Platform edge monitoring and early warning method and system

Abstract

The invention discloses a platform edge monitoring and early warning method and system. The method comprises the following steps that whether a heavy object exists on the edge of a platform or not is detected; if yes, an image of the edge of the platform is collected; preprocessing the image to improve the image quality; segmenting a target and a background in the image; performing big data comparison judgment on the target data; if the target is a person, positioning and tracking the edge of the passenger in real time; if the target is an article, triggering a yellow alarm mechanism; measuring the distance between the human body key node and the edge of the platform; judging whether the distance between the passenger and the edge of the platform is smaller than a safety threshold; if yes, a red alarm mechanism is triggered; if not, a yellow alarm mechanism is triggered; the platform edge, the passenger density and the safety distance between the passengers and the platform can be monitored in real time, the monitoring efficiency and accuracy are improved, and the risk of safety accidents is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of platform monitoring, and in particular to a platform edge monitoring and early warning method and system. Background Art

[0002] In recent years, urban rail transit has become the backbone of public transportation in major cities due to its advantages such as large capacity, high efficiency, and punctuality. Passenger traffic at stations continues to rise, especially during peak hours in the morning and evening, when platforms are densely populated. Traditional manual security inspections are unable to promptly detect and address all safety hazards, necessitating the urgent need for intelligent security monitoring.

[0003] Safety accidents such as passengers falling onto the tracks at the edge of the platform, trampling during crowding, and sudden equipment failures occur frequently, which not only endanger the lives and property of passengers, but also seriously affect the normal operation of rail transit, causing huge social and economic losses and negative public opinion impact, highlighting the urgency of strengthening platform safety monitoring. Summary of the Invention

[0004] The purpose of the present invention is to provide a platform edge monitoring and early warning method and system, which can realize real-time monitoring of the platform edge, passenger density and the safe distance between passengers and the platform, improve monitoring efficiency and accuracy, and reduce the risk of safety accidents.

[0005] To achieve this object, the present invention adopts the following technical solutions: A platform edge monitoring and early warning method is provided, comprising the following steps: Check whether there are heavy objects on the edge of the platform; If so, collect images of the platform edge; Preprocess the image to improve image quality; Segment the target from the background in the image; Conduct big data comparison and judgment on target data; If the target is a person, the passenger edge is located and tracked in real time; If the target is an object, the yellow alarm mechanism is triggered; Calculate the distance between the key nodes of the human body and the edge of the platform; Determine whether the distance between the passenger and the platform edge is less than the safety threshold; If so, a red alarm mechanism is triggered; If not, the yellow alarm mechanism is triggered.

[0006] As a preferred solution of the platform edge monitoring and early warning method, the step of preprocessing the image to improve the image quality includes: image grayscale processing and bilateral filtering noise reduction processing.

[0007] As a preferred solution of the platform edge monitoring and early warning method, the step of measuring the distance between the key nodes of the human body and the platform edge includes: Select reference joint points: Based on the needs and actual conditions of platform safety monitoring, select passenger joint points as reference points for calculating the distance to the platform edge; Determination of distance calculation method: Get the coordinates of the selected reference joint point and the position information of the platform edge. When the platform edge is a straight line, set its equation to ax+by+c=0, and obtain the straight line parameters a, b, and c through edge positioning. The coordinates of the passenger reference joint point are (x0, y0). According to the distance formula from point to straight line Calculate the distance between passengers and the edge of the platform; Distance calculation loop: traverse each detected passenger, calculate its distance to the platform edge, and obtain an array or list containing the distances between all passengers and the platform edge, for example , where di represents the distance between the ith passenger and the edge of the platform.

[0008] As a preferred solution for the platform edge monitoring and early warning method, in the step of determining the distance calculation method, for a platform edge with a complex shape, it is divided into multiple line segments, and the distance from the passenger to each line segment is calculated separately, and the minimum value is taken as the final distance.

[0009] As a preferred solution for the platform edge monitoring and early warning method, the red alarm mechanism includes: red light warning, voice warning and pushing alarm information to the central control room.

[0010] As a preferred solution for the platform edge monitoring and early warning method, the yellow alarm mechanism includes: a yellow light reminder and a voice reminder.

[0011] The present invention also provides a platform edge monitoring and early warning system, which is characterized by comprising: Pressure sensor, used to detect whether there is heavy object on the edge of the platform; Camera, used to capture images of the platform edge; Image processing module, used for preprocessing images to improve image quality; Semantic segmentation model module, used to segment the target from the background in the image; The first judgment module is used to perform big data comparison and judgment on the target data; Positioning and tracking module, used for real-time positioning and tracking of passenger edges; Yellow alarm module, used to trigger the yellow alarm mechanism; Distance measurement module, used to measure the distance between the key nodes of the human body and the edge of the platform; The second judgment module is used to judge whether the distance between the passenger and the edge of the platform is less than a safety threshold; Red alarm module, used to trigger the red alarm mechanism.

[0012] Beneficial effects of the present invention: The platform edge monitoring and early warning method and system proposed in the present invention can realize real-time monitoring of the platform edge, passenger density and the safe distance between passengers and the platform, improve monitoring efficiency and accuracy, and reduce the risk of safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0014] Figure 1 This is a flow chart of a platform edge monitoring and early warning method according to an embodiment of the present invention; Figure 2 It is a schematic block diagram of the structure of a platform edge monitoring and early warning system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0016] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0017] Reference Figure 1 An embodiment of the present invention provides a platform edge monitoring and early warning method, comprising the following steps: S1: Detect whether there is any heavy object on the edge of the platform; S2: If yes, collect images of the platform edge; S3: preprocess the image to improve image quality; S4: Segment the target and background in the image; S5: Perform big data comparison and judgment on the target data; S6: If the target is a person, real-time positioning and tracking of the passenger edge is performed; S7: If the target is an object, the yellow alarm mechanism is triggered; S8: Calculate the distance between the key nodes of the human body and the edge of the platform; S9: Determine whether the distance between the passenger and the platform edge is less than a safety threshold; S10: If yes, trigger the red alarm mechanism; S11: If not, trigger the yellow alarm mechanism.

[0018] Image preprocessing: An optical camera is used to capture the original image of the platform target, and Zeiss coating and anti-shake gimbal are used to offset interference from ambient light and equipment jitter. Perspective transformation is used to correct image distortion, ensure edge linearity, and improve image edge accuracy. Finally, a series of image noise reduction and enhancement algorithms are synchronized with cloud computing to improve the final image quality.

[0019] (1) Image grayscale processing steps Principle understanding: Read the collected RGB color image and convert it into a grayscale image. The purpose of grayscale conversion is to reduce the data dimension and the amount of calculation for subsequent processing while retaining the basic information of the image. The following are two methods and calculation formulas for grayscale conversion: Weighted average method principle: The weighted average method takes into account the human eye's varying sensitivity to different colors. The human eye is most sensitive to green, followed by red, and least sensitive to blue. Therefore, when converting an RGB color image to grayscale, the green channel is given the highest weight, followed by the red channel, and the blue channel the lowest. Commonly used standard weights are 0.299 for red (R), 0.587 for green (G), and 0.114 for blue (B).

[0020] step: First, determine the color image to be grayscaled and obtain the RGB value of each pixel in the image.

[0021] For each pixel, use the weighted average formula to calculate the gray value. Gray=0.299*R+0.587*G+0.114*B Repeat the above calculation process for each pixel in the image, and replace the original RGB value with the obtained grayscale value to complete the grayscale conversion of the entire image.

[0022] Average method principle: The average value method assumes that the three RGB channels contribute equally to the brightness of the pixel, so the grayscale value is obtained by calculating the average value of the three RGB channel values ​​of each pixel.

[0023] step: For each pixel in a color image, get its RGB value.

[0024] Calculate the gray value according to the formula. Gray=(R+G+B) / 3 Calculate each pixel in the image according to this formula, and replace the original RGB value with the obtained grayscale value to complete the grayscale conversion of the image.

[0025] (2) Bilateral filtering noise reduction steps Understanding the Principle: Bilateral filtering is a nonlinear filtering method that considers both spatial distance and differences in pixel grayscale values. Its core concept is to use two weights to determine the contribution of each pixel to the filtering process: one based on spatial distance and the other based on grayscale similarity. This enables bilateral filtering to effectively preserve image edge information while removing noise.

[0026] Implementation steps: Step 1: Determine filter parameters Window Size Selection: Similar to Gaussian filtering, bilateral filtering requires a window size, which defines how many neighboring pixels around each pixel are considered during the filtering process. This window size is typically an odd number, such as 3×3 or 5×5. The choice of window size affects both filtering effectiveness and computational complexity. Smaller windows (such as 3×3) can better preserve detail in richly detailed images, but may have limited noise smoothing effects. Larger windows (such as 7×7) can more effectively remove noise but may blur some small details in the image.

[0027] Spatial distance standard deviation (σ_d): This parameter controls the weight distribution based on spatial distance. It determines the weight assigned to pixels within the filter window based on their distance from the center pixel. A smaller σ_d causes the spatial weight to drop off more rapidly near the center pixel, meaning that only pixels very close to the center pixel have a higher weight. A larger σ_d results in a more even distribution of spatial weights, allowing pixels farther away to have a greater impact on the filtering result of the center pixel. An appropriate value can generally be determined through experimentation, typically ranging from 0.5 to 10, depending on the image content and noise level.

[0028] Grayscale similarity standard deviation (σ_r): This parameter controls the weight distribution based on grayscale similarity. It determines the weight assigned to pixels within the filtering window based on their grayscale value difference from the center pixel. A smaller σ_r makes the grayscale similarity weight more sensitive to grayscale value differences, with only pixels with grayscale values ​​very close to the center pixel receiving a higher weight. A larger σ_r makes the grayscale similarity weight more tolerant of grayscale value differences, allowing more pixels with different grayscale values ​​to have a greater impact on the filtering result. The appropriate range for this parameter needs to be determined experimentally based on the specific image, but is typically between 10 and 100.

[0029] Step 2: Filter each pixel in the image Locate the pixel and its neighborhood window: For each pixel in the image, use the pixel as the center and determine the range of its neighborhood pixels based on the selected window size. For example, if a 3×3 window is selected, a 3×3 pixel area centered on the current pixel is considered.

[0030] Calculate spatial distance weight and grayscale similarity weight Spatial distance weight (W_d): For each pixel in the neighborhood window (let its coordinates be (xi, yi), and the center pixel coordinates be (xc, yc)), a weight is calculated based on the spatial distance. The calculation formula of the spatial distance weight is usually based on the Gaussian function, such as This formula indicates that the farther the pixel is from the center pixel, the smaller its spatial distance weight is.

[0031] Grayscale similarity weight (W_r): Calculate the grayscale value difference between each pixel in the neighborhood window and the central pixel, and then calculate the grayscale similarity weight based on this difference. The calculation formula can be , where I(xi,yi) is the grayscale value of the pixel at coordinates (xi,yi), and I(xc,yc) is the grayscale value of the center pixel. This formula indicates that the greater the difference in grayscale value between a pixel and the center pixel, the smaller its grayscale similarity weight.

[0032] Calculate comprehensive weights and update pixel values Comprehensive weight (W): Multiply the spatial distance weight and the grayscale similarity weight to obtain the comprehensive weight of each pixel .

[0033] Update pixel value: Use the comprehensive weight to perform weighted summation on the pixels in the neighborhood window to obtain the filtered pixel value. Let the original image be I and the filtered image be J. For the center pixel (xc, yc), the calculation formula for its filtered pixel value is The summation here is performed on all pixels within the neighborhood window. This formula means that the farther the pixel is from the center pixel, the smaller its spatial distance weight is.

[0034] Step 3: Traverse the entire image Following the method in step 2 above, starting from the pixel in the upper left corner of the image, bilateral filtering is performed on each pixel row by row and column by column until the pixel in the lower right corner of the image is processed. In this way, the bilateral filtering denoising process for the entire image is completed, and the denoised image is obtained. In this process, since bilateral filtering requires calculating the spatial and grayscale weights of each pixel and involves a large number of exponential and multiplication operations, the computational complexity is relatively large, but it can better preserve the edge and detail information of the image while removing noise.

[0035] Platform target segmentation: Use a TOF lens and a depth sensor to directly obtain depth information within the platform target range. At the same time, combined with a deep learning-based semantic segmentation model (such as U-Net), the target in the platform image is separated from the platform background, each pixel of the target is labeled, and the target edge area is accurately extracted. Combined with big data segmentation, the outline of a single target is identified, and the outline of objects or people is analyzed more accurately. Principle and implementation steps of human target segmentation A: Principle of Semantic Segmentation Model (U-Net) Based on Deep Learning U-Net is a classic convolutional neural network model with an encoder-decoder structure that performs well in image semantic segmentation tasks.

[0036] The encoder layer gradually extracts image features through a series of convolutional and pooling layers. As the network layers deepen, the resolution of the feature map decreases, but the semantic information of the features gradually becomes richer. For example, when processing platform images, the initial convolutional layers can extract low-level features such as edges and textures, while deeper convolutional layers can learn more abstract features such as the shape and posture of the crowd.

[0037] The decoder gradually upsamples the high-level semantic features obtained in the encoder to the same resolution as the original image. During the upsampling process, the feature maps of the corresponding resolution in the encoder are fused with the upsampling results in the decoder through skip connections. This allows the decoder to obtain high-level semantic information while preserving the details of the original image, thereby accurately classifying each pixel, separating the crowd from the background in the image, and assigning a corresponding label to each pixel (for example, crowd pixels are labeled as one category, background pixels are labeled as another).

[0038] B: Steps for separating target and background using U-Net (1) Data preparation Collect a large amount of image data containing platform scenes. These images should have different lighting conditions, crowd density, scene layout, etc. to cover various situations in practical applications.

[0039] The collected images are annotated by manually labeling each pixel with its category (target or background) to generate an annotation mask. The annotation mask corresponds one-to-one with the original image and is used to train the model.

[0040] (2) Model training The labeled image dataset is divided into a training set, a validation set, and a test set. The training set typically takes up the largest portion and is used to learn the model's parameters; the validation set is used to evaluate the model's performance and adjust hyperparameters during training; and the test set is used to ultimately evaluate the model's generalization capabilities.

[0041] The training set images are fed into the U-Net model, which predicts the class probability distribution for each pixel based on the input image. By comparing the predicted results with the annotated mask, a loss function (such as the cross-entropy loss function) is calculated, which reflects the accuracy of the model's predictions.

[0042] Using the backpropagation algorithm, the model parameters are adjusted based on the loss function, allowing the model to gradually learn the feature representations that accurately segment people from the background. During training, these steps are repeated until the model's performance on the validation set stabilizes or meets a preset stopping condition (such as achieving a certain accuracy or completing a certain number of training rounds).

[0043] (3) Model evaluation and optimization Evaluate the trained model using the test set and calculate metrics such as precision, recall, and F1 score to comprehensively assess model performance. If model performance is unsatisfactory, analyze the reasons and consider further adjustments to the model structure, increasing the amount of training data, optimizing hyperparameters, and retraining the model until satisfactory performance is achieved.

[0044] (4) Crowd and background separation in practical applications During actual operation, the station feeds real-time captured images into a trained U-Net model, which outputs the probability of each pixel belonging to the crowd or background. Based on a set threshold (for example, pixels with a probability greater than 0.5 are considered crowd pixels), the crowd and background areas in the image are separated, resulting in a binary mask image of the crowd, where white areas represent the crowd and black areas represent the background.

[0045] C: Combine instance segmentation to identify individual passenger outlines Based on the crowd region, instance segmentation is further used to identify the outlines of individual passengers. Instance segmentation can not only distinguish different object categories (such as crowds and background), but also separate different instances of the same category (such as individual passengers).

[0046] Advanced instance segmentation models, such as Mask R-CNN, can detect individual passengers in a crowd while generating precise contour masks for each passenger. These models learn richer feature representations during training, enabling them to distinguish the boundaries between different passengers, thereby achieving more accurate character contour analysis. For example, when dealing with crowded platform scenarios, instance segmentation models can accurately identify the position, posture, and outline shape of each passenger, providing more detailed information for subsequent behavior analysis and traffic statistics. Similarly, similar methods can be used to identify and analyze the contours of items carried by passengers, further improving our understanding and management of various objects in platform scenarios.

[0047] Passenger body contour edge positioning and tracking: Use deep learning models such as YOLO or Faster R-CNN to initially locate the passenger's body contour (presented as a bounding box) after training on a large data set; then use algorithms such as Canny to accurately extract the edges within the box and optimize them through morphological operations; then use optical flow methods (such as the Lucas-Kanade algorithm) to calculate the motion vectors of edge pixels for dynamic tracking, or combine with Kalman filtering to set edge positions and other variables as state variables. In the prediction and update stages, the optical flow estimate and the newly detected position are integrated to improve stability and accuracy, continuously and accurately track the edges of passenger body contours, and adapt to the complex environment of the platform and changes in passenger activities. (1) Target detection and contour extraction based on deep learning for initial positioning Model Selection and Training: Select advanced object detection models such as YOLO (You Only Look Once) or Faster R-CNN. Train on large-scale image datasets containing human figures, annotated with body categories and key points or bounding boxes, so the model can learn the appearance and contour patterns of the human body. For example, during training, the model automatically extracts features of human figures in images using a convolutional neural network. It gradually develops accurate recognition capabilities for human figures in various poses, viewing angles, and lighting conditions, accurately locating the human body in the image and providing an approximate outline (presented as a bounding box).

[0048] Image input and preliminary contour extraction: The pre-processed images collected from the platform are fed into the trained object detection model. The model scans and analyzes the images, outputting information about the detected human objects, including their location (bounding box coordinates), a confidence score, and a category label (e.g., "passenger"). Based on the bounding box coordinates, the approximate area of ​​the passenger's body contour can be initially determined. However, the contour is relatively rough at this point, consisting of a rectangular frame, and requires further refinement.

[0049] (2) Accurate contour extraction and edge detection refinement Application of Contour Refinement Algorithms: Within the preliminarily determined human contour area (within the bounding box), precise contour extraction is performed using algorithms such as the Active Contour Model (also known as the Snake model) or a modified version of the Canny edge detection algorithm. Taking the Canny algorithm as an example, image noise is first removed through bilateral filtering (see the "Image Preprocessing" section). The image's gradient magnitude and direction are then calculated, and possible edge points are screened using non-maximum suppression. Finally, a dual-threshold algorithm is used to determine the true edge points and connect them to form a contour. This algorithm extracts a more refined pixel-level curve of the human body's edges, clearly outlining the body's outer boundaries, including details such as the body contour and limb edges.

[0050] Morphological optimization: Morphological operations, such as erosion and dilation, are performed on the extracted edge contours. Erosion removes small noise points and burrs from the edge contour, making it smoother and more continuous. Dilation fills small holes or gaps within the contour, ensuring the integrity of the human body contour, avoiding outline breakage or loss caused by noise or local image defects, and further optimizing the quality of the human body contour edge.

[0051] (3) Dynamic tracking based on optical flow or Kalman filtering 1. Principles and Applications of Optical Flow (Taking the Lucas-Kanade Optical Flow Algorithm as an Example): For consecutive image frames in a video sequence, the grayscale values ​​of pixels in the image are assumed to remain constant over a short period of time. The optical flow field is calculated based on this assumption and the pixel displacement relationship between adjacent frames. In passenger body contour edge tracking, the extracted edge pixels of the human body contour in the current frame are used as the research objects. By calculating the optical flow vectors of these pixels between the previous and current frames, the direction and speed of their motion in the image plane are determined. For example, for an edge point (x, y) on the human body contour, the Lucas-Kanade algorithm calculates its optical flow vector (u, v), and the predicted position of this point in the next frame is (x+u, y+v). By calculating and tracking optical flow at multiple edge points, dynamic tracking of the human body contour is achieved, and the contour position is updated in real time to adapt to the changes in passenger movement within the platform.

[0052] 2. Principle and application of Kalman filtering: (1) Overview of Kalman Filter Principle: Kalman filtering is a recursive algorithm that uses the linear system state equation to optimally estimate the system state through system input and observation data. In edge tracking, the position, velocity, and other state information of the edge segment are used as system state variables, and the dynamic tracking of the edge position is achieved by continuously updating the state estimate.

[0053] (2) State variable definition and initialization: Define the state variables of the edge segment. For example, for a straight edge in a two-dimensional plane, the state can be represented by the line's endpoint coordinates (x1, y1) and (x2, y2) or the line's parameters (p, q) and their rate of change (p^, q^). Initialize the state variable values ​​based on the initial edge segment position detected by the Hough transform.

[0054] (3) Prediction stage: Based on the dynamic model of the system (for example, assuming that the edge segment performs uniform linear motion or uniformly accelerated linear motion in a short period of time, and selecting the appropriate motion model according to the actual situation), the state estimate value at the previous moment is used to predict the state of the edge segment at the current moment. For example, if the uniform linear motion model is used, the prediction formula is , where x^k is the predicted state vector at the current moment, F is the state transfer matrix, x^k-1 is the state estimate at the previous moment, B is the control input matrix, and uk is the control input vector (uk can be a zero vector if there is no external control input).

[0055] (4) Update phase Get the observation value at the current moment, that is, the new position information of the edge segment obtained through image acquisition and processing (such as new edge point coordinates or new line parameters). Calculate the residual between the observation value and the predicted value, where zk is the observation value and H is the observation matrix.

[0056] According to the Kalman gain formula Calculate the Kalman gain Kk, where Pk is the prediction error covariance matrix and R is the observation noise covariance matrix.

[0057] Update state estimates using Kalman gain , and update the prediction error covariance matrix .

[0058] (5) Real-time update and cyclic tracking: The prediction and update phases are repeated continuously. As time goes by, the Kalman filter can continuously adjust the state estimate of the edge segment based on the new observation data and update the edge position in real time. Even if the edge is slightly displaced by factors such as lighting changes and vehicle vibrations, it can accurately track the actual position change of the edge, providing reliable edge position information for subsequent safety monitoring and analysis. In each iteration, the updated state estimate is used for the next prediction, forming a continuous dynamic tracking process. Passenger distance measurement and alarm: Based on the human posture recognition model, the key joint points of passengers are located and the distance between the passengers and the edge of the platform is calculated. A safety threshold is set. If the distance is less than the set safety threshold, the light alarm is instantly triggered. At the same time, the alarm information is pushed to the control room, including the specific platform location, time, and degree of danger (stay time), so that security personnel can intervene quickly.

[0059] Human posture recognition model locates key joints of passengers Model Selection and Preparation: Choose an appropriate human pose recognition model, such as OpenPose. These models are pre-trained on large-scale human pose datasets and can identify multiple key human joints, such as the head, neck, shoulders, elbows, wrists, hips, knees, and ankles. Load the model into the system and adjust necessary parameters based on the specific characteristics of the platform scenario, such as adjusting the input image size and detection confidence threshold.

[0060] Image Input and Joint Recognition: The platform image, which has undergone preprocessing (including grayscale conversion, bilateral filtering for noise reduction, edge detection, and edge location tracking), is input into the human posture recognition model. The model analyzes the image and outputs the coordinates of each passenger's key joints. For each detected passenger, the joint coordinates are stored in a data structure (such as an array or list). For example, for a scene containing n passengers, the resulting array might be a three-dimensional array, where the first dimension represents the passenger number, the second dimension represents the joint number, and the third dimension represents the two-dimensional (x, y) coordinates of the joint in the image.

[0061] Calculate the distance between passengers and the edge of the platform Select a reference joint: Based on the needs and actual conditions of platform safety monitoring, select an appropriate passenger joint as the reference point for calculating the distance to the platform edge. Typically, the foot joint (such as the ankle) is selected because it is in contact with the platform floor and can better reflect the relative position of the passenger to the platform edge.

[0062] Distance calculation method: Get the coordinates of the selected reference joint point and the position information of the platform edge (the platform edge position can be determined by the edge segment equation or edge point coordinates obtained by edge positioning and tracking). If the platform edge is a straight line, set its equation to ax+by+c=0 (the straight line parameters a, b, c are obtained by edge positioning), and the coordinates of the passenger reference joint point are (x0, y0), then according to the distance formula from point to straight line Calculate the distance between the passenger and the platform edge. For complex platform edges (such as curves or polylines), split them into multiple line segments. Calculate the distance from the passenger to each line segment separately, and take the minimum value as the final distance.

[0063] Distance calculation loop: traverse each detected passenger, calculate its distance to the platform edge, and obtain an array or list containing the distances between all passengers and the platform edge, for example , where di represents the distance between the ith passenger and the edge of the platform.

[0064] Alarm triggering mechanism and setting safety thresholds Alarm triggering: The distance between each passenger and the platform edge is monitored in real time, and the calculated distance is compared with a safety threshold. For each passenger's distance di, if di < the safety threshold, the alarm mechanism is triggered. If it does not exceed the safety threshold, the human posture recognition model is used again in the first step to locate the passenger's key joints. (The safety threshold is set based on factors such as the actual platform layout, train operation rules, passenger behavior characteristics, and safety standards. For example, considering the braking distance when the train enters the station, the passenger's reaction time, and the need to prevent passengers from accidentally falling off the platform, the safety threshold may be set at 0.80 meters from the platform edge. (The specific value is subject to rigorous safety assessment and testing based on actual conditions.) Light alarm trigger and information push Light Alarm Control: When a passenger's distance is detected to be less than a safety threshold, a trigger signal is immediately sent to a connected light alarm. This light alarm can be a yellow warning light installed at the edge of the platform. The flashing light attracts the attention of passengers and surrounding personnel, reminding them to stay away from the dangerous area at the edge of the platform.

[0065] Alarm Information Generation and Push: The system also generates detailed alarm information, including the location of the passenger who triggered the alarm (which can be pinpointed to the specific area using coordinate information from the image combined with the actual platform layout, such as the platform number and the location of the nearest carriage), the alarm trigger time (accurate to the second or millisecond, recording the moment the system detected the dangerous situation), and the risk level assessment (the degree to which the passenger's distance from the platform edge deviates from the safety threshold; for example, the closer the distance to the threshold, the higher the risk, with levels of high, medium, and low). This alarm information is pushed to the central control room's monitoring system using UWB technology via network communication technologies (such as TCP / IP).

[0066] Control room response and security personnel intervention Alarm Reception and Display in the Central Control Room: Upon receiving an alarm, the monitoring system in the central control room presents it to security personnel in an intuitive manner. For example, a pop-up window containing the alarm information appears on the monitoring screen, accompanied by an audible prompt. The pop-up window displays a screenshot of the passenger's location (with the passenger and platform edge highlighted), a detailed location description, the time of day, and the severity of the danger, enabling security personnel to quickly understand the situation.

[0067] Security personnel deployment: Based on the location indicated by the alarm, security personnel quickly respond to the scene and intervene. They collaborate using communication devices like intercoms and take appropriate measures based on the severity of the danger. These measures include guiding passengers away from dangerous areas, setting up temporary warning signs, and assisting passengers in boarding and exiting the vehicle, ensuring passenger safety and preventing potential incidents. After addressing the alarm, security personnel provide feedback through the central control room system, which records the relevant information for subsequent safety analysis and statistics.

[0068] Reference Figure 2 An embodiment of the present invention further provides a platform edge monitoring and early warning system, comprising: Pressure sensor, used to detect whether there is heavy object on the edge of the platform; Camera, used to capture images of the platform edge; Image processing module, used for preprocessing images to improve image quality; Semantic segmentation model module, used to segment the target from the background in the image; The first judgment module is used to perform big data comparison and judgment on the target data; Positioning and tracking module, used for real-time positioning and tracking of passenger edges; Yellow alarm module, used to trigger the yellow alarm mechanism; Distance measurement module, used to measure the distance between the key nodes of the human body and the edge of the platform; The second judgment module is used to judge whether the distance between the passenger and the edge of the platform is less than a safety threshold; Red alarm module, used to trigger the red alarm mechanism.

[0069] In the description of the present invention, it should be understood that the terms "middle", "length", "upper", "lower", "front", "back", "vertical", "horizontal", "inner", "outer", "radial", "circumferential", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0070] In the present invention, unless otherwise expressly specified or limited, a first feature "on" a second feature may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediary. "Multiple" means at least two, such as two or three, unless otherwise expressly specified or limited.

[0071] In the present invention, unless otherwise specified or limited, the terms "installed," "connected," "connect," "fixed," etc. should be understood in a broad sense. For example, they can refer to fixed connection, detachable connection, or integration; mechanical connection, electrical connection, or communication; direct connection or indirect connection through an intermediate medium; internal communication between two elements or interaction between two elements, unless otherwise specified. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0072] The above is only for explaining the embodiments of the present invention and is not intended to limit the present invention. For those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present invention without creative work should be included in the scope of protection of the present invention.

Claims

1. A platform edge monitoring and early warning method, characterized in that: The following steps are involved: Check whether there are heavy objects on the edge of the platform; If so, collect images of the platform edge; Preprocess the image to improve image quality; Segment the target from the background in the image; Conduct big data comparison and judgment on target data; If the target is a person, the passenger edge is located and tracked in real time; If the target is an object, the yellow alarm mechanism is triggered; Calculate the distance between the key nodes of the human body and the edge of the platform; Determine whether the distance between the passenger and the platform edge is less than the safety threshold; If so, a red alarm mechanism is triggered; If not, the yellow alarm mechanism is triggered.

2. The platform edge monitoring and early warning method according to claim 1, characterized in that: The step of pre-processing the image to improve the image quality includes: image grayscale processing and bilateral filtering noise reduction processing.

3. The platform edge monitoring and early warning method according to claim 1, characterized in that: The step of measuring the distance between the key nodes of the human body and the edge of the platform includes: Select reference joint points: Based on the needs and actual conditions of platform safety monitoring, select passenger joint points as reference points for calculating the distance to the platform edge; Determination of distance calculation method: Get the coordinates of the selected reference joint point and the position information of the platform edge. When the platform edge is a straight line, set its equation to ax+by+c=0, and obtain the straight line parameters a, b, and c through edge positioning. The coordinates of the passenger reference joint point are (x0, y0). According to the distance formula from point to straight line Calculate the distance between passengers and the edge of the platform; Distance calculation loop: traverse each detected passenger, calculate its distance to the platform edge, and obtain an array or list containing the distances between all passengers and the platform edge, for example , where di represents the distance between the ith passenger and the edge of the platform.

4. The platform edge monitoring and early warning method according to claim 3, characterized in that: In the determining step of the distance calculation method, for a platform edge with a complex shape, it is divided into multiple line segments, and the distance from the passenger to each line segment is calculated respectively, and the minimum value is taken as the final distance.

5. The platform edge monitoring and early warning method according to claim 1, characterized in that: The red alarm mechanism includes: red light warning, voice warning and pushing alarm information to the central control room.

6. The platform edge detection and early warning method according to claim 1, characterized in that: The yellow alarm mechanism includes: a yellow light reminder and a voice reminder.

7. A platform edge monitoring and early warning system, characterized in that: include: Pressure sensor, used to detect whether there is heavy object on the edge of the platform; Camera, used to capture images of the platform edge; Image processing module, used for preprocessing images to improve image quality; Semantic segmentation model module, used to segment the target from the background in the image; The first judgment module is used to perform big data comparison and judgment on the target data; Positioning and tracking module, used for real-time positioning and tracking of passenger edges; Yellow alarm module, used to trigger the yellow alarm mechanism; Distance measurement module, used to measure the distance between the key nodes of the human body and the edge of the platform; The second judgment module is used to judge whether the distance between the passenger and the edge of the platform is less than a safety threshold; Red alarm module, used to trigger the red alarm mechanism.

Images