Railway train detection method and system based on reference frame difference and frequency domain energy

Abstract

The application discloses a railway train detection method and system based on reference frame difference and frequency domain energy, and belongs to the technical fields of computer vision, video image analysis and railway intelligent monitoring. The method comprises the following steps: S1, collecting images of a normal scene without a train, and pre-processing the collected images to establish a reference frame representing a stable background state; S2, obtaining a current image to be detected, and performing difference operation on the image to be detected and the reference frame to obtain a difference image; S3, performing frequency domain transformation on the difference image, extracting frequency domain features and calculating frequency domain energy values; and S4, adopting a hierarchical double-threshold judgment mechanism composed of a train event alarm threshold and a reference frame update threshold, and comparing the frequency domain energy values with a preset train event alarm threshold. The railway train detection method and system based on reference frame difference and frequency domain energy can reduce false alarms of static similar objects, have low computing power requirements and are suitable for edge deployment, can adapt to environmental changes, and can guarantee long-term stable operation.

Description

Technical Field

[0001] This invention relates to the fields of computer vision, video image analysis and intelligent railway monitoring technology, and in particular to a railway train detection method and system based on reference frame difference and frequency domain energy. Background Technology

[0002] With the continuous increase in railway operating mileage, monitoring train traffic status along railway lines and detecting entry and exit at station entrances have become core aspects of railway security and maintenance management. Current mainstream train video detection solutions mostly employ deep learning-based object detection models (such as YOLO), which identify and determine trains by extracting their appearance features. These methods rely on appearance information such as contours, textures, and colors in static images. When static objects similar in shape to trains exist in the monitored scene, false identification and false alarms are easily triggered. Furthermore, the railway environment is complex and variable; various building structures, obstructions, reflective areas, and complex natural backgrounds may appear within the camera's field of view. These interfering scenarios are difficult to fully cover in the model training data, resulting in a high false alarm rate for object detection models in actual deployment, making it difficult to meet the requirements for long-term stable operation.

[0003] Furthermore, to ensure detection accuracy, target detection models typically have a large number of parameters and high computational complexity. However, most railway monitoring equipment is an edge device with limited computing power, storage, and power consumption resources, making it difficult to support real-time inference of large-scale deep learning models. At the same time, detection models with fixed parameters are not adaptable to long-term environmental changes such as changes in lighting, weather, slow changes in background facilities, and slight camera drift. They are prone to performance degradation as operating time increases, requiring frequent manual maintenance and calibration.

[0004] Some traditional solutions attempt to detect moving targets using simple inter-frame difference methods, but these methods only determine pixel changes in the spatial domain and are easily affected by local noise, instantaneous fluctuations in illumination, and scattered interference, resulting in poor detection stability. Moreover, most solutions lack a stable and effective background adaptive update mechanism, which cannot meet the needs of long-term uninterrupted operation along railway lines and is difficult to implement in practical engineering. Summary of the Invention

[0005] The purpose of this invention is to provide a railway train detection method and system based on reference frame difference and frequency domain energy, thereby solving the above-mentioned technical problems.

[0006] To achieve the above objectives, this invention provides a railway train detection method based on reference frame difference and frequency domain energy, comprising the following steps: Includes the following steps: S1. Acquire images of a normal scene without trains, and preprocess the acquired images to establish a reference frame representing a stable background state. S2. Obtain the current image to be detected, and perform a difference operation between the image to be detected and the reference frame to obtain the difference image; S3. Perform frequency domain transformation on the difference image, extract frequency domain features, and calculate frequency domain energy values; S4. A hierarchical dual-threshold determination mechanism consisting of a train event alarm threshold and a reference frame update threshold is adopted. The frequency domain energy value is compared with the preset train event alarm threshold. When the frequency domain energy value is not less than the alarm threshold, it is determined that a train entering or leaving event exists. When the frequency domain energy value is lower than the preset reference frame update threshold for a consecutive preset number of frames, the current image is used to update the reference frame to achieve environmental adaptive adjustment.

[0007] Preferably, the preprocessing in S1 is one or more of the following: grayscale conversion, size normalization, region of interest cropping, and brightness normalization.

[0008] Preferably, the difference operation in S2 is the absolute value difference operation between the corresponding pixels of the current image and the reference frame, and the calculation formula is: ; In the formula, Indicates time At the current frame in coordinates Pixel value at; Indicates the reference frame in coordinates Pixel value at; Indicates the difference image in coordinates The difference value at that point.

[0009] Preferably, the frequency domain transformation in S3 uses Fast Fourier Transform (FFT) to transform the difference image from the spatial domain to the frequency domain, obtaining the frequency domain coefficient matrix. ,in Represents frequency domain coordinates; The frequency domain energy value is calculated based on the sum of squares of the amplitudes of the frequency domain coefficients, the sum of the absolute values ​​of the amplitudes, or the weighted summation result; the formula for calculating the frequency domain energy value using the sum of squares of the amplitudes is as follows: ; In the formula, Represents frequency domain coordinates The amplitude at that point; This represents the frequency domain energy value corresponding to the current difference image.

[0010] Preferably, in S4, the dual thresholds are divided into high and low levels based on function and value. The high threshold is the train event alarm threshold, denoted as... This is used to determine the triggering of train entry and departure events. The low threshold is the baseline frame update threshold, denoted as... This is used to determine the timing of the baseline frame update, and both conditions must be met. Numerical relationships; When satisfied When a significant dynamic change occurs in the scene, the detection of a train entering or leaving the scene is triggered.

[0011] Preferably, in S4, when the frequency domain energy value continuous The frame is below the baseline frame update threshold When the current scene is determined to have returned to a stable background state, a baseline frame update is performed; whereby... The preset consecutive frame counting threshold is set to 10 to 30 frames, preferably 15 frames.

[0012] Preferably, when performing a reference frame update, the current frame directly replaces the original reference frame, or a weighted fusion method is used to smoothly update the original reference frame; the updated reference frame is used to adapt to long-term operating environments with changes in illumination, weather, and slow changes in background facilities.

[0013] A system for railway train detection based on reference frame difference and frequency domain energy includes: The image acquisition module is used to acquire surveillance images along the railway line; The reference frame management module is used to establish reference frames in train-free scenarios and update the reference frames when update conditions are met. The difference processing module is used to perform difference operations between the current image and the reference frame to obtain a difference image; The frequency domain analysis module is used to perform frequency domain transformation on the difference image, extract frequency domain features, and calculate frequency domain energy values; The decision module is used to compare the frequency domain energy value with the train event alarm threshold and the reference frame update threshold, and output the decision result of the train entering event, the train leaving event, or the background stable state. The output module is used to output detection results or send alarm information to the upper-level system.

[0014] Preferably, the monitoring images are acquired by cameras along the railway line. These cameras are deployed along the railway line area to acquire continuous images and transmit them to the image acquisition module.

[0015] Preferably, it also includes an alarm module, which is used to issue an alarm signal or record event information when the judgment module determines that a train entry event or a train departure event has occurred.

[0016] Therefore, the railway train detection method and system based on reference frame difference and frequency domain energy described above have the following beneficial effects: 1. Low false alarm rate: This method does not rely on the recognition of train appearance features, but achieves detection through the overall dynamic changes of the scene caused by the train entering and leaving. In principle, it avoids the problem of false recognition caused by static similar objects and is suitable for complex background scenes along railway lines.

[0017] 2. Adaptable to edge deployment: The core consists of inter-frame difference, fast Fourier transform and energy calculation. It does not require a deep learning model with a large number of parameters, has low computational load and low resource consumption, and can realize real-time detection on low computing power edge devices along railway lines.

[0018] 3. Strong long-term environmental adaptability: Through the reference frame update mechanism of "dual threshold + continuous frame judgment", the reference background can be automatically updated with the slow changes of lighting, weather, background and slight camera drift, ensuring the stability of the method to run continuously for a long time and reducing operation and maintenance costs.

[0019] 4. Strong system compatibility: It can be used independently as a train entry and exit detection module, or as a front-end trigger module for existing monitoring and alarm systems and upper-level analysis systems. It does not require reconstruction of the original system architecture and is easy to implement and deploy.

[0020] 5. Excellent real-time performance: The algorithm logic is simple and efficient, and it can process video streams frame by frame on a conventional edge computing platform, which fully meets the real-time requirements of railway line monitoring.

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 This is an overall flowchart of the railway train detection method based on reference frame difference and frequency domain energy of the present invention; Figure 2 This is a framework diagram of the railway train detection system based on reference frame difference and frequency domain energy according to the present invention; Figure 3 This is a frequency domain energy variation curve corresponding to the video sequence acquired in this invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0024] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0025] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] like Figures 1-3 As shown, this invention provides a railway train detection method and system based on reference frame difference and frequency domain energy. Unlike existing YOLO-based target detection methods, this invention does not rely on identifying the static appearance features of the train for judgment, but rather utilizes the overall dynamic changes caused by the train entering and leaving the monitoring scene for detection, thereby reducing false alarms caused by statically similar objects in the scene. Specifically, for fixed camera scenarios along railway lines, train entry and departure events are detected through reference frame difference combined with frequency domain energy analysis, and long-term operational stability is ensured through a dual-threshold adaptive update mechanism, including the following steps: S1. Reference Frame Initialization and Image Preprocessing: Acquire normal scene images without trains passing by, and establish a reference frame representing a stable background state.

[0027] Selecting images without trains as the initial reference can maximize the highlighting of scene changes caused by trains entering and leaving, providing a stable reference for subsequent differential detection and avoiding detection failure caused by the initial reference containing train targets.

[0028] Before establishing the baseline frame, the acquired images are preprocessed, including one or more of the following: grayscale conversion, size normalization, region of interest (ROI) cropping, and brightness normalization. Grayscale conversion transforms the three-channel image into a single channel, significantly reducing the amount of data required for subsequent computations. ROI cropping removes irrelevant background areas outside the railway line, focusing on the core monitoring area, further reducing computation and minimizing irrelevant interference. Brightness normalization counteracts the effects of instantaneous fluctuations in illumination, improving the environmental adaptability of the detection method.

[0029] S2. Obtain the difference image by inter-frame difference operation: Obtain the image to be detected at the current time, perform difference operation between the image to be detected and the reference frame to obtain the difference image.

[0030] Specifically, the absolute value difference operation of the corresponding pixels is used, and the calculation formula is as follows: ; In the formula, Indicates time At the current frame in coordinates Pixel value at; Indicates the reference frame in coordinates Pixel value at; Indicates the difference image in coordinates The difference value at each location indicates a more significant change in the scene. By performing absolute value difference calculations, the pixel change area of ​​the current scene relative to the reference background can be directly extracted, intuitively representing the degree of scene change. This calculation does not rely on train appearance features for matching and recognition, thus avoiding the false recognition problem caused by static similar objects in principle and effectively reducing the false alarm rate.

[0031] S3. Frequency Domain Transformation and Frequency Domain Energy Value Calculation: Perform frequency domain transformation on the difference image, extract frequency domain features, and calculate the frequency domain energy value.

[0032] Specifically, the Fast Fourier Transform (FFT) is used to perform the frequency domain transformation, converting the difference image in the spatial domain to the frequency domain, thus obtaining the frequency domain coefficient matrix. ,in Represents frequency domain coordinates; frequency domain energy values ​​are calculated based on the frequency domain coefficient matrix, and calculation methods such as sum of squared amplitudes, sum of absolute amplitudes, or weighted accumulation can be selected; among these, the sum of squared amplitudes is preferred, and the formula is: ; In the formula, Represents frequency domain coordinates The amplitude at that point; This represents the frequency domain energy value corresponding to the current differential image. It is used to quantify the overall change intensity of the scene from a global perspective. Compared with the method of directly summing the differential pixels in the spatial domain, the frequency domain energy can characterize scene changes from a global frequency perspective. It has stronger robustness to local noise, small light jitter, and scattered interference points. It can effectively filter out small disturbances that are not at the train level, avoid sporadic interference triggering false alarms, and improve the stability of detection.

[0033] S4. Dual-threshold event determination and adaptive update of reference frame: A hierarchical dual-threshold determination mechanism is adopted to compare the frequency domain energy value with the preset high and low thresholds respectively, and to simultaneously complete the train event determination and reference frame update control.

[0034] (1) Determination of train entry and departure events The high threshold is the train event alarm threshold, denoted as: This is specifically used for determining the trigger of train events. When the conditions are met... When a significant dynamic change occurs in the scene, the system detects the arrival or departure of a train. Setting a high train event alarm threshold filters out minor background fluctuations and non-train-level disturbances, preventing false alarms. Both train arrival and departure cause significant changes in the scene relative to the baseline background; therefore, both types of events can be detected using the same energy threshold. The system can further differentiate between arrival and departure states by combining energy change trends across consecutive frames. After determining the event, the system outputs the detection results to the upper-level system, monitoring platform, or local alarm module via the output module.

[0035] (2) Adaptive update of the reference frame The low threshold is the baseline frame update threshold, denoted as . It is specifically used for determining the timing of base frame updates and satisfies... The numerical relationship forms a hierarchical judgment logic that makes event detection and background updates independent of each other. By setting an update threshold below the alarm threshold, it can be ensured that the reference frame update is only initiated after the train has completely left and the scene energy has fully dropped to a stable level. Logically, this avoids erroneous updates to the reference frame when the train has not completely left and the scene is still changing, thus ensuring detection accuracy.

[0036] When the frequency domain energy value continuous The frame is below the baseline frame update threshold When the current scene has been determined to have returned to a stable, train-free background state, a baseline frame update is performed; among which... The preset consecutive frame counting threshold ranges from 10 to 30 frames, with 15 frames being preferred. The mechanism of determining consecutive multiple frames effectively avoids erroneous updates of the reference frame caused by single-frame fluctuations, instantaneous shadows, random noise, or temporary occlusion. Combined with the dual-threshold hierarchical mechanism, this further enhances the reliability of background updates. When updating the reference frame, the current frame can directly replace the original reference frame for rapid background updates; alternatively, a weighted fusion method can be used to smoothly update the original reference frame, avoiding detection fluctuations caused by abrupt background changes.

[0037] Through the aforementioned adaptive update mechanism, the reference frame can be adjusted synchronously with long-term environmental changes such as changes in lighting, weather, slow changes in background facilities, and slight camera drift, so that the method can maintain stable detection performance during long-term uninterrupted operation without the need for frequent manual maintenance and calibration.

[0038] Furthermore, the overall computation of this method only includes frame difference, FFT transformation and energy summation. It does not require the deployment of large-scale deep learning models, has low computational load and low resource consumption, and can be deployed on edge computing devices along railway lines to realize real-time detection of train entry and departure status under low computing power conditions. It is suitable for various scenarios such as railway line monitoring, station entrance monitoring, and dedicated line passage detection.

[0039] Combination Figure 3 The frequency domain energy curves can intuitively verify the detection logic of this method: when there is no train in the scene and the background is stable, the frequency domain energy value remains at a low and stable level, with only minor fluctuations caused by illumination disturbances and image noise. For minor dynamic changes such as people moving in the scene, which are not at the train level, the resulting increase in frequency domain energy is limited and far below the train event alarm threshold, thus not triggering false alarms, demonstrating the method's ability to suppress scattered interference. When the train begins to enter the monitoring area, the pixel change range of the scene relative to the reference frame increases rapidly, and the frequency domain energy value rises rapidly, forming a steep rising edge. When the train is in a large area of ​​the monitoring screen and continues to pass through the scene, the frequency domain energy value fluctuates in a high range. When the train gradually leaves the monitoring area, the scene change range gradually shrinks, and the frequency domain energy value falls back. After the train has completely left and the scene has stabilized, the frequency domain energy value returns to a low and stable state.

[0040] This invention determines train entry / departure events during the energy rise phase by using a train event alarm threshold. By combining a reference frame update threshold with a continuous frame determination mechanism, the reference frame is updated after the energy drops to a low level and remains stable. This fully covers the entire process of train "entry-passage-departure-scene recovery" detection, while ensuring the accuracy of background adaptive updates and avoiding erroneous updates when the train has not completely departed.

[0041] This invention also provides a railway train detection system that implements the above-mentioned detection method. This system automates the detection of train entry and departure events based on reference frame differential and frequency domain energy analysis. Each module works in concert to execute the steps described above. The specific composition and operation are as follows: Along-line cameras: Deployed in railway line areas, such as railway monitoring points, station entrances and exits, and dedicated line crossings, to collect continuous monitoring images of the railway line monitoring area and transmit the collected video frames to the image acquisition module in real time, providing the raw image data source for subsequent detection processes.

[0042] Image acquisition module: Receives continuous monitoring images transmitted from cameras along the line, completes image format standardization, frame buffering and preliminary verification processing, and transmits image data that meets the calculation requirements to the reference frame management module and the differential processing module in sequence to ensure stable and compliant image input in subsequent detection processes.

[0043] The reference frame management module is responsible for the initialization and adaptive update management of reference frames. In the initial stage of system startup, under normal scenarios where no trains are passing, it selects clear images to establish initial reference frames and stores them persistently. During subsequent operation, it receives background stability judgment results and update instructions output by the decision module. When the update conditions are met, it updates the stored reference frames by directly replacing the current frame or by weighted fusion, and synchronizes the latest reference frames to the differential processing module in real time, providing a reference background for the frame difference operation with a dynamically adapted environment.

[0044] Differential processing module: Receives the current image to be detected and the latest reference frame, performs absolute value difference operation on the corresponding pixels of the two to generate a differential image representing the degree of scene change, and transmits the differential image to the frequency domain analysis module; This module directly extracts the scene change area through frame difference operation, without relying on template matching based on train appearance features, reducing the risk of false alarms caused by static similar objects from the source of detection principle.

[0045] Frequency domain analysis module: After receiving the differential image, it uses Fast Fourier Transform (FFT) to convert the spatial domain differential image to the frequency domain, obtaining the corresponding frequency domain coefficient matrix; and calculates the frequency domain energy value corresponding to the current frame based on the frequency domain coefficient matrix, and transmits the calculated frequency domain energy value to the decision module in real time; this module quantifies the overall scene change intensity from a global perspective through frequency domain energy, which has stronger anti-interference ability compared with spatial domain pixel summation, and can effectively filter detection errors caused by local noise and small illumination fluctuations.

[0046] The decision module incorporates a hierarchical dual-threshold decision logic that includes a train event alarm threshold and a reference frame update threshold. After receiving the frequency domain energy value, it first compares it with the train event alarm threshold to determine whether the current frame triggers a train entry or departure event. Simultaneously, it continuously counts the number of consecutive frames with frequency domain energy values ​​lower than the reference frame update threshold. When the number of consecutive frames meets the preset threshold requirement, it outputs a background stability state and a reference frame update command. This module achieves independent control of event detection and background update through a high-low separation dual-threshold mechanism, logically avoiding false updates and false alarms.

[0047] Output module: Receives the train event judgment results and background status information output by the judgment module, uploads the detection results, event time and other data to the upper-level monitoring system and operation and maintenance management platform, or sends them to the local storage unit for data retention, so as to realize the synchronization and traceability of detection data.

[0048] Alarm module: Receives train entry / departure event trigger signals output by the decision module. When a train event is detected, it can trigger audible and visual alarm signals to remind on-site maintenance personnel, or simultaneously record information such as the event time, corresponding image frame, and energy peak value, providing event reminders and data traceability for railway operation and maintenance management and security monitoring.

[0049] The system has a lightweight overall architecture, does not require a large-scale deep learning inference framework or high-performance computing hardware, and can be directly deployed in edge computing devices along railway lines to achieve real-time train detection and adaptive background updates under low computing power conditions, making it suitable for long-term unattended operation scenarios along railway lines.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A railway train detection method based on reference frame difference and frequency domain energy, characterized in that, Includes the following steps: S1. Acquire images of a normal scene without trains, and preprocess the acquired images to establish a reference frame representing a stable background state. S2. Obtain the current image to be detected, and perform a difference operation between the image to be detected and the reference frame to obtain the difference image; S3. Perform frequency domain transformation on the difference image, extract frequency domain features, and calculate frequency domain energy values; S4. A hierarchical dual-threshold determination mechanism consisting of a train event alarm threshold and a reference frame update threshold is adopted. The frequency domain energy value is compared with the preset train event alarm threshold. When the frequency domain energy value is not less than the alarm threshold, it is determined that a train entering or leaving event exists. When the frequency domain energy value is lower than the preset reference frame update threshold for a consecutive preset number of frames, the current image is used to update the reference frame to achieve environmental adaptive adjustment.

2. The railway train detection method based on reference frame difference and frequency domain energy according to claim 1, characterized in that: The preprocessing in S1 includes one or more of the following: grayscale conversion, size normalization, region of interest cropping, and brightness normalization.

3. The railway train detection method based on reference frame difference and frequency domain energy according to claim 1, characterized in that: The difference operation in S2 is the absolute value difference operation between the corresponding pixels of the current image and the reference frame. The calculation formula is: ； In the formula, Indicates time At the current frame in coordinates Pixel value at; Indicates the reference frame in coordinates Pixel value at; Indicates the difference image in coordinates The difference value at that point.

4. The railway train detection method based on reference frame difference and frequency domain energy according to claim 1, characterized in that: The frequency domain transformation in S3 uses Fast Fourier Transform (FFT) to transform the difference image from the spatial domain to the frequency domain, obtaining the frequency domain coefficient matrix. ,in Represents frequency domain coordinates; The frequency domain energy value is calculated based on the sum of squares of the amplitudes of the frequency domain coefficients, the sum of the absolute values ​​of the amplitudes, or the weighted summation result; the formula for calculating the frequency domain energy value using the sum of squares of the amplitudes is as follows: ； In the formula, Represents frequency domain coordinates The amplitude at that point; This represents the frequency domain energy value corresponding to the current difference image.

5. The railway train detection method based on reference frame difference and frequency domain energy according to claim 4, characterized in that: In S4, the dual thresholds are divided into high and low levels based on function and value. The high threshold is the train event alarm threshold, denoted as... This is used to determine the triggering of train entry and departure events. The low threshold is the baseline frame update threshold, denoted as... This is used to determine the timing of the baseline frame update, and both conditions must be met. Numerical relationships; When satisfied When a dynamic change occurs in the scene, the system detects the arrival or departure of a train.

6. The railway train detection method based on reference frame difference and frequency domain energy according to claim 5, characterized in that: In S4, when the frequency domain energy value continuous The frame is below the baseline frame update threshold When the current scene is determined to have returned to the background state, a baseline frame update is performed; whereby... The preset consecutive frame count threshold is 10 to 30 frames.

7. The railway train detection method based on reference frame difference and frequency domain energy according to claim 6, characterized in that: When performing a reference frame update, the current frame can be used to directly replace the original reference frame, or a weighted fusion method can be used to smoothly update the original reference frame. The updated reference frame is used to adapt to the operating environment of changes in lighting, weather, and background facilities.

8. The system for railway train detection based on reference frame difference and frequency domain energy according to any one of claims 1-7, characterized in that, include: The image acquisition module is used to acquire surveillance images along the railway line; The reference frame management module is used to establish reference frames in train-free scenarios and update the reference frames when update conditions are met. The difference processing module is used to perform difference operations between the current image and the reference frame to obtain a difference image; The frequency domain analysis module is used to perform frequency domain transformation on the difference image, extract frequency domain features, and calculate frequency domain energy values; The decision module is used to compare the frequency domain energy value with the train event alarm threshold and the reference frame update threshold, and output the decision result of the train entering event, the train leaving event, or the background stable state. The output module is used to output detection results or send alarm information to the upper-level system.

9. The system of the railway train detection method based on reference frame difference and frequency domain energy according to claim 8, characterized in that: The surveillance images are collected by cameras along the railway line. These cameras are deployed along the railway line area to collect continuous images and transmit them to the image acquisition module.

10. The system of the railway train detection method based on reference frame difference and frequency domain energy according to claim 8, characterized in that: It also includes an alarm module, which is used to issue an alarm signal or record event information when the decision module determines that a train arrival event or a train departure event has occurred.

Images