency upscaling strategies, wherein the frequency upscaling process dynamically executes load balancing scheduling based on a sliding monitoring window.

[0047] The track topology refers to the track structure on which trains travel, including the shape of the track, the location of each section, turning angles, gradients, and other geographical and geometric information. Each track section boundary positioning beacon is a fixed positioning point used to accurately identify the boundary of the track section. These boundary positioning beacons are located using high-precision positioning technologies (such as GNSS, lidar, ground base stations, etc.) to determine the train's position within the track section in real time, as well as the distance between the train and these boundaries.

[0048] Based on multiple zone boundary beacons, the relationship between real-time train positioning and each track zone is calculated. The dynamic Euclidean distance, the straight-line distance between two points, is used here to determine the spatial relationship between the train and the track zone by calculating the Euclidean distance between the train's position and the track zone boundary beacon. This parameter indicates the train's distance from each boundary, providing fundamental data for frequency-upgraded monitoring and early warning. The relative azimuth angle refers to the directional angle from the train's position to the track zone boundary beacon, usually measured from north or a reference direction. By calculating the relative azimuth angle, the train's direction relative to the track zone can be determined, thus indicating the train's travel direction and whether it is approaching a specific track zone. The remaining arrival time refers to the estimated time for the train to reach a certain zone boundary beacon. This time is calculated based on the train's current speed and distance to the beacon. This parameter helps predict in advance when a train might enter a specific track zone, thereby adjusting monitoring frequency or triggering early warnings.

[0049] Multi-dimensional trigger thresholds refer to standards set based on multiple different spatial topology parameters (including dynamic Euclidean distance, relative azimuth, and remaining time of arrival). Each parameter has a threshold, and frequency upscaling is triggered when the orbital data exceeds this threshold. By calculating the spatial topology parameters in real time, it is verified whether the current orbital conditions meet the frequency upscaling requirements. Through the verification of the multi-dimensional trigger thresholds, M frequency upscaling partitions that meet the frequency upscaling conditions are selected, where M is a positive integer.

[0050] Each upsampling zone has corresponding spatial topology parameters and a zone risk level. The zone risk level is assessed based on the zone's geographical characteristics and train operation conditions; for example, areas with higher risk levels include curves, steep slopes, and tunnels. Upsampling strategy template matching is performed. The template includes detailed requirements for monitoring frequencies and sensor configurations under different zone risk levels. The matching process is adjusted according to the specific circumstances of each zone (e.g., whether it is in a high-risk area, whether it is near obstacles, etc.). Based on the template matching results, a real-time upsampling strategy is generated for each upsampling zone.

[0051] Adaptive frequency upscaling refers to the monitoring system automatically adjusting the monitoring frequency based on real-time operating status. For each upscaling zone, its baseline monitoring frequency is increased to a higher level according to a real-time upscaling strategy. A sliding monitoring window refers to a dynamic time window within which the monitoring frequency is automatically adjusted. This sliding monitoring window slides with the real-time position of the train; therefore, the monitoring frequency adjustment is dynamically updated based on the train passing through different areas. For example, when the train approaches a high-risk area, the sliding monitoring window shrinks rapidly, increasing the monitoring frequency; conversely, when the train leaves the area, the window expands, decreasing the frequency. Load balancing scheduling refers to ensuring that the load on each monitoring zone is not overly concentrated during the upscaling process, based on a real-time monitoring load distribution strategy.

[0052] Furthermore, based on the original surveying data and route design information, multi-level risk zoning of the target track is performed to obtain multiple track zones. The method includes:

[0053] A track BIM model is constructed based on the original surveying data and route design information; the track BIM model is segmented based on a preset grid scale to obtain multiple grid-level track models; multiple terrain feature vectors are extracted from the multiple grid-level track models; the track grids of the target track are clustered and merged based on the multiple terrain feature vectors to obtain the multiple track partitions.

[0054] Based on the original survey data and route design information, a digital track BIM model is constructed. The track BIM model not only includes the geometry of the track (such as track route, curves, gradients, etc.), but also integrates information such as track materials, structure, and facilities, representing the specific structural layout of the track, serving as the basis for subsequent analysis and monitoring.

[0055] The preset grid scale refers to the dividing standard for the track area into multiple small grids. This dividing standard can be adjusted according to different needs, usually taking into account the complexity of the area and the required monitoring accuracy. For example, if a section of the track has a complex curve or gradient, a smaller grid scale is used, while a larger grid scale is selected for straight sections or simple terrain. The track BIM model is divided according to the set grid scale to obtain multiple grid-level track models. Each grid-level track model represents a region of the track and usually includes information such as geographic coordinates, spatial location, and terrain.

[0056] Each grid-level orbital model represents a region of the orbit. The corresponding terrain feature vector is extracted from each grid-level orbital model. The terrain feature vector consists of elevation gradient, surface curvature, vegetation cover density, water body distribution distance, and proximity of artificial structures. These features are used to describe the terrain and environmental conditions of the region.

[0057] By clustering analysis of terrain feature vectors, grid regions with similar features are merged into a single track partition. Each track partition represents a track region with similar terrain and environmental features. For example, if multiple grids have similar features such as elevation gradient and vegetation cover, these grids are grouped together to form a single track partition. This allows the entire track to be divided into multiple different levels of regions based on terrain features, enabling more accurate monitoring.

[0058] Furthermore, the method involves clustering and merging the orbital grids of the target orbit based on the multiple terrain feature vectors to obtain the multiple orbital partitions, and includes:

[0059] Based on track safety requirements, terrain feature risk weights are configured, and weighted normalization is performed on the multiple terrain feature vectors according to the terrain feature risk weight configuration to obtain multiple standardized feature spaces. The multiple standardized feature spaces are traversed based on multi-feature risk thresholds to verify the risk homogeneity of adjacent grids, and a risk homogeneity identifier array is output. The terrain similarity of adjacent grids is calculated based on the multiple standardized feature spaces, and a terrain similarity quantification array is output. After superimposing the risk homogeneity identifier array and the terrain similarity quantification array, spatial continuity topological clustering is performed to output the multiple track partitions.

[0060] Safety requirements vary across different track areas, depending on factors such as train speed, track curvature, gradient, and environmental conditions (e.g., adverse weather). Each terrain feature (e.g., elevation gradient, curvature, vegetation cover) has a different impact on track safety. Therefore, different weights are assigned to different terrain features based on track safety requirements; for example, areas with steeper gradients and unstable ground require higher weights. After assigning risk weights to each terrain feature, multiple terrain feature vectors are weighted and normalized, integrating terrain features with different weights into a unified, standardized feature space. This normalization process allows terrain features of different scales and types to be compared under the same standard.

[0061] The multi-feature risk threshold is defined based on the weighted values ​​of terrain features and track safety requirements. For example, some areas have excessively large elevation gradients, requiring a lower threshold to identify high-risk areas. All raster cells are traversed, and the normalized feature space of each cell is checked to see if it exceeds the preset multi-feature risk threshold. If the feature value of a raster cell exceeds the threshold, it indicates that the area has a high risk and requires special attention. After threshold filtering, the risk consistency between adjacent raster cells is checked. If adjacent raster cells have similar risk features and all exceed the risk threshold, these raster cells are considered to have risk homogeneity. The output risk homogeneity identifier array identifies which raster cells belong to homogeneous areas, i.e., which areas have similar terrain features and risk levels. This helps determine which areas share common security threats, thus enabling more accurate monitoring and early warning.

[0062] Based on the standardized feature space, the terrain similarity between adjacent rasters is calculated. For example, Euclidean distance and cosine similarity are used to evaluate whether the terrain features (such as elevation gradient, vegetation density, etc.) of adjacent rasters are similar. If the terrain features of adjacent rasters are very similar, their terrain similarity will be higher. The calculated terrain similarity of adjacent rasters is quantified into a terrain similarity quantification array, where each element represents the degree of similarity between adjacent rasters.

[0063] By combining the risk homogeneity identifier array with the terrain similarity quantification array, and superimposing these two arrays, both risk similarity and terrain similarity can be considered simultaneously, forming a more accurate basis for regional division. Based on the superimposed array, spatial continuity topological clustering is performed, which is a process of aggregating similar grid regions. The goal is to find those spatially continuous regions with similar risk and terrain features, and to divide these regions into the same track partition. Finally, after cluster analysis, multiple track partitions are output, each representing a track region that is similar in terms of terrain and risk features.

[0064] Furthermore, the method involves matching monitoring strategies based on multiple fused feature vectors from the multiple orbital partitions to output multiple benchmark monitoring strategies.

[0065] Based on the grid structure of the first track partition, a first set of terrain feature vectors is extracted; PCA dimensionality reduction is performed on the first set of terrain feature vectors based on the variance contribution rate constraint to obtain a first principal component feature subset; the first principal component feature subset is input into the fuzzy rule decision engine, and risk level mapping is performed according to the partition type of the first track partition to output the risk level of the first partition; a first benchmark monitoring strategy is matched with the risk level of the first partition in the dynamic strategy mapping matrix.

[0066] The first track partition is any one of multiple track partitions as the current analysis object. Each track partition is divided into multiple small grid cells. The grid composition refers to the multiple grid cells within the first track partition. Each grid cell represents a small area. In each grid cell, a series of terrain features can be extracted, including elevation gradient, surface curvature, vegetation cover density, water body distribution distance, and proximity of artificial structures, forming the first set of terrain feature vectors.

[0067] Principal Component Analysis (PCA) dimensionality reduction is used to reduce high-dimensional features to a lower-dimensional space while preserving as much original information as possible. Here, PCA is used to reduce the dimensionality of the first set of terrain feature vectors, reducing data complexity. By extracting the first few principal components, the part with the largest variance in the data is retained, reducing redundant information. The variance contribution rate represents the proportion of each principal component's contribution to the total variance of the data. To ensure that too much valuable information is not lost after dimensionality reduction, a threshold is usually set, requiring the total variance contribution rate of the selected principal components to reach a preset value, such as above 95%. In this way, the first principal component feature subset is extracted, which consists of several main features that can explain most of the data variability, reducing computational load while retaining effective information.

[0068] The fuzzy rule decision engine is based on predefined rules. For example, if the slope is steep and the elevation changes significantly, the area is considered a high-risk zone. Fuzzy logic is used to assess the risk level of these feature combinations. The first principal component feature subset obtained after dimensionality reduction is input into the fuzzy rule decision engine. The risk level of the first track partition is assessed through fuzzy inference rules. The output risk level of the first partition represents the risk level of the first track partition, such as dividing it into high-risk area (Level S), medium-risk area (Level A), low-risk area (Level B), and buffer zone (Level C).

[0069] The dynamic strategy mapping matrix is ​​a pre-defined table that defines the baseline monitoring strategy for different risk levels in different zones. This includes the baseline monitoring frequency and the monitoring sensor configuration. The baseline monitoring frequency is the frequency at which the sensors need to collect data, and the frequency will vary depending on the risk level. The monitoring sensor configuration includes different combinations of sensors, which are set according to different risk levels.

[0070] For example, for high-risk areas (Level S), a higher monitoring baseline frequency is set, such as 100Hz+, while using multiple sensors, such as radar, visible light, and infrared three-sensor fusion; for medium-risk areas (Level A), a moderate monitoring baseline frequency is set, such as 30Hz, while using only radar and visible light dual-sensor fusion; for low-risk areas (Level B), a lower monitoring baseline frequency is set, such as 10Hz, while using only a single sensor, such as single radar scanning; for buffer zones (Level C), the lowest monitoring baseline frequency is set, such as 5Hz or event-triggered, while using only a single sensor, such as video motion detection.

[0071] Furthermore, the method involves inputting the first principal component feature subset into a fuzzy rule decision engine, mapping risk levels according to the partition type of the first track partition, and outputting the risk level of the first partition.

[0072] Based on the partition type of the first track partition, dynamic feature weighting is performed on the first principal component feature subset to generate a first weighted feature vector; fuzzy membership function transformation is performed on the first weighted feature vector to construct a first fuzzy input vector; the first fuzzy input vector is loaded into a pre-constructed hierarchical fuzzy rule base to perform cross-level risk reasoning calculation and output a first fuzzy risk value, wherein the hierarchical structure design of the hierarchical fuzzy rule base includes basic risk factor rules, composite risk coupling rules, and special scenario correction rules; the first fuzzy risk value is defuzzified to output the risk level of the first partition.

[0073] Zoning type refers to the division of rail zones based on specific terrain, usage, or function. For example, some rail zones involve higher traffic volumes or more complex terrain, while others are flatter and have lower traffic volumes. Dynamic feature weighting is applied to the first principal component feature subset based on the zoning type. Dynamic feature weighting means adjusting the weights of different terrain features according to the zoning type of the first rail zone. For example, in high-altitude mountainous areas, elevation is given higher weight, while in urban plains, features such as vegetation density and water distribution are emphasized. After adjusting the different terrain features according to their weights, a first weighted feature vector is formed, providing a more accurate data foundation for subsequent risk assessment.

[0074] Each feature in the first weighted feature vector is converted into a fuzzy membership value that is suitable for fuzzy logic rules. For example, the slope feature is converted into a fuzzy linguistic value such as "flat" or "steep" through a fuzzy membership function to represent the degree of membership of the feature under different risk scenarios. The converted fuzzy membership values ​​are organized into a first fuzzy input vector, which contains the features of the track partition and their corresponding memberships, representing the terrain and environmental conditions of the area.

[0075] The hierarchical fuzzy rule base contains multi-level rules. Basic risk factor rules handle risk assessments based on simple topographic features, such as the impact of a single slope or curvature value on risk. Composite risk coupling rules involve combinations of multiple features; for example, slope and vegetation cover together may have a greater impact on track safety. Special scenario correction rules address specific or extreme scenarios, such as risk assessments under extreme weather or special terrain conditions. Based on the hierarchical fuzzy rule base, cross-level reasoning is performed on the first fuzzy input vector—that is, the corresponding rules are searched and calculated—to ultimately obtain the first fuzzy risk value. This first fuzzy risk value is a fuzzy assessment of track zone safety, incorporating a comprehensive consideration of multiple factors.

[0076] The obtained first fuzzy risk value is converted into a specific risk level through a defuzzification method. The defuzzification process is based on the maximum membership method or the weighted average method to convert the fuzzy value into an actual risk level. For example, if the first fuzzy risk value indicates that the risk of the area is high, then after defuzzification, the risk level of the area is determined to be a medium-risk area (level A) or a high-risk area (level S), depending on the setting of the rule base.

[0077] Furthermore, the terrain feature vector consists of elevation gradient, surface curvature, vegetation cover density, water body distribution distance, and proximity of artificial structures.

[0078] Elevation gradient indicates the slope and undulation of the area, usually obtained by calculating the elevation difference between adjacent grids. Areas with steep slopes may affect the stability of train operation. Surface curvature describes the surface features of the track area, helping to identify features such as steep slopes and curves. Vegetation cover density is the vegetation cover of the area around the track calculated using remote sensing or geographic information system technology. Areas with excessive vegetation may lead to reduced visibility and the risk of obstructing the observation of obstacles. Water body distribution distance is obtained by calculating the distance between the track and nearby water bodies (such as rivers, lakes, etc.). Water bodies may weaken the track foundation or obstruct the passage of trains. The proximity of artificial structures is obtained by assessing the distance between the track and nearby artificial structures such as buildings, bridges, tunnels, and roads. These structures may pose potential obstacle threats to the track area.

[0079] Furthermore, after deploying multiple integrated sensor arrays for the multiple track zones according to multiple monitoring sensor configurations, and performing dynamic sensing monitoring of obstacle intrusions in the multiple track zones based on multiple monitoring reference frequencies, the method includes:

[0080] Multiple edge obstacle intrusion detection nodes are pre-deployed in the multiple track zones, wherein the multiple integrated sensor arrays are connected to the multiple edge obstacle intrusion detection nodes via low-latency communication links; the multiple integrated sensor arrays send time-series sensor data streams to the multiple edge obstacle intrusion detection nodes in a polling manner, constrained by the multiple monitoring reference frequencies, to perform dynamic sensing and monitoring of obstacle intrusions in the multiple track zones.

[0081] In multiple track zones, based on the characteristics of each track (such as risk level and terrain complexity), multiple edge obstacle intrusion detection nodes are pre-deployed. These nodes are distributed small computing units located close to the sensors, reducing data transmission latency. These edge obstacle intrusion detection nodes can analyze obstacle data in real time, thus reacting promptly when an obstacle intrusion occurs. Each node is responsible for obstacle detection and intrusion event identification in its assigned track zone. To ensure data real-time performance and accuracy, multiple integrated sensor arrays are connected to the multiple edge obstacle intrusion detection nodes via low-latency communication links. This link ensures that data collected from the sensors can be transmitted to the edge obstacle intrusion detection nodes at the fastest possible speed, thereby significantly reducing the system's response time.

[0082] Each track zone's integrated sensor array collects environmental data at a set monitoring baseline frequency and periodically sends the data stream to the corresponding edge obstacle intrusion detection node. Each integrated sensor array periodically sends its collected data stream in a time-series manner to ensure the data stream has a temporal order, facilitating analysis and response. The edge obstacle intrusion detection node analyzes the sensor data in real time to determine whether an obstacle intrusion event has occurred.

[0083] Furthermore, when the Kth track zone identifies a real-time obstacle, a directional and hierarchical early warning command is generated based on the real-time train positioning and the spatial topology relationship of the Kth track zone. The method includes:

[0084] Multi-dimensional features and spatial coordinates of obstacles are extracted from real-time obstacles based on multimodal sensing data; the multi-dimensional features are weighted, normalized, and fused to quantify and output a real-time threat score, wherein the multi-dimensional features include geometric dimensions, material type, and motion vector features; the real-time train speed is interactively obtained, and the collision time of track geometric constraints is calculated based on the train speed and obstacle spatial coordinates; a graded warning matching is performed based on the coupling relationship between the real-time threat score and the collision time, and the directional graded warning command is output.

[0085] Multimodal sensing data is real-time obstacle information collected by different types of sensors, including images, sounds, distances, speeds, and other aspects of obstacles. Multi-dimensional features of obstacles are extracted from these data, including geometric dimensions, material types, and motion vectors. The spatial coordinates of the obstacles are determined, i.e., the real-time position of the obstacles relative to the track and train, which is obtained through sensor localization.

[0086] Multidimensional features of obstacles include geometric dimensions, material type, and motion vector features. Geometric dimensions indicate the physical properties of obstacles, such as size, shape, and volume. Material type features indicate the material of obstacles, such as metal, wood, and plastic. These features help to determine their potential impact or danger. Motion vector features indicate the motion state of obstacles, including their speed, direction, and acceleration. If an obstacle is moving, its danger is higher.

[0087] The multi-dimensional features of obstacles are weighted and normalized, assigning different weights to each feature according to its importance or contribution. For example, larger obstacles pose a greater threat than minor ones and are therefore given higher weights. Normalization unifies the values ​​of different features to the same standard range, allowing them to be fused at the same scale. This prevents any single feature from dominating the result due to its large numerical range. The weighted and normalized features are then fused into a unified real-time threat score using methods such as weighted averaging. This real-time threat score is a quantifiable value representing the degree of threat that obstacles pose to train safety in real time. Obstacles with higher real-time threat scores require more urgent responses.

[0088] The system obtains the real-time train speed, which comes from the train's real-time positioning system or through interaction with the railway dispatching system. Based on the train speed and the spatial coordinates of the obstacle, the collision time is estimated. The collision time refers to the expected moment when the train encounters the obstacle. Track geometric constraints refer to the fact that factors such as the curve and gradient of the track affect the train's braking distance and collision time.

[0089] Based on the coupling relationship between real-time threat scores and collision times, a tiered early warning matching system is implemented. If the real-time threat score is high and the collision time is close, an emergency warning is triggered; if the real-time threat score is low and the collision time is far off, an immediate warning is not required. Based on the warning matching results, specific targeted tiered early warning commands are output, thus providing intelligent dynamic responses for train and track management systems. This enables precise early warnings before hazards occur, maximizing track safety.

[0090] In summary, the dynamic detection and early warning method for track obstacles provided in this application has the following technical effects:

[0091] By retrieving the original survey data and route design information of the target track, precise track geometry and design parameters can be obtained, providing accurate foundational data for subsequent risk analysis and monitoring strategy development. Based on the original survey data and route design information, the track is divided into multiple risk zones. This hierarchical zone management allows for customized monitoring strategies to be developed according to the characteristics of different areas, achieving more efficient and accurate risk warnings. By matching monitoring strategies to the fused feature vectors of multiple track zones, a baseline monitoring strategy that meets actual needs can be automatically generated. This strategy matching can set corresponding monitoring frequencies and sensor configurations based on the specific characteristics of each track zone, thereby achieving personalized monitoring and warnings and improving the system's flexibility and accuracy. Based on the configuration of multiple monitoring sensors, integrated sensor arrays are deployed in each track zone to comprehensively monitor obstacles within the track area. The system dynamically detects obstruction intrusions on the track based on a baseline monitoring frequency. This dynamic monitoring provides real-time feedback and adjusts the monitoring frequency according to the needs of different risk areas, thereby improving monitoring efficiency and response speed. By combining real-time train positioning with the spatial topology of track zones, the system triggers adaptive frequency upsampling of the baseline monitoring frequency. This allows for dynamic adjustment of the monitoring frequency based on train location and the risk status of the track area. This frequency upsampling mechanism ensures efficient monitoring while avoiding unnecessary monitoring burdens, improving system responsiveness and resource utilization efficiency. When an obstruction is identified within any track zone, a directional, graded early warning command is generated based on the real-time train positioning and the spatial topology of the obstruction. This graded early warning adjusts the warning level according to the severity of the threat, improving the accuracy of the warning and the timeliness of the response, thus ensuring effective protection of rail transit safety.

[0092] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A dynamic detection and early warning device for track obstacles, characterized in that, The device includes: The data retrieval module is used to retrieve the original survey data and route design information of the target track; The multi-level risk zoning module is used to perform multi-level risk zoning of the target track based on the original surveying data and line design information, thereby obtaining multiple track zonings; The monitoring strategy matching module is used to perform monitoring strategy matching based on multiple fused feature vectors of the multiple track partitions and output multiple benchmark monitoring strategies. The benchmark monitoring strategies include benchmark monitoring frequency and monitoring sensor configuration. Each track partition has multiple feature vectors. These feature vectors are fused to obtain a complete and comprehensive fused feature vector. The sensing and monitoring module is used to perform dynamic sensing and monitoring of obstacle intrusion in the multiple track zones after deploying multiple integrated sensing arrays according to multiple monitoring sensor configurations and based on multiple monitoring reference frequencies. An adaptive frequency upsampling trigger module is used to trigger adaptive upsampling of the multiple monitoring reference frequencies based on the real-time train positioning of the real-time rail train and the spatial topology relationship of the multiple track partitions. The warning command generation module is used to generate directional and hierarchical warning commands based on the spatial topological relationship between the real-time train positioning and the Kth track zone when a real-time obstacle is identified in the Kth track zone. The adaptive up-frequency triggering module includes: A zone positioning unit is used to locate multiple zone boundary positioning beacons in the multiple track zones according to the real-time track train's operating line topology. The dynamic calculation unit is used to dynamically calculate multiple spatial topology parameters based on the real-time train positioning and multiple zone boundary positioning beacons, wherein the spatial topology parameters include dynamic Euclidean distance, relative azimuth angle and remaining arrival time; The up-frequency partition positioning unit is used to verify whether the multiple spatial topology parameters meet the up-frequency conditions based on multi-dimensional trigger thresholds, and to filter and locate M up-frequency partitions. The template matching unit is used to perform upsampling strategy template matching based on the M spatial topology parameters and M partition risk levels of the M upsampling partitions to obtain M real-time upsampling strategies; An adaptive upsampling unit is used to adaptively upsampling the M monitoring reference frequencies of the M upsampling partitions according to the M real-time upsampling strategies, wherein the upsampling process dynamically performs load balancing scheduling based on a sliding monitoring window. The early warning instruction generation module includes: The coordinate extraction unit is used to extract the multi-dimensional features of the obstacle and the spatial coordinates of the obstacle based on multimodal sensing data; The scoring quantization unit is used to weighted normalize and fuse the multi-dimensional features of the obstacle, and quantify and output a real-time threat score. The multi-dimensional features of the obstacle include geometric size features, material type features, and motion vector features. The collision time calculation unit is used to interactively obtain the real-time train speed of the track train, and calculate the collision time of the track geometric constraints based on the train speed and the spatial coordinates of the obstacle. The graded warning matching unit is used to perform graded warning matching based on the coupling relationship between the real-time threat score and the collision time, and output the directional graded warning command.

2. The dynamic detection and early warning device for track obstacles as described in claim 1, characterized in that, The multi-level risk zoning module includes: The model building unit is used to build a track BIM model based on the original surveying data and route design information. The model segmentation unit is used to segment the track BIM model based on a preset grid scale to obtain multiple grid-level track models. A vector extraction unit is used to extract multiple terrain feature vectors from the multiple grid-level track models; The clustering and merging unit is used to cluster and merge the orbital grids of the target orbit based on the multiple terrain feature vectors to obtain the multiple orbital partitions.

3. The dynamic detection and early warning device for track obstacles as described in claim 2, characterized in that, The multi-level risk zoning module includes: The weight configuration unit is used to configure terrain feature risk weights according to track safety requirements, and to perform weighted normalization processing on the multiple terrain feature vectors according to the terrain feature risk weight configuration to obtain multiple standardized feature spaces. The homogeneity verification unit is used to traverse the multiple standardized feature spaces based on multi-feature risk thresholds, perform risk homogeneity verification of adjacent grids, and output a risk homogeneity identifier array. The similarity calculation unit is used to calculate the terrain similarity of adjacent grids based on the multiple standardized feature spaces and output a terrain similarity quantification array. The topological clustering unit is used to perform spatial continuity topological clustering after superimposing the risk homogeneity identifier array and the terrain similarity quantification array, and output the multiple orbital partitions.

4. The dynamic detection and early warning device for track obstacles as described in claim 3, characterized in that, The monitoring strategy matching module includes: The feature vector extraction unit is used to extract the first set of terrain feature vectors based on the grid structure of the first track partition; The dimensionality reduction unit is used to perform PCA dimensionality reduction on the first set of terrain feature vectors based on the variance contribution rate constraint to obtain the first principal component feature subset. The risk level mapping unit is used to input the first principal component feature subset into the fuzzy rule decision engine, perform risk level mapping according to the partition type of the first track partition, and output the risk level of the first partition. The strategy matching unit is used to match the first benchmark monitoring strategy of the first partition risk level in the dynamic strategy mapping matrix.

5. The dynamic detection and early warning device for track obstacles as described in claim 4, characterized in that, The monitoring strategy matching module includes: The dynamic feature weighting unit is used to perform dynamic feature weighting on the first principal component feature subset based on the partition type of the first track partition, and generate a first weighted feature vector; The function transformation unit is used to perform fuzzy membership function transformation on the first weighted feature vector to construct the first fuzzy input vector; The reasoning and calculation unit is used to load the first fuzzy input vector into a pre-built hierarchical fuzzy rule base, perform cross-level risk reasoning and calculation, and output the first fuzzy risk value. The hierarchical structure design of the hierarchical fuzzy rule base includes basic risk factor rules, composite risk coupling rules, and special scenario correction rules. The defuzzing processing unit is used to defuzzify the first fuzzy risk value and output the first partition risk level.

6. The dynamic detection and early warning device for track obstacles as described in claim 2, characterized in that, The terrain feature vector consists of elevation gradient, surface curvature, vegetation cover density, water body distribution distance, and proximity of artificial structures.

7. The dynamic detection and early warning device for track obstacles as described in claim 1, characterized in that, The sensing and monitoring module includes: A node deployment unit is used to pre-deploy multiple edge obstacle intrusion detection nodes in the multiple track partitions, wherein the multiple integrated sensor arrays and the multiple edge obstacle intrusion detection nodes are connected by a low-latency communication link; The sensing and monitoring unit is used to send time-series sensing data streams to the multiple edge obstacle intrusion identification nodes in a polling manner, constrained by the multiple monitoring reference frequencies, to perform dynamic sensing and monitoring of obstacle intrusion in the multiple track partitions.

8. A method for dynamic detection and early warning of track obstacles, characterized in that, Based on the implementation of the dynamic detection and early warning device for track obstacles according to any one of claims 1-7, the method includes: Retrieve the original survey data and route design information of the target track; Based on the original survey data and route design information, the target track is divided into multiple risk zones, resulting in multiple track zones. Based on multiple fused feature vectors of the multiple track partitions, a monitoring strategy is matched to output multiple benchmark monitoring strategies, wherein the benchmark monitoring strategy includes a benchmark monitoring frequency and a monitoring sensor configuration; After deploying multiple integrated sensor arrays for the multiple track zones according to the configuration of multiple monitoring sensors, dynamic sensing and monitoring of obstacle intrusion in the multiple track zones are carried out based on multiple monitoring reference frequencies. Based on the real-time train positioning of the real-time rail train and the spatial topology relationship of the multiple track zones, the adaptive upsampling of the multiple monitoring reference frequencies is triggered. When the Kth track zone identifies a real-time obstacle, a directional and hierarchical early warning command is generated based on the real-time train positioning and the spatial topology relationship of the Kth track zone.

Images