Scene adaptive ice column detection method and system based on vehicle-mounted mobile monitoring

Abstract

The application discloses a scene adaptive ice column detection method and system based on vehicle-mounted mobile monitoring, relates to the technical field of railway safety monitoring, and discloses a scene adaptive ice column detection method and system based on vehicle-mounted mobile monitoring. The disclosed scene adaptive ice column detection method and system based on vehicle-mounted mobile monitoring activate a detection module by acquiring environmental temperature data, identify a scene type, process image data, analyze features, and review a model to generate an alarm signal, realize scene adaptive ice column detection under vehicle-mounted mobile monitoring, solve the problem that the prior art cannot identify ice column hidden dangers in real time, and can improve railway operation safety and efficiency.

Description

Technical Field

[0001] This application relates to the field of railway safety monitoring technology, and in particular to a scene-adaptive icicle detection method and system based on vehicle-mounted mobile monitoring. Background Technology

[0002] With the rapid development of high-speed railway networks, the train operating environment is becoming increasingly complex and variable. In the harsh winter conditions, icicles easily form on the arches of tunnels and the edges of bridges around railway lines. Once these icicles break off, they can pose a serious threat to high-speed trains and even cause safety accidents. Currently, the condition monitoring of the overhead contact system and its surrounding environment mainly relies on onboard overhead contact system operation status monitoring devices (commonly known as "3C" devices). However, existing 3C devices and their accompanying algorithms are only designed to monitor the geometric parameters of the overhead contact system, pantograph wear, and pantograph-catenary arcing status. Their core function focuses on assessing the health status of the overhead contact system itself, and they are completely unable to identify potential icicle hazards in the surrounding environment. This not only results in a huge waste of onboard hardware resources but also creates blind spots in safety monitoring. At the same time, the industry lacks real-time icicle detection technology adapted to the high-speed movement of high-speed trains, because high-speed movement leads to unstable image acquisition, drastic changes in ambient light, and different scenarios such as inside tunnels, tunnel entrances, and open-air bridges have drastically different background characteristics and interference sources. For example, tunnel scenes are often dimly lit with a monotonous background, while tunnel entrance scenes present abrupt changes in brightness and dynamic transitions. Open-air bridge scenes also face challenges such as interference from the sky background and complex temperature gradients, making it difficult for traditional detection methods to effectively distinguish icicles from environmental disturbances. Therefore, railway departments still primarily rely on inefficient and limited-coverage manual inspections, failing to achieve real-time early warning and proactive handling of icicle hazards, thus hindering the safety and efficiency of railway operations.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this application is to provide a scene-adaptive icicle detection method and system based on vehicle-mounted mobile monitoring, aiming to improve the accuracy and real-time performance of icicle detection.

[0005] To achieve the above objectives, this application proposes a scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring, the method comprising: The system acquires ambient temperature data. When the ambient temperature data is lower than a preset temperature threshold, the system activates the icicle detection module and acquires infrared thermal image data collected by the vehicle-mounted infrared thermal imager and visible light image data collected by the vehicle-mounted visible light camera. Based on the visible light image data, the scene type of the current train is identified, and the scene identification result is obtained; the scene type includes tunnel scene, tunnel entrance scene and open bridge scene. Based on the infrared thermal image data, grayscale histogram statistics and adaptive threshold segmentation are performed to obtain the data of the first suspected icicle region. Morphological and brightness feature analyses were performed on the data of the first suspected icicle region to obtain the data of the second suspected icicle region. Based on the scene recognition results and the data of the second suspected icicle region, a multidimensional temperature statistical analysis is performed on the data of the second suspected icicle region to obtain the data of the third suspected icicle region. The data of the third suspected icicle region is input into a pre-trained deep learning model for verification to obtain icicle detection result data, and a graded alarm signal is generated based on the icicle detection result data.

[0006] In one embodiment, the step of identifying the scene type of the current train based on the visible light image data and obtaining the scene identification result includes: Based on the visible light image data, calculate the average brightness change rate data and image texture feature data; Obtain train location information data; Based on the image average brightness change rate data, the image texture feature data, and the train position information data, the current scene type is determined through state machine logic to obtain the scene recognition result.

[0007] In one embodiment, the step of determining the current scene type and obtaining the scene recognition result based on the image average brightness change rate data, the image texture feature data, and the train position information data through state machine logic includes: Based on the average brightness change rate data of the image, when the brightness value is detected to decrease by more than the first preset threshold within a preset time window, it is determined to be a scene of entering a tunnel entrance. Based on the image texture feature data, when the image texture complexity is detected to be lower than the second preset threshold, it is determined to be a scene inside a tunnel; Based on the train location information data and combined with a pre-stored line geographic information database, the preset scenario type corresponding to the current train location is determined. Based on the scene determination results at the tunnel entrance, the scene determination results inside the tunnel, and the preset scene type, the final scene recognition result is determined through state machine state transition logic. Based on the scene recognition results, the corresponding detection parameter set is dynamically loaded from the preset parameter database; the detection parameter set includes infrared preprocessing parameters, temperature analysis thresholds and interference filtering rules for different scenes.

[0008] In one embodiment, the step of performing morphological feature analysis and brightness feature analysis on the first suspected icicle region data to obtain the second suspected icicle region data includes: Contour feature calculation is performed on each of the first suspected icicle regions to obtain contour feature data; Based on the contour feature data, regions with inverted triangle features (wider at the top and narrower at the bottom) or elongated features are selected to obtain morphological screening result data. Calculate the average brightness value of the region corresponding to each of the morphological screening results to obtain the region average brightness data; Based on the average brightness data of the region, low-brightness feature areas are retained and high-brightness noise areas are removed to obtain the second suspected icicle region data.

[0009] In one embodiment, the step of performing multidimensional temperature statistical analysis on the second suspected icicle region data based on the scene recognition result and the second suspected icicle region data to obtain the third suspected icicle region data includes: Map the data of the second suspected icicle region back to the original infrared temperature data matrix to obtain the temperature data of each region; Based on the scene recognition results and the temperature data of each region, when the scene is an open-air bridge scene, the open-air bridge scene temperature analysis process is executed; when the scene is a tunnel scene, the tunnel scene temperature analysis process is executed, and the temperature analysis result data is obtained. Based on the temperature analysis results, interference areas were eliminated to obtain the data for the third suspected icicle region.

[0010] In one embodiment, the temperature analysis process for the open-air bridge scene includes: Based on the temperature data of each region, the minimum temperature value, average temperature value and temperature variance value of each region are calculated to obtain temperature statistical characteristic data. Based on the temperature statistical feature data, when the lowest temperature value of a certain area is lower than the preset deep space temperature threshold, the area is determined to be a sky background interference area. Based on the temperature statistical feature data and the ambient temperature data, the average temperature value of each region is compared with the ambient temperature data to determine whether it conforms to the temperature characteristics of an ice-water mixture. Based on the results of the sky background interference area determination and the results of the ice-water mixture temperature characteristic determination, areas that meet the temperature characteristics of ice columns are selected to obtain the outdoor scene temperature analysis results data.

[0011] In one embodiment, the tunnel scene temperature analysis process includes: Based on the temperature data of each region, the temperature variance value of each region and the temperature variance value of the surrounding background are calculated to obtain temperature variance feature data. Based on the temperature variance feature data, when the temperature variance value of the surrounding background of a certain area is less than the preset background variance threshold and there is a temperature gradient difference between the area and the background, it is determined to be a highly suspected icicle area. Based on the temperature data of each region and the ambient temperature data, the difference between the average temperature value of each region and the ambient temperature data is calculated to determine whether it conforms to the temperature characteristics of ice columns. Based on the results of the highly suspected icicle area determination and the results of the icicle temperature characteristic determination, areas that meet the icicle temperature characteristics are selected to obtain tunnel scene temperature analysis results data.

[0012] In one embodiment, the steps of inputting the data of the third suspected icicle region into a pre-trained deep learning model for verification to obtain icicle detection result data, and generating a graded alarm signal based on the icicle detection result data include: The visible light image sub-region and the infrared thermal image sub-region corresponding to the third suspected icicle region data are fused to obtain multimodal input data; The multimodal input data is input into a pre-trained target detection neural network model for deep learning verification to obtain deep learning verification result data; Based on the deep learning verification results data, false targets with similar shapes but inconsistent texture features are filtered out to obtain icicle detection results data; Based on the icicle detection results, the location information and confidence information of the icicle in the image are obtained; Based on the location and confidence information of the icicles in the image, a graded alarm signal is generated; The graded alarm signals are pushed to the driver and passengers or the ground data center in real time through the vehicle terminal.

[0013] In one embodiment, the step of generating a graded alarm signal based on the location information and confidence information of the icicle in the image includes: Based on the position information of the icicle in the image and the pre-established coordinate mapping relationship, the position information of the icicle in the image is converted into the actual spatial position information of the icicle. Based on the actual spatial location information of the icicle and the preset data of the contact wire safety area, it is determined whether the actual spatial location of the icicle exceeds the boundary of the safety area, and the intrusion judgment result is obtained. Based on the actual spatial location information of the icicle and the location data of the center line of the contact network, the distance between the actual spatial location of the icicle and the center line of the contact network is calculated, and it is determined whether the icicle is located directly above the contact network, thus obtaining the result of the direct-above determination. Based on the confidence level information, the detection reliability level is determined; Based on the intrusion limit judgment result, the direct upward judgment result, and the detection reliability level, a graded alarm signal is generated; the graded alarm signal includes an emergency alarm signal, a warning signal, and a prompt signal.

[0014] Furthermore, to achieve the above objectives, this application also proposes a scene-adaptive icicle detection system based on vehicle-mounted mobile monitoring. The scene-adaptive icicle detection system based on vehicle-mounted mobile monitoring includes: a memory, a processor, and a scene-adaptive icicle detection program based on vehicle-mounted mobile monitoring stored in the memory and executable on the processor. The scene-adaptive icicle detection program based on vehicle-mounted mobile monitoring is configured to implement the steps of the scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring.

[0015] The proposed method and system for scene-adaptive icicle detection based on vehicle-mounted mobile monitoring achieves scene-adaptive icicle detection under vehicle-mounted mobile monitoring by acquiring ambient temperature data to activate the detection module, identifying scene types, processing image data, analyzing features, and verifying the model to generate alarm signals. This solves the problem that existing technologies cannot identify icicle hazards in real time and can improve the safety and efficiency of railway operations. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring provided in this application. Figure 2 This is a schematic diagram of a scenario-adaptive icicle detection system based on vehicle-mounted mobile monitoring, as provided in this application.

[0019] Explanation of icon numbers: 10. Memory; 20. Processor.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] It should be understood that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] In existing technologies, while monitoring the geometric parameters of the overhead contact system, the wear of the pantograph contactor, and the arcing status of the pantograph-catenary system, onboard overhead contact system monitoring devices cannot identify potential icicle hazards in the surrounding environment, creating blind spots in safety monitoring. Furthermore, the industry lacks real-time icicle detection technology adapted to the high-speed movement of high-speed trains, forcing railway departments to still rely on inefficient manual inspections, thus failing to achieve real-time early warning and proactive handling of icicle hazards.

[0024] Based on this, this application provides a scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring, referring to... Figure 1 The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring includes steps S100 to S600, wherein: Step S100: Obtain ambient temperature data. When the ambient temperature data is lower than a preset temperature threshold, activate the icicle detection module and obtain infrared thermal image data collected by the vehicle-mounted infrared thermal imager and visible light image data collected by the vehicle-mounted visible light camera. Step S200: Based on the visible light image data, identify the scene type where the train is currently located and obtain the scene identification result; the scene type includes tunnel scene, tunnel entrance scene and open bridge scene; Step S300: Based on the infrared thermal image data, perform grayscale histogram statistics and adaptive threshold segmentation to obtain the data of the first suspected icicle region. Step S400: Perform morphological feature analysis and brightness feature analysis on the data of the first suspected icicle region to obtain the data of the second suspected icicle region. Step S500: Based on the scene recognition result and the second suspected icicle region data, perform multidimensional temperature statistical analysis on the second suspected icicle region data to obtain the third suspected icicle region data; Step S600: Input the data of the third suspected icicle region into the pre-trained deep learning model for verification to obtain icicle detection result data, and generate a graded alarm signal based on the icicle detection result data.

[0025] In this embodiment, ambient temperature data refers to the real-time temperature information around the train's operating environment, typically acquired through onboard temperature sensors, used to determine whether the low-temperature conditions for icicle formation are met. The preset temperature threshold is a pre-set temperature limit value; when the ambient temperature falls below this threshold, it indicates a potential risk of icicle formation, thus triggering the icicle detection process. The icicle detection module refers to the software or hardware unit in the system responsible for executing the icicle detection algorithm and processing related data; its function is to be activated when specific conditions are met to initiate the icicle recognition process. The onboard infrared thermal imager is a device installed on the train to collect infrared radiation energy from object surfaces and convert it into infrared thermal image data, reflecting the temperature distribution of the object's surface. The onboard visible light camera is installed on the train to collect image data within the visible spectrum, providing visual information about the scene and assisting in scene recognition and target localization. Scene type refers to different geographical environments encountered during train operation, such as inside tunnels, tunnel entrances / exits, and open-air bridges; the formation characteristics and detection strategies for icicles may differ in different scenarios.

[0026] In this embodiment, grayscale histogram statistics refers to a method for statistically analyzing the distribution of pixel grayscale values ​​in infrared thermal images. This is used to understand the overall brightness distribution and contrast information of the image, providing a basis for subsequent image processing. Adaptive threshold segmentation is a processing method that automatically determines the segmentation threshold based on local or global image features to separate the foreground from the background, helping to extract potential icicle regions from infrared thermal images. Morphological feature analysis refers to analyzing the geometric features of image regions, such as shape, size, and connectivity, through mathematical morphological operations (such as erosion, dilation, opening, and closing operations) to identify targets with specific shapes. Brightness feature analysis refers to statistically analyzing the brightness values ​​of image regions, such as calculating average brightness and brightness variance, to distinguish targets from the background or identify objects with specific brightness characteristics. Multidimensional temperature statistical analysis refers to performing multi-angle, multi-parameter statistical analysis on the temperature data of the target area, such as calculating the minimum temperature, average temperature, and temperature variance, and combining this with environmental information for comprehensive judgment. The pre-trained deep learning model is a neural network model that has been trained on a large-scale dataset and possesses specific recognition or classification capabilities. In this application, it is used for the final verification and confirmation of suspected icicle areas. The graded alarm signal refers to generating different levels of alarm information based on information such as the degree of danger, location, and confidence level of the icicles, so that drivers or ground control centers can take appropriate measures.

[0027] In this embodiment, to acquire ambient temperature data and activate the icicle detection module, a temperature sensor can be installed on the train to periodically collect real-time temperature information of the train's operating environment. When the detected ambient temperature data is lower than a preset temperature threshold, the system is triggered, activating the icicle detection module. Subsequently, the onboard infrared thermal imager and onboard visible light camera are activated to collect infrared thermal image data and visible light image data, respectively. As an optional implementation, the ambient temperature data can also be manually read periodically or collected using a simple temperature sensor, with the operator manually activating the icicle detection module.

[0028] In this embodiment, regarding the identification of the current train scene type based on visible light image data, simple image processing can be performed on the visible light image, such as analyzing the overall brightness, color distribution, or edge features, to roughly determine the current scene. An image with extremely low brightness and a single color may be initially identified as a tunnel scene, while an image with high brightness and rich colors may be initially identified as an open-air bridge scene. The scene identification result can include tunnel scenes, tunnel entrance scenes, and open-air bridge scenes. Alternatively, scene type identification can also be accomplished by manually observing the visible light image and manually inputting the current scene type.

[0029] In this embodiment, regarding grayscale histogram statistics and adaptive threshold segmentation based on infrared thermal image data, a global grayscale histogram can be calculated for the acquired infrared thermal image data to understand the overall brightness distribution of the image. Subsequently, a globally fixed threshold can be used for segmentation, marking pixels with grayscale values ​​higher than the threshold as foreground (suspected icicle regions) and pixels with grayscale values ​​lower than the threshold as background, thus obtaining the first suspected icicle region data. Alternatively, grayscale histogram statistics can be performed using a fixed number of grayscale levels. Adaptive threshold segmentation can employ a simple local mean thresholding method, for example, comparing the grayscale values ​​of all pixels in the image with a preset fixed threshold; pixels higher than the threshold are marked as foreground, and pixels lower than the threshold are marked as background.

[0030] Furthermore, in performing morphological and brightness feature analysis on the data of the first suspected icicle regions, for each region, its area and perimeter can be calculated, and filtering can be performed based on these simple geometric parameters. Regions that are too small or too large may be directly eliminated. Simultaneously, the average brightness value of each region can be calculated and compared with a preset fixed brightness range; regions exceeding this range may be considered interference. This yields the data for the second suspected icicle regions. Alternatively, morphological feature analysis can determine whether a region is an icicle simply by calculating its area and perimeter. Brightness feature analysis can calculate only the average brightness value of the region and compare it with a fixed brightness range; regions exceeding this range are considered interference.

[0031] Furthermore, regarding the multidimensional temperature statistical analysis of the second suspected icicle region data based on scene recognition results and second suspected icicle region data, the second suspected icicle region data can be mapped back to the original infrared temperature data matrix to obtain the temperature information of each region. Subsequently, the average temperature of each region can be calculated and compared with a globally fixed freezing point temperature (e.g., 0°C). If the average temperature is close to the freezing point, the region is initially identified as an icicle. In this process, no specific scene type is distinguished; all scenes use the same temperature judgment standard. Thus, the third suspected icicle region data is obtained. As another implementation method, multidimensional temperature statistical analysis can calculate only the average temperature of the second suspected icicle region and compare it with a globally fixed freezing point temperature.

[0032] In this embodiment, regarding the input of the data of the third suspected icicle region into a pre-trained deep learning model for verification to obtain icicle detection result data, and the generation of graded alarm signals based on the icicle detection result data, the features (such as shape, size, average temperature, etc.) corresponding to the data of the third suspected icicle region can be input into a pre-trained simple classifier, such as a support vector machine (SVM) or decision tree model, for verification to determine whether it is a real icicle. Based on the output of the classifier, the icicle detection result data is obtained. Subsequently, based on the number of detected icicles or the presence or absence of icicles, a single alarm signal is generated, such as "icicle detected" or "icicle not detected," and a notification is issued through the vehicle terminal. Alternatively, the alarm signal can be generated solely based on the number of detected icicles or the presence or absence of icicles, without graded generation.

[0033] In this embodiment, by combining an environmental temperature triggering mechanism, multimodal image data acquisition, and scene-adaptive image processing and temperature analysis strategies, the blind spots in existing icicle detection technologies can be effectively overcome. This method further utilizes a deep learning model for verification, improving the accuracy and reliability of detection. It can also generate tiered alarm signals based on the detection results, thereby achieving real-time, accurate early warning and handling of icicle hazards around high-speed moving trains, and enhancing the safety of railway operations.

[0034] In one feasible implementation, the step of identifying the scene type of the current train based on the visible light image data and obtaining the scene identification result includes: calculating the average brightness change rate data and image texture feature data based on the visible light image data; obtaining train position information data; and determining the current scene type through state machine logic based on the average brightness change rate data, the image texture feature data, and the train position information data to obtain the scene identification result.

[0035] In this embodiment, the average brightness change rate data is used to reflect the overall brightness change trend of the visible light image over time, which is particularly suitable for detecting scenes with drastic changes in lighting environment, such as trains entering and exiting tunnels. It can be calculated by summing and averaging the pixel brightness of consecutive frames of the visible light image, and then calculating the difference or rate of change in average brightness between adjacent frames. For example, when a train enters a tunnel from a bright open environment, the average brightness of the image will decrease rapidly, and vice versa. Image texture feature data is used to describe the regularity or complexity of pixel distribution in the visible light image, effectively distinguishing the visual structural characteristics of different scenes. For example, the interior of a tunnel usually has relatively simple and repetitive textures, while an open bridge scene may contain richer and more complex background textures. Image texture features can be extracted using various algorithms, such as Gray-Level Co-occurrence Matrix (GLCM), Local Binary Pattern (LBP), or Gabor filters, to quantify texture attributes such as roughness, contrast, and directionality of the image.

[0036] In this embodiment, train location information data is acquired. This data provides geospatial contextual information for scene recognition, serving as a strong prior for auxiliary judgment. Train location information data can be acquired through onboard Global Positioning System (GPS), odometer, Inertial Navigation System (INS), or multi-sensor fusion technology. This data can provide real-time information such as the train's current latitude, longitude, altitude, and direction of travel, and can be matched with a pre-stored line geographic information database to determine the specific line segment and possible scene type where the train is located.

[0037] Based on this, the current scene type is determined through state machine logic using the image average brightness change rate data, image texture feature data, and train position information data, resulting in scene recognition. State machine logic is a decision-making mechanism based on predefined states, events, and state transition rules. It integrates image average brightness change rate data, image texture feature data, and train position information data to achieve robust judgment of the current scene type. This state machine can define states such as "open-air bridge scene," "tunnel entrance scene," and "tunnel interior scene," and uses changes in the aforementioned multi-source data as trigger events to drive the state machine to switch between different scene states. For example, when the image average brightness change rate data indicates a sharp drop in brightness, and the train position information data matches the tunnel entrance area, the state machine can transition from the "open-air bridge scene" to the "tunnel entrance scene."

[0038] In this embodiment, by introducing image average brightness change rate data and image texture feature data, changes in ambient lighting and structure can be captured more sensitively, such as sudden brightness changes when a train enters or exits a tunnel, and differences in texture complexity inside and outside the tunnel. Simultaneously, combining train location information data provides strong geospatial constraints and prior knowledge for scene recognition. By fusing and judging these multi-source heterogeneous data through state machine logic, the limitations of single-feature recognition can be effectively avoided, improving the accuracy and robustness of scene recognition. This ensures that the subsequent icicle detection module can dynamically adjust detection parameters according to the actual scene, enhancing the adaptability and reliability of the overall detection method.

[0039] In one feasible implementation, the step of determining the current scene type and obtaining the scene recognition result based on the image average brightness change rate data, the image texture feature data, and the train position information data through state machine logic includes: based on the image average brightness change rate data, when the brightness value is detected to decrease by more than a first preset threshold within a preset time window, it is determined to be an entry into a tunnel scene; based on the image texture feature data, when the image texture complexity is detected to be lower than a second preset threshold, it is determined to be an inside-tunnel scene; based on the train position information data, combined with a pre-stored line geographic information database, a preset scene type corresponding to the current train position is determined; based on the entry into a tunnel scene determination result, the inside-tunnel scene determination result, and the preset scene type, the final scene recognition result is determined through state machine state transition logic; based on the scene recognition result, a corresponding detection parameter set is dynamically loaded from a preset parameter database; the detection parameter set includes infrared preprocessing parameters, temperature analysis thresholds, and interference filtering rules for different scenes.

[0040] In this embodiment, to accurately identify the critical moment when a train enters a tunnel, the system continuously monitors the average brightness value of the visible light image. When the brightness value decreases by more than a preset first threshold within a preset time window, it is determined that the train is entering a tunnel. This mechanism effectively captures the abrupt change in light from bright to dim. Simultaneously, to identify whether the train is inside a tunnel, the system analyzes the texture feature data of the visible light image. The tunnel environment is typically characterized by uniform lighting and a simple structure, resulting in low image texture complexity. When the detected image texture complexity is below a preset second threshold, the current scene is determined to be a tunnel scene. Furthermore, to provide a stable scene reference, the system also utilizes train location information data obtained from the onboard positioning system and compares it with a pre-stored railway geographic information database to determine the preset scene type corresponding to the current train location. This database contains the precise geographic coordinates and type identifiers of various sections along the railway line (such as tunnels, bridges, and open sections).

[0041] In this embodiment, the aforementioned tunnel entrance scene determination result, tunnel interior scene determination result, and preset scene type obtained based on train position information will be used as inputs to determine the final scene recognition result through state machine state transition logic. The state machine typically contains multiple states (e.g., "open-air bridge scene," "tunnel entrance scene," "tunnel interior scene," etc.) and a series of predefined state transition rules. For example, when the "tunnel entrance scene determination result" is true, the state machine may transition from "open-air bridge scene" to "tunnel entrance scene"; when the "tunnel interior scene determination result" remains true and the train position information also indicates that the train is inside the tunnel, the state machine may transition to "tunnel interior scene." This state machine-based logic effectively handles the uncertainty during scene switching, avoids misjudgments caused by instantaneous fluctuations of a single feature, and ensures the robustness and accuracy of scene recognition. Once the final scene recognition result is determined, the system dynamically loads a set of detection parameters that best matches the current scene from a preset parameter database based on this result. This set of detection parameters is pre-configured for different scene characteristics. For example, it may include infrared preprocessing parameters, temperature analysis thresholds and interference filtering rules for tunnel scenes, or different parameters for open-air bridge scenes.

[0042] In this embodiment, the above-described technical solution effectively addresses the issues of misjudgment and instability that easily occur when recognizing scenes using single or simple combinations of features in complex railway environments. Based on multi-dimensional inputs of image average brightness change rate data, image texture feature data, and train position information data, combined with state machine state transition logic, robust and accurate recognition of tunnel entrance scenes, tunnel interior scenes, and open-air bridge scenes is achieved. The state machine logic effectively filters out interference from instantaneous noise and environmental fluctuations, ensuring the stability and reliability of scene recognition results. Furthermore, the corresponding detection parameter set is dynamically loaded based on the accurate scene recognition results, enabling subsequent icicle detection steps such as infrared preprocessing, temperature analysis, and interference filtering to adapt to the characteristics of different scenes. For example, in tunnel interior scenes, a stricter texture complexity threshold and filtering rules targeting tunnel wall heat sources can be used; in open-air bridge scenes, the temperature analysis threshold can be adjusted to distinguish icicles from the sky background. This adaptive parameter adjustment mechanism improves the accuracy of icicle detection, reduces false alarm and false negative rates, thereby enhancing the reliability and practicality of the entire vehicle-mounted mobile monitoring system.

[0043] In one feasible implementation, the step of performing morphological feature analysis and brightness feature analysis on the first suspected icicle region data to obtain the second suspected icicle region data includes: calculating the contour features of each of the first suspected icicle region data to obtain contour feature data; based on the contour feature data, filtering out regions with an inverted triangle feature that is wider at the top and narrower at the bottom or a long strip feature to obtain morphological filtering result data; calculating the average brightness value of the region corresponding to each of the morphological filtering result data to obtain the region average brightness data; and based on the region average brightness data, retaining low brightness feature regions and removing high brightness noise regions to obtain the second suspected icicle region data.

[0044] In this embodiment, for each initially identified first suspected icicle region, its boundary needs to be accurately depicted and quantized. Edge detection algorithms (such as the Canny operator or the Sobel operator) or contour extraction algorithms (such as the Suzuki algorithm) can be used to identify and track the boundary pixels of the region, thereby obtaining the contour of the region. This contour information, including the coordinate sequence of contour points, the perimeter, area, minimum bounding rectangle, minimum bounding circle, etc., is integrated into contour feature data. This contour feature data provides a basic geometric description for subsequent morphological analysis.

[0045] In this embodiment, icicles typically exhibit specific geometric shapes in their natural state, such as an inverted triangle shape (conical icicle) that is wider at the top and narrower at the bottom, or a long, thin strip shape (columnar icicle). Based on the obtained contour feature data, the shape of each region can be analyzed. For inverted triangle features, the convex hull of the contour can be calculated, and the width ratio of its top and bottom can be analyzed, or a polygon can be fitted to determine whether it conforms to the geometric characteristics of an inverted triangle. For long strip features, the aspect ratio of the region can be calculated by the minimum bounding rectangle or the rotated bounding box. For example, when the aspect ratio is greater than a certain preset threshold, it can be determined to be long strip. In this way, regions with these typical icicle morphological features can be filtered out, thereby obtaining morphological screening result data and effectively eliminating interference objects with inconsistent shapes.

[0046] In this embodiment, icicles typically exhibit low brightness or reflectivity in visible light images, potentially showing a brightness difference compared to their surrounding environment (such as metallic structures or reflective surfaces). For each region after morphological screening, its corresponding average brightness value in the visible light image needs to be calculated. Specifically, all pixels in the visible light image for that region can be extracted, and the average grayscale or brightness value of these pixels can be calculated to obtain the region's average brightness data. This data provides overall brightness information for the region in the visible light band, laying the foundation for subsequent brightness feature analysis.

[0047] In this embodiment, based on the calculated average brightness data of the region, it is possible to further distinguish between genuine icicle targets and high-brightness noise. Due to their material properties, icicles typically exhibit relatively low brightness under visible light. Conversely, some metallic reflections, light reflections, or other high-brightness background interferences will exhibit high brightness. Therefore, a brightness threshold can be set, retaining areas with average brightness values ​​below this threshold as low-brightness feature areas. Simultaneously, areas with average brightness values ​​above this threshold are identified as high-brightness noise areas and discarded. Through this brightness feature analysis, suspected icicle areas can be further refined, yielding more accurate data on second suspected icicle areas.

[0048] In this embodiment, by calculating the contour features of each first suspected icicle region and filtering out regions with inverted triangular or elongated features (wider at the top and narrower at the bottom) based on the contour feature data, a large number of non-target regions whose shapes do not conform to the characteristics of icicles can be effectively eliminated, reducing false detections. Secondly, by calculating the average brightness value of the corresponding regions in the morphological filtering results data, and retaining low-brightness feature regions and eliminating high-brightness noise regions based on the average brightness data, further fine-tuning the suspected regions using the brightness information of the visible light image effectively suppresses high-brightness interference caused by reflections, illumination, etc. This multi-dimensional analysis method combining morphological and brightness features greatly improves the accuracy of identifying real icicle targets from the first suspected icicle region data, reduces the computational burden of subsequent processing, and makes the obtained second suspected icicle region data purer and more reliable. This provides high-quality input for subsequent multi-dimensional temperature statistical analysis and deep learning verification, thereby improving the robustness and accuracy of the entire icicle detection method.

[0049] In one feasible implementation, the step of performing multidimensional temperature statistical analysis on the second suspected icicle region data based on the scene recognition result and the second suspected icicle region data to obtain the third suspected icicle region data includes: mapping the second suspected icicle region data back to the original infrared temperature data matrix to obtain temperature data for each region; based on the scene recognition result and the temperature data for each region, when the scene is an open-air bridge scene, executing an open-air bridge scene temperature analysis process, and when the scene is a tunnel scene, executing a tunnel scene temperature analysis process to obtain temperature analysis result data; and based on the temperature analysis result data, removing interference areas to obtain the third suspected icicle region data.

[0050] In this embodiment, the data of the second suspected icicle region is mapped back to the original infrared temperature data matrix to obtain the temperature data of each region. This step aims to associate the suspected icicle regions, filtered by morphological and brightness analysis, with their corresponding actual temperature information in the original infrared thermal image. Specifically, the system searches for and extracts the temperature values ​​of all pixels contained in these regions from the original infrared temperature data matrix based on the pixel coordinates or geometric positions of the second suspected icicle region data in the image, thereby providing basic, quantified thermal data for subsequent temperature statistical analysis.

[0051] In this embodiment, based on the scene recognition results and the temperature data of each region, when the scene is an open-air bridge scene, a temperature analysis process for the open-air bridge scene is executed; when the scene is a tunnel scene, a temperature analysis process for the tunnel scene is executed, thereby obtaining temperature analysis result data. This step is crucial for achieving scene-adaptive icicle detection. Because the temperature characteristics of icicles and background interference patterns differ in different scenes—for example, an open-air bridge scene may be affected by environmental factors such as deep-space radiation and wind speed, while a tunnel scene may have a relatively stable background temperature but a significant temperature difference between the icicle and the background—the system dynamically selects and executes a temperature analysis process specifically optimized for the current scene based on the previously obtained scene recognition results. For example, for an open-air bridge scene, the focus may be on eliminating deep-space background interference or considering the influence of ambient temperature on the surface temperature of the icicle; while for a tunnel scene, more attention may be paid to the local temperature difference and temperature gradient between the icicle and the tunnel wall. In this way, it is possible to more accurately assess whether the temperature characteristics of the suspected area conform to the physical properties of the icicle and effectively suppress false targets in specific scenes.

[0052] In this embodiment, based on the temperature analysis results, interfering regions are eliminated to obtain the third suspected icicle region data. After scene-adaptive temperature analysis, although most non-icicle interference has been eliminated, a small number of interference regions may still remain. This step aims to refine the temperature analysis results further, removing regions that are still confused with icicles in terms of temperature characteristics. For example, additional elimination rules can be set based on the typical temperature range of icicles, temperature uniformity, or differences from other known interference sources. Through this series of processes, the final third suspected icicle region data will be a highly reliable icicle candidate region, providing high-quality input for subsequent deep learning verification.

[0053] In this embodiment, through the above technical solution, this application can dynamically adjust the temperature statistical analysis strategy of icicles according to the specific scenario of the train (such as an open-air bridge scenario or a tunnel scenario). This scenario-adaptive temperature analysis method can effectively cope with the complex differences in thermal characteristics between icicles and the background under different environments, improving the accuracy and robustness of icicle detection. For example, in an open-air bridge scenario, it can effectively avoid misidentifying deep-space backgrounds as icicles; while in a tunnel scenario, it can more accurately identify icicles with a temperature difference from the tunnel wall. This not only reduces the false alarm rate but also lowers the risk of missed detection, making the icicle detection results more reliable and providing a solid foundation for subsequent graded alarms and safety decisions.

[0054] In one feasible implementation, the temperature analysis process for the open-air bridge scene includes: calculating the minimum temperature value, average temperature value, and temperature variance value of each region based on the temperature data of each region to obtain temperature statistical feature data; determining that a region is a sky background interference region when the minimum temperature value of a certain region is lower than a preset deep space temperature threshold based on the temperature statistical feature data; comparing the difference between the average temperature value of each region and the ambient temperature data based on the temperature statistical feature data and the ambient temperature data to determine whether it conforms to the temperature characteristics of an ice-water mixture; and filtering out regions that conform to the temperature characteristics of ice columns based on the determination results of the sky background interference region and the determination results of the temperature characteristics of the ice-water mixture to obtain the open-air scene temperature analysis result data.

[0055] In this embodiment, under the scenario of an open-air bridge, the minimum temperature, average temperature, and temperature variance of each region are first calculated based on the temperature data of each region to obtain temperature statistical characteristic data. This step aims to quantify the temperature characteristics of each suspected icicle area from multiple dimensions. The minimum temperature value reflects the extreme low temperature point within the region and is of great significance for identifying the core temperature characteristics of icicles. The average temperature value provides the overall temperature level of the region, facilitating macroscopic comparison with the ambient temperature. The temperature variance characterizes the uniformity of temperature distribution within the region; for example, a uniform icicle area may have a low temperature variance. The extraction of these statistical features provides a rich data foundation for subsequent refined judgment.

[0056] In this embodiment, secondly, based on the aforementioned temperature statistical characteristic data, when the lowest temperature value of a certain area is lower than a preset deep-space temperature threshold, that area is determined to be a sky background interference area. In open-air bridge scenarios, especially at night or in high-altitude areas, infrared thermal imagers may capture extremely low deep-space background temperatures, which may be misidentified as icicles. The preset deep-space temperature threshold is an empirical value or a value determined through calibration, used to distinguish between true icicles (typically with temperatures near or slightly below 0°C) and the extremely cold deep-space background. By setting this threshold, misjudgments caused by deep-space background can be effectively eliminated, improving the accuracy of detection.

[0057] In this embodiment, again, based on temperature statistical characteristic data and ambient temperature data, the average temperature value of each region is compared with the ambient temperature data to determine whether it conforms to the temperature characteristics of an ice-water mixture. As an ice-water mixture, icicles typically exhibit the characteristic that their temperature approaches or remains around 0°C when the ambient temperature changes. By comparing the average temperature of a suspected region with the real-time acquired ambient temperature, it can be determined whether the temperature behavior of that region conforms to the physical characteristics of an ice-water mixture. For example, if the average temperature of a region remains near 0°C when the ambient temperature is above 0°C, it is highly suspected to be an icicle; conversely, if its temperature remains consistent with the ambient temperature, it may be another object. This difference comparison mechanism utilizes the unique thermophysical properties of icicles, further enhancing the reliability of the identification.

[0058] In this embodiment, finally, based on the results of the sky background interference region determination and the temperature characteristics determination of the ice-water mixture, regions that meet the temperature characteristics of icicles are selected to obtain the outdoor scene temperature analysis results data. This step is a comprehensive application of the two aforementioned determinations. First, regions that have been determined to be sky background interference are removed from the suspected icicle regions. Then, among the remaining regions, those regions that are determined to meet the temperature characteristics of ice-water mixtures after comparison with the ambient temperature are further selected. Through this multi-condition screening, common non-icicle interference in outdoor bridge scenes can be effectively eliminated, thereby obtaining more accurate outdoor scene temperature analysis results data, providing high-quality input for subsequent deep learning verification.

[0059] In this embodiment, multidimensional temperature statistical characteristics are calculated, and a preset deep-space temperature threshold is used to effectively filter out background interference from the sky. Simultaneously, the temperature characteristics of the ice-water mixture are compared with the ambient temperature to further distinguish real icicles from other low-temperature objects. This multi-layered and refined temperature analysis method improves the anti-interference capability and accuracy of icicle detection in complex open-air environments, effectively reduces the false alarm rate, and ensures the reliability of icicle detection results, thus providing a more solid guarantee for train operation safety.

[0060] In one feasible implementation, the tunnel scene temperature analysis process includes: calculating the temperature variance value of each region and the temperature variance value of the surrounding background based on the temperature data of each region to obtain temperature variance feature data; based on the temperature variance feature data, when the temperature variance value of the surrounding background of a certain region is less than a preset background variance threshold and there is a temperature gradient difference between the region and the background, it is determined to be a highly suspected icicle region; based on the temperature data of each region and the ambient temperature data, calculating the difference between the average temperature value of each region and the ambient temperature data, and determining whether it meets the icicle temperature characteristics; based on the highly suspected icicle region determination result and the icicle temperature characteristic determination result, filtering out regions that meet the icicle temperature characteristics to obtain tunnel scene temperature analysis result data.

[0061] In this embodiment, the temperature variance value of each region and the temperature variance value of the surrounding background are calculated to obtain temperature variance feature data, which aims to assist in identification by quantifying the thermal uniformity within and around the region. The temperature variance value reflects the uniformity or fluctuation of the temperature distribution within the region. Icicles typically have relatively uniform low temperatures, while their surrounding background (such as tunnel walls) may have uneven temperature distribution due to factors such as material, lighting, and reflection. Calculating the temperature variance value of each suspected icicle region helps to assess the consistency of its internal temperature. At the same time, calculating the temperature variance value of the surrounding background helps to assess the stability of the background. For each second suspected icicle region data, it can first be mapped back to the original infrared thermal image data to obtain the temperature values ​​of all pixels within the region, and the variance of these pixel temperature values ​​can be calculated using statistical methods. For the surrounding background, a ring-shaped or rectangular neighborhood around the suspected icicle region can be defined, the pixel temperature values ​​within the neighborhood can be obtained, and their variance can be calculated.

[0062] Based on this, when the variance of the surrounding background temperature in a certain area is less than a preset background variance threshold and there is a temperature gradient difference between that area and the background, it is determined to be a highly suspected icicle area. This judgment logic aims to utilize the typical thermal differences between icicles and the background in the tunnel. Icicles are usually relatively independent cold sources, with low and relatively uniform temperatures, while the adjacent tunnel wall background may also be relatively uniform in some cases. When the background temperature variance is low, it indicates that the background thermal environment is stable; at this time, if there is a temperature gradient difference between the suspected icicle area and this stable background (i.e., the suspected icicle area is significantly colder than the background), then the area is very likely to be an icicle. The preset background variance threshold can be statistically analyzed and empirically set based on infrared thermal image data of the actual tunnel environment. The temperature gradient difference can be quantified by comparing the average temperature of the suspected icicle area with the average temperature of the surrounding background area.

[0063] In this embodiment, the difference between the average temperature value of each region and the ambient temperature data is calculated to determine whether it conforms to the temperature characteristics of icicles. The formation and existence of icicles are closely related to the ambient temperature, which is usually around 0°C or slightly lower than the ambient temperature (when the ambient temperature is below freezing). By comparing the average temperature of suspected icicle areas with the real-time acquired ambient temperature data, it can be further verified whether the area has the typical temperature characteristics of icicles. For each area determined to be highly suspected of being an icicle, the average temperature value of all pixels within it is calculated. Then, this average value is subtracted from the ambient temperature data acquired in real time by the vehicle system to obtain the temperature difference. Based on the physical properties of ice, a temperature range or a range of differences from the ambient temperature can be set. If the difference falls within this range, it is considered to conform to the temperature characteristics of icicles.

[0064] In this embodiment, based on the results of the highly suspected icicle region determination and the icicle temperature characteristic determination, regions that meet the icicle temperature characteristics are selected to obtain the tunnel scene temperature analysis results. This step is the final logical integration, which combines the two independent determination results to ensure that only regions that simultaneously meet the structural thermal characteristics and actual temperature characteristics are confirmed as icicles. This is a logical "AND" operation; only when a region is both determined to be a "highly suspected icicle region" and determined to "meet the icicle temperature characteristics" is it selected as the final result of the tunnel scene temperature analysis.

[0065] In this embodiment, by calculating the temperature variance of the suspected area and its surrounding background, and combining this with the temperature gradient difference between the area and the background, icicles with independent thermal characteristics can be effectively distinguished from the complex tunnel thermal environment, avoiding misidentification of hot spots or reflections on the tunnel wall as icicles. Simultaneously, comparing the average temperature of the suspected area with the ambient temperature further verifies whether it possesses the typical temperature characteristics of an ice-water mixture, thus eliminating interference from other low-temperature objects or thermal noise. This multi-dimensional, scene-adaptive temperature analysis method improves the accuracy and robustness of icicle detection in tunnels, reduces false alarm and false negative rates, provides more reliable input for subsequent deep learning verification, and ultimately ensures train operation safety.

[0066] In one feasible implementation, the steps of inputting the data of the third suspected icicle region into a pre-trained deep learning model for verification to obtain icicle detection result data, and generating a graded alarm signal based on the icicle detection result data, include: fusing the visible light image sub-region and the infrared thermal image sub-region corresponding to the data of the third suspected icicle region to obtain multimodal input data; inputting the multimodal input data into a pre-trained target detection neural network model for deep learning verification to obtain deep learning verification result data; filtering out false targets with similar shapes but inconsistent texture features based on the deep learning verification result data to obtain icicle detection result data; obtaining the location information and confidence information of the icicle in the image based on the icicle detection result data; generating a graded alarm signal based on the location information and confidence information of the icicle in the image; and pushing the graded alarm signal to the driver or a ground data center in real time through an in-vehicle terminal.

[0067] In this embodiment, the visible light image sub-region and the infrared thermal image sub-region corresponding to the third suspected icicle region data are fused to obtain multimodal input data. Specifically, the visible light image provides rich texture, color, and shape details, while the infrared thermal image provides precise temperature distribution information of the target region. By fusing these two different modalities of data, for example, by stacking the RGB channels of the visible light image and the grayscale channels of the infrared thermal image to form multi-channel input data, more comprehensive and discriminative information can be provided to the subsequent deep learning model, thereby enhancing the model's ability to understand icicle features.

[0068] In this embodiment, the multimodal input data is fed into a pre-trained object detection neural network model for deep learning verification, yielding deep learning verification result data. This object detection neural network model can be based on various advanced architectures of convolutional neural networks (CNNs), such as YOLO (You Only Look Once), Faster R-CNN, or SSD (Single Shot MultiBox Detector). This model has been thoroughly trained on a multimodal dataset containing a large number of real icicles and various interference objects, enabling it to learn and recognize the complex feature patterns of icicles. Verification using a deep learning model leverages its powerful feature extraction and classification capabilities to perform deeper analysis and judgment on the initially screened suspected areas.

[0069] Based on this, and using the deep learning verification results, false targets with similar shapes but inconsistent texture features are filtered out to obtain icicle detection results. Deep learning models may still make some misjudgments in complex scenes. This step aims to further refine the model's output. For example, a threshold can be set based on the confidence score of the model output; detection results below this threshold will be considered low reliability and discarded. Furthermore, prior knowledge about icicles can be incorporated; for example, icicles typically have relatively smooth or specific ice crystal textures, while water droplets or background debris may have different texture features. By analyzing the feature maps or feature vectors output by the model, false targets that are similar in shape to icicles but do not conform to the texture details of icicles can be identified and filtered out.

[0070] Furthermore, based on the icicle detection results, the location information and confidence information of the icicle in the image are obtained. Location information is typically represented by a bounding box, i.e., the pixel coordinate range of the icicle in the image (such as the coordinates of the top left and bottom right corners), or the coordinates of its center point. Confidence information represents the probability assessment by the deep learning model that the detection result is a real icicle, reflecting the reliability of the detection. This information is crucial for subsequent alarm signal generation and spatial localization.

[0071] In this embodiment, a tiered alarm signal is generated based on the location and confidence level of the icicle in the image. This tiered alarm mechanism allows the system to provide different levels of warning based on the potential hazard of the icicle and the reliability of the detection results. For example, an emergency alarm can be generated for icicles with high confidence level and located in critical areas (such as near overhead contact lines); a warning signal can be generated for icicles with relatively high confidence level but slightly lower hazard; and a prompt signal is generated for icicles with moderate confidence level or located in non-critical areas. This tiered processing helps avoid over-alarming while ensuring timely response to high-risk situations.

[0072] In this embodiment, the graded alarm signals are finally pushed to the driver and passengers or a ground data center in real time via an onboard terminal. The onboard terminal can be a display screen in the train driver's cab, an audible and visual alarm, or an onboard computer system. The real-time push mechanism rapidly transmits the alarm information to the driver and passengers or a remote ground data center via a wireless communication network (such as 5G, 4G, or a dedicated wireless network). This ensures that relevant personnel can obtain the icicle detection results immediately, thereby taking necessary countermeasures in a timely manner and ensuring train operation safety.

[0073] In this embodiment, based on the preliminary screening of the third suspected icicle region data, this application further introduces a multimodal deep learning verification mechanism through the above-described technical solution. This mechanism, by fusing complementary information from visible light and infrared thermal images, provides richer and more comprehensive input to the deep learning model. The pre-trained target detection neural network model can learn and recognize the complex feature patterns of icicles, thereby performing high-precision verification of suspected areas, effectively distinguishing real icicles from false targets with similar shapes but inconsistent texture features, improving the accuracy and robustness of icicle detection, and significantly reducing the false alarm rate. Furthermore, based on the location and confidence information of the icicles, a graded alarm signal is generated and pushed to the driver or ground data center in real time. This allows the alarm information to be intelligently graded according to the actual degree of danger and detection reliability, avoiding over-alarms or under-alarms, ensuring the timely transmission of critical information, and thus providing safer and more reliable icicle detection and early warning protection for train operation.

[0074] In one feasible implementation, the step of generating a graded alarm signal based on the position information and confidence information of the icicle in an image includes: converting the position information of the icicle in the image into the actual spatial position information of the icicle based on the position information of the icicle in the image and a pre-established coordinate mapping relationship; determining whether the actual spatial position of the icicle exceeds the boundary of the safety zone based on the actual spatial position information of the icicle and preset data of the contact wire safety zone, thereby obtaining an intrusion judgment result; calculating the distance between the actual spatial position of the icicle and the contact wire centerline based on the actual spatial position information of the icicle and the contact wire centerline position data, thereby determining whether the icicle is located directly above the contact wire, thereby obtaining a direct-above judgment result; determining the detection reliability level based on the confidence information; and generating a graded alarm signal based on the intrusion judgment result, the direct-above judgment result, and the detection reliability level; the graded alarm signal includes an emergency alarm signal, a warning signal, and a prompt signal.

[0075] In this embodiment, based on the positional information of the icicle in the image and a pre-established coordinate mapping relationship, the positional information of the icicle in the image is converted into the actual spatial position information of the icicle. This conversion process typically involves precise camera calibration of the vehicle-mounted infrared thermal imager and the vehicle-mounted visible light camera to obtain their internal and external parameters. Using these parameters, combined with the real-time position and attitude information of the train, the two-dimensional pixel coordinates in the image can be accurately mapped to a three-dimensional actual spatial coordinate system, thereby determining the precise physical position of the icicle in the railway environment. For example, monocular or binocular visual ranging technology can be used, combined with the train's own positioning system (such as GNSS, odometer, etc.), to project the position of the icicle in the image coordinate system to the world coordinate system or the train coordinate system.

[0076] In this embodiment, based on the actual spatial location information of the icicle and the preset safety zone data of the overhead contact line, it is determined whether the actual spatial location of the icicle exceeds the boundary of the safety zone, thus obtaining the intrusion judgment result. The preset safety zone data of the overhead contact line typically defines a three-dimensional safety envelope or geometric model around the working area of ​​the overhead contact line and pantograph. Any object encroaching on this area is considered a potential threat. This preset data can be stored as a three-dimensional point cloud, a CAD model, or geometric boundary parameters. By comparing the actual spatial location of the icicle with these preset safety boundaries, it can be determined whether the icicle poses an intrusion risk to train operation.

[0077] In this embodiment, based on the actual spatial location information of the icicle and the position data of the contact wire centerline, the distance between the actual spatial location of the icicle and the contact wire centerline is calculated to determine whether the icicle is located directly above the contact wire, thus obtaining a result indicating that it is directly above the contact wire. The contact wire centerline position data can be obtained from a line geographic information database or high-precision measurement data, and is usually represented in the form of a three-dimensional coordinate sequence or a parametric curve. By calculating the spatial distance between the icicle and the contact wire centerline, and combining this with the vertical height of the icicle, it is possible to accurately determine whether the icicle is suspended directly above the contact wire, because the icicle at this location poses the highest risk of impact to the pantograph.

[0078] In this embodiment, the detection reliability level is determined based on the confidence information. The confidence information is typically the probability value output by a deep learning model, reflecting the model's certainty about the detection result. This confidence value can be divided into different levels; for example, a confidence level above 0.9 is considered high reliability, between 0.7 and 0.9 is considered medium reliability, and between 0.5 and 0.7 is considered low reliability, thus quantifying the reliability level of the detection result. Finally, based on the intrusion determination result, the direct overhead determination result, and the detection reliability level, a graded alarm signal is generated; the graded alarm signal includes an emergency alarm signal, a warning signal, and a prompt signal. For example, when an icicle is determined to be an intrusion and located directly above the overhead contact line, and the detection reliability level is high, the system will generate an emergency alarm signal, requiring immediate action; when the icicle has a potential intrusion risk or the detection reliability is medium, a warning signal is generated to remind drivers and passengers; and for icicles with low detection reliability or that do not pose a direct threat, a prompt signal is generated for subsequent observation or data recording.

[0079] In this embodiment, by converting the position information of the icicle in the image into its actual spatial position information, and combining this with preset data on the safety zone of the overhead contact system and the position data of the centerline of the overhead contact system for encroachment and direct overhead detection, the potential threat of the icicle to train operation can be accurately assessed. Simultaneously, the introduction of detection reliability levels ensures that alarm issuance considers not only the presence of the icicle but also the accuracy of the detection results. This multi-dimensional and refined assessment mechanism enables the system to generate graded alarm signals, including emergency alarm signals, warning signals, and alert signals. This effectively solves the problem that a single alarm signal cannot distinguish the degree of icicle hazard, avoiding unnecessary emergency stops or missed reports of high-risk icicles. Drivers or ground data centers can quickly and accurately determine the hazard level of the icicle based on different alarm signal levels, thereby taking targeted countermeasures, improving the safety and operational efficiency of railway operations, and reducing potential accident risks.

[0080] In the embodiments of this application, the scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring activates the detection module by acquiring ambient temperature data, identifying scene types, processing image data, analyzing features, and verifying the model to generate alarm signals. This achieves scene-adaptive icicle detection under vehicle-mounted mobile monitoring, solves the problem that existing technologies cannot identify icicle hazards in real time, and can improve the safety and efficiency of railway operations.

[0081] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the scene adaptive icicle detection method based on vehicle-mounted mobile monitoring in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0082] This application also provides a scene-adaptive icicle detection system based on vehicle-mounted mobile monitoring, referenced... Figure 2 The scene-adaptive icicle detection system based on vehicle-mounted mobile monitoring includes: a memory 10, a processor 20, and a scene-adaptive icicle detection program based on vehicle-mounted mobile monitoring stored on the memory 10 and executable on the processor 20. The scene-adaptive icicle detection program based on vehicle-mounted mobile monitoring is configured to implement the steps of the scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring.

[0083] The scene-adaptive icicle detection system based on vehicle-mounted motion monitoring provided in this application adopts the scene-adaptive icicle detection method based on vehicle-mounted motion monitoring in the above embodiments, which can improve the accuracy and real-time performance of icicle detection. Compared with the prior art, the beneficial effects of the scene-adaptive icicle detection system based on vehicle-mounted motion monitoring provided in this application are the same as those of the scene-adaptive icicle detection method based on vehicle-mounted motion monitoring provided in the above embodiments, and other technical features of the scene-adaptive icicle detection system based on vehicle-mounted motion monitoring are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0084] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0085] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. All equivalent structural transformations made under the technical concept of this application using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the scope of patent protection of this application.

Claims

1. A scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring, characterized in that, The method includes: The system acquires ambient temperature data. When the ambient temperature data is lower than a preset temperature threshold, the system activates the icicle detection module and acquires infrared thermal image data collected by the vehicle-mounted infrared thermal imager and visible light image data collected by the vehicle-mounted visible light camera. Based on the visible light image data, the scene type of the current train is identified, and the scene identification result is obtained; the scene type includes tunnel scene, tunnel entrance scene and open bridge scene. Based on the infrared thermal image data, grayscale histogram statistics and adaptive threshold segmentation are performed to obtain the data of the first suspected icicle region. Morphological and brightness feature analyses were performed on the data of the first suspected icicle region to obtain the data of the second suspected icicle region. Based on the scene recognition results and the data of the second suspected icicle region, a multidimensional temperature statistical analysis is performed on the data of the second suspected icicle region to obtain the data of the third suspected icicle region. The data of the third suspected icicle region is input into a pre-trained deep learning model for verification to obtain icicle detection result data, and a graded alarm signal is generated based on the icicle detection result data.

2. The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in claim 1, characterized in that, The steps for identifying the scene type of the current train based on the visible light image data and obtaining the scene identification result include: Based on the visible light image data, calculate the average brightness change rate data and image texture feature data; Obtain train location information data; Based on the image average brightness change rate data, the image texture feature data, and the train position information data, the current scene type is determined through state machine logic to obtain the scene recognition result.

3. The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in claim 2, characterized in that, Based on the image average brightness change rate data, the image texture feature data, and the train position information data, the steps for determining the current scene type and obtaining the scene recognition result through state machine logic include: Based on the average brightness change rate data of the image, when the brightness value is detected to decrease by more than the first preset threshold within a preset time window, it is determined to be a scene of entering a tunnel entrance. Based on the image texture feature data, when the image texture complexity is detected to be lower than the second preset threshold, it is determined to be a scene inside a tunnel; Based on the train location information data and combined with a pre-stored line geographic information database, the preset scenario type corresponding to the current train location is determined. Based on the scene determination results at the tunnel entrance, the scene determination results inside the tunnel, and the preset scene type, the final scene recognition result is determined through state machine state transition logic. Based on the scene recognition results, the corresponding detection parameter set is dynamically loaded from the preset parameter database; the detection parameter set includes infrared preprocessing parameters, temperature analysis thresholds and interference filtering rules for different scenes.

4. The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in claim 1, characterized in that, The steps for performing morphological and brightness feature analysis on the data of the first suspected icicle region to obtain the data of the second suspected icicle region include: Contour feature calculation is performed on each of the first suspected icicle regions to obtain contour feature data; Based on the contour feature data, regions with inverted triangle features (wider at the top and narrower at the bottom) or elongated features are selected to obtain morphological screening result data. Calculate the average brightness value of the region corresponding to each of the morphological screening results to obtain the region average brightness data; Based on the average brightness data of the region, low-brightness feature areas are retained and high-brightness noise areas are removed to obtain the second suspected icicle region data.

5. The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in claim 1, characterized in that, Based on the scene recognition results and the data of the second suspected icicle region, the steps for performing multidimensional temperature statistical analysis on the data of the second suspected icicle region to obtain the data of the third suspected icicle region include: Map the data of the second suspected icicle region back to the original infrared temperature data matrix to obtain the temperature data of each region; Based on the scene recognition results and the temperature data of each region, when the scene is an open-air bridge scene, the open-air bridge scene temperature analysis process is executed; when the scene is a tunnel scene, the tunnel scene temperature analysis process is executed, and the temperature analysis result data is obtained. Based on the temperature analysis results, interference areas were eliminated to obtain the data for the third suspected icicle region.

6. The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in claim 5, characterized in that, The temperature analysis process for the open-air bridge scene includes: Based on the temperature data of each region, the minimum temperature value, average temperature value and temperature variance value of each region are calculated to obtain temperature statistical characteristic data. Based on the temperature statistical feature data, when the lowest temperature value of a certain area is lower than the preset deep space temperature threshold, the area is determined to be a sky background interference area. Based on the temperature statistical feature data and the ambient temperature data, the average temperature value of each region is compared with the ambient temperature data to determine whether it conforms to the temperature characteristics of an ice-water mixture. Based on the results of the sky background interference area determination and the results of the ice-water mixture temperature characteristic determination, areas that meet the temperature characteristics of ice columns are selected to obtain the outdoor scene temperature analysis results data.

7. The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in claim 5, characterized in that, The temperature analysis process inside the tunnel includes: Based on the temperature data of each region, the temperature variance value of each region and the temperature variance value of the surrounding background are calculated to obtain temperature variance feature data. Based on the temperature variance feature data, when the temperature variance value of the surrounding background of a certain area is less than the preset background variance threshold and there is a temperature gradient difference between the area and the background, it is determined to be a highly suspected icicle area. Based on the temperature data of each region and the ambient temperature data, the difference between the average temperature value of each region and the ambient temperature data is calculated to determine whether it conforms to the temperature characteristics of ice columns. Based on the results of the highly suspected icicle area determination and the results of the icicle temperature characteristic determination, areas that meet the icicle temperature characteristics are selected to obtain tunnel scene temperature analysis results data.

8. The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in claim 1, characterized in that, The steps of inputting the data of the third suspected icicle region into a pre-trained deep learning model for verification to obtain icicle detection result data, and generating a graded alarm signal based on the icicle detection result data, include: The visible light image sub-region and the infrared thermal image sub-region corresponding to the third suspected icicle region data are fused to obtain multimodal input data; The multimodal input data is input into a pre-trained target detection neural network model for deep learning verification to obtain deep learning verification result data; Based on the deep learning verification results data, false targets with similar shapes but inconsistent texture features are filtered out to obtain icicle detection results data; Based on the icicle detection results, the location information and confidence information of the icicle in the image are obtained; Based on the location and confidence information of the icicles in the image, a graded alarm signal is generated; The graded alarm signals are pushed to the driver and passengers or the ground data center in real time through the vehicle terminal.

9. The scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in claim 8, characterized in that, The steps for generating graded alarm signals based on the location and confidence level of the icicles in the image include: Based on the position information of the icicle in the image and the pre-established coordinate mapping relationship, the position information of the icicle in the image is converted into the actual spatial position information of the icicle. Based on the actual spatial location information of the icicle and the preset data of the contact wire safety area, it is determined whether the actual spatial location of the icicle exceeds the boundary of the safety area, and the intrusion judgment result is obtained. Based on the actual spatial location information of the icicle and the location data of the center line of the contact network, the distance between the actual spatial location of the icicle and the center line of the contact network is calculated, and it is determined whether the icicle is located directly above the contact network, thus obtaining the result of the direct-above determination. Based on the confidence level information, the detection reliability level is determined; Based on the intrusion limit judgment result, the direct upward judgment result, and the detection reliability level, a graded alarm signal is generated; the graded alarm signal includes an emergency alarm signal, a warning signal, and a prompt signal.

10. A scene-adaptive icicle detection system based on vehicle-mounted mobile monitoring, characterized in that, The scene-adaptive icicle detection system based on vehicle-mounted mobile monitoring includes: a memory, a processor, and a scene-adaptive icicle detection program based on vehicle-mounted mobile monitoring stored in the memory and executable on the processor. The scene-adaptive icicle detection program based on vehicle-mounted mobile monitoring is configured to implement the steps of the scene-adaptive icicle detection method based on vehicle-mounted mobile monitoring as described in any one of claims 1 to 9.

Images