Safety monitoring system and method for railway station platform

Abstract

The invention discloses a railway station platform safety monitoring system and method, and relates to the field of platform safety monitoring. A railway station platform safety monitoring system comprises an environment data acquisition module, a station platform target screening module, an abnormal target identification module and a safety monitoring intervention module. The platform environment data is collected in real time by deploying various sensing devices, and the initial platform environment data set is constructed, so that the full-scene coverage and dynamic monitoring capability of targets such as personnel, articles and flyers is improved, and comprehensive and accurate data support is provided for subsequent intelligent analysis; through target detection and type division, platform target data is divided into three subsets of flyers, personnel and articles, independent modeling, accurate identification and targeted risk control can be carried out on different types of safety events, and the refinement capability of a monitoring system is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of platform safety monitoring, and in particular to a railway station platform safety monitoring system and method. Background Art

[0002] Train platform safety is a critical component of the railway transportation system. Its safety is directly related to the safety of passengers and property, as well as the normal operation of railways. With the continuous development of railway transportation and the continuous growth of passenger volume, the demand for train platform safety monitoring is also increasing. Existing safety monitoring methods mainly identify and address dangerous behaviors that have already occurred, but lack the ability to analyze and predict the precursors of dangerous behaviors. Most existing safety monitoring methods use a unified detection and analysis strategy, without establishing differentiated detection methods and anomaly assessment criteria for different types of targets.

[0003] To this end, the present invention proposes a railway platform safety monitoring system and method to solve the above problems. Summary of the Invention

[0004] In response to the above problems, the present invention proposes a railway platform safety monitoring system and method to perform differentiated safety monitoring on different types of targets on the railway platform.

[0005] A railway platform safety monitoring method comprises the following steps: Assume that a train platform includes several sensing devices, based on which platform environment data is collected in real time, and several pieces of platform environment data are obtained each time; the several pieces of platform environment data are combined to obtain an initial platform environment data set; The initial platform environment data set is screened to obtain a safe target detection data set; the safe target detection data set is divided into safe detection data subsets Z according to target type. i , i=1, 2, 3; wherein, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the item data subset; For the security detection data subset Z i Abnormal feature screening is performed separately to obtain abnormal flying object dataset, abnormal personnel dataset and abnormal item dataset; Anomaly assessments are performed on the abnormal flying object dataset, abnormal personnel dataset, and abnormal object dataset to obtain the comprehensive flying object hazard level, comprehensive personnel hazard level, and comprehensive object hazard level; Based on the comprehensive flying object hazard level, comprehensive personnel hazard level and comprehensive object hazard level, corresponding safety intervention operations are performed to complete a railway station platform safety monitoring.

[0006] As a preferred technical solution of the present invention, the security target detection data set is divided into security detection data subsets Z according to target type. i The specific steps include: Perform target detection on the platform environment data in the initial platform environment dataset to obtain target identification labels; retain all platform environment data with target identification labels as yes to form a safe target detection dataset; Traverse the platform environment data in the safety target detection data set and segment it by target to obtain several platform target data; Target types are set to flying objects, personnel, and objects; Divide all station target data by target type to obtain the security detection data subset Z i , i=1, 2, 3; among them, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the item data subset.

[0007] As a preferred technical solution of the present invention, the security detection data subset Z i The specific steps for screening abnormal features include: For the security detection data subset Z i Extract the features of the platform target data in the dataset to obtain the features of the platform target data; The platform target data features are identified using the pre-trained abnormal target recognition model; if the platform target data features are abnormal, the platform target data is identified according to the security detection data subset Z to which the platform target data belongs. i , update the platform target data to abnormal flying object data or abnormal personnel data or abnormal item data, add the abnormal flying object data to the abnormal flying object data set; add the abnormal personnel data to the abnormal personnel data set; add the abnormal item data to the abnormal item data set; if the platform target data feature is normal, delete the corresponding platform target data; Traverse all security detection data subsets Z i , we obtain the abnormal flying object dataset, abnormal personnel dataset and abnormal item dataset after abnormal feature screening.

[0008] As a preferred technical solution of the present invention, the specific steps of performing anomaly assessment on an abnormal flying object dataset include: For any abnormal flying object data in the abnormal flying object dataset, perform the following operations: Obtain the current train platform weather data and identify the current risk level threshold based on the train platform weather data; Analyze the trajectory change rate of abnormal flying object data to obtain the speed change rate and altitude change rate of the flying object; A virtual flight area division model is established based on the train platform, and the proximity of abnormal flying object data to the dangerous area in the virtual flight area division model is identified to obtain the flying object area proximity; Construct an aircraft anomaly scoring matrix based on the aircraft speed change rate, aircraft altitude change rate and aircraft area proximity; Perform weighted calculation on the flight object anomaly scoring matrix to obtain the flight object anomaly score; Compare the flight object anomaly score with the current risk level threshold to obtain the flight object hazard level of the abnormal flight object data; Traverse the abnormal flying object data set to obtain the comprehensive flying object hazard level.

[0009] As a preferred technical solution of the present invention, the specific steps of performing anomaly assessment on an abnormal personnel dataset include: For any abnormal person data in the abnormal person dataset, perform the following operations: Extract the first K frames of abnormal personnel data corresponding to the abnormal personnel data and obtain the personnel behavior unit sequence P, P={P k |k=1, 2, ..., K+1}, K+1 is the total length of the personnel behavior unit sequence P, P k Abnormal personnel behavior data; Abnormal personnel behavior data P k Extract the time series behavior characteristics and obtain the abnormal time series behavior characteristics T k ; Build a behavior pattern library that includes normal behavior patterns and abnormal behavior precursor patterns; The abnormal timing behavior characteristics T k Match with the behavior pattern library and calculate the abnormal personnel behavior score F k ; For the front Abnormal personnel behavior score F k , use the first weight coefficient to perform weighted averaging to obtain the first abnormal personnel behavior average score; for the latter Abnormal personnel behavior score F k ; Use the second weight coefficient to perform weighted averaging on the scores to obtain the second average score of abnormal personnel behavior; represents the floor function; The average of the first abnormal person behavior score and the second abnormal person behavior score is taken as the person abnormality score; the person danger level is matched according to the person abnormality score; Traverse the abnormal personnel data set to obtain the comprehensive personnel danger level.

[0010] As a preferred technical solution of the present invention, the specific steps of performing anomaly assessment on an abnormal item dataset include: For any abnormal item data in the abnormal item dataset, perform the following operations: Identify the location of abnormal items based on abnormal item data; Evaluate the item anomaly score based on the abnormal item data and the abnormal item placement location; identify the item's danger level based on the item anomaly score; Traverse the item hazard levels to obtain the comprehensive item hazard level.

[0011] A railway platform safety monitoring system, comprising: The environmental data acquisition module is used to assume that the train platform includes several sensing devices, collect platform environmental data in real time based on the several sensing devices, and obtain several pieces of platform environmental data each time; combine the several pieces of platform environmental data to obtain an initial platform environmental data set; The platform target screening module is used to screen the initial platform environment data set to obtain a safe target detection data set; the safe target detection data set is divided into safe detection data subsets Z according to the target type i , i=1, 2, 3; wherein, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the item data subset; Abnormal target recognition module, used to detect the security data subset Z i Abnormal feature screening is performed separately to obtain abnormal flying object datasets, abnormal personnel datasets, and abnormal object datasets; abnormal evaluation is performed on the abnormal flying object datasets, abnormal personnel datasets, and abnormal object datasets respectively to obtain comprehensive flying object hazard levels, comprehensive personnel hazard levels, and comprehensive object hazard levels; The safety monitoring intervention module is used to perform corresponding safety intervention operations based on the comprehensive flying object hazard level, the comprehensive personnel hazard level and the comprehensive object hazard level to complete a railway station platform safety monitoring.

[0012] The present invention has the following advantages: The present invention deploys a variety of sensing devices to collect platform environmental data in real time and constructs an initial platform environmental data set, thereby improving the full-scene coverage and dynamic monitoring capabilities of targets such as people, objects, and flying objects, and providing comprehensive and accurate data support for subsequent intelligent analysis; through target detection and type classification, the platform target data is divided into three subsets: flying objects, people, and objects, which helps to independently model, accurately identify and carry out targeted risk control of security incidents of different categories, and enhance the refinement capability of the monitoring system; by linking various comprehensive hazard levels with platform safety strategies, it automatically executes safety intervention measures such as broadcasting, scheduling, blockade, and manual inspections, and realizes an integrated closed-loop process of perception, identification, evaluation, and control, which significantly improves the intelligence and automation level of platform safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The figure is a schematic structural diagram of a railway platform safety monitoring system adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0015] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0016] Example 1, a railway platform safety monitoring method, comprising the following steps: Assume that a train platform includes several sensing devices, based on which platform environment data is collected in real time, and several pieces of platform environment data are obtained each time; the several pieces of platform environment data are combined to obtain an initial platform environment data set; The initial platform environment data set is screened to obtain a safe target detection data set; the safe target detection data set is divided into safe detection data subsets Z according to target type. i , i=1, 2, 3; wherein, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the item data subset; Train platforms are equipped with a variety of sensing devices for real-time collection of environmental information. These sensing devices include multiple video cameras (to monitor human activities, flying object intrusions, and abandoned items), infrared thermal imagers (to identify human heat sources and objects), etc. These devices work together to collect and generate multiple pieces of platform environmental data, which are combined to form the initial platform environmental data set.

[0017] The security target detection dataset is divided into security detection data subsets Z according to target type. i The specific steps include: Perform target detection on the platform environment data in the initial platform environment dataset to obtain target identification labels; retain all platform environment data with target identification labels as yes to form a safe target detection dataset; Target detection models, such as YOLOv5 / YOLOv8 models, can be used to detect targets in the station environment data to identify whether the data contains valid information and retain all station environment data that "contains at least one target"; The platform environment data in the safety target detection dataset is traversed and segmented by target to obtain several pieces of platform target data. In the obtained safety target detection dataset, each piece of data may contain multiple targets, so it is necessary to segment each detected target. Target segmentation can be performed by cropping or extracting the original image based on the bounding box in the target detection result to generate image data for each independent target. Target types are set to flying objects, personnel, and objects; Divide all station target data by target type to obtain the security detection data subset Z i , i=1, 2, 3; among them, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the object data subset; traverse all target data, filter them according to the category labels given during detection, and classify them into corresponding safety detection data subsets: flying object targets enter Z1, personnel targets enter Z2, and object targets enter Z3; finally, all target data are effectively classified into three subsets according to their types, which is convenient for subsequent safety analysis and processing.

[0018] For the security detection data subset Z i Abnormal feature screening is performed separately to obtain abnormal flying object dataset, abnormal personnel dataset and abnormal item dataset; For the security detection data subset Z i The specific steps for screening abnormal features include: For the security detection data subset Z i The platform target data in the image is extracted to obtain the platform target data features; the extraction method can adopt a deep convolutional neural network, input each target image into the network, and obtain a high-dimensional feature vector by removing the feature extraction part after the classification layer, as the structured feature expression of the target; The platform target data features are identified using the pre-trained abnormal target recognition model; if the platform target data features are abnormal, the platform target data is identified according to the security detection data subset Z to which the platform target data belongs. i , update the platform target data to abnormal flying object data or abnormal personnel data or abnormal item data, add the abnormal flying object data to the abnormal flying object data set; add the abnormal personnel data to the abnormal personnel data set; add the abnormal item data to the abnormal item data set; if the platform target data feature is normal, delete the corresponding platform target data; Traverse all security detection data subsets Zi , obtain the abnormal flying object dataset, abnormal personnel dataset and abnormal item dataset after abnormal feature screening; The pre-trained abnormal target recognition model is trained on a large amount of normal target data from the stations to learn the characteristic distribution pattern of normal targets. The training process usually adopts a feature extraction network based on ResNet, MobileNet or Vision Transformer, combined with a multi-layer perceptron (MLP), support vector machine (SVM) or autoencoder as a classifier or anomaly detection module. The specific feature extraction network and anomaly detection module are manually determined by professional technicians; the training data mainly consists of samples of flying objects, personnel or objects marked as normal. When the training method is supervised learning, some labeled abnormal samples are introduced; when the training method is unsupervised learning, only normal samples are relied upon for modeling; the training goal is to maximize the recognition accuracy of normal samples or minimize the reconstruction error, so that the model has the ability to recognize deviations from normal patterns; the training process is guided by the performance of the validation set, and adopts an early stopping strategy or reaches the set maximum number of rounds as the termination condition, and finally obtains a pre-trained model that can be used to quickly identify abnormal target features.

[0019] Anomaly assessments are performed on the abnormal flying object dataset, abnormal personnel dataset, and abnormal object dataset to obtain the comprehensive flying object hazard level, comprehensive personnel hazard level, and comprehensive object hazard level; The specific steps for anomaly assessment of the abnormal flying object dataset include: For any abnormal flying object data in the abnormal flying object dataset, perform the following operations: Obtain the current train platform weather data and identify the current risk level threshold based on the train platform weather data; First, the station obtains real-time weather data, such as wind speed, rainfall, and visibility, from the station's environmental perception system. Based on pre-set weather-risk association rules (e.g., the higher the wind speed, the more likely an aircraft is to lose control), the station calculates the current aircraft risk level threshold. This threshold serves as a reference for determining whether an aircraft's abnormality constitutes a danger. Analyze the trajectory change rate of abnormal flying object data to obtain the speed change rate and altitude change rate of the flying object; Perform time series analysis on the trajectory data of each abnormal flying object to extract the changes in its flight speed and altitude over time. By calculating the speed and altitude differences between consecutive time points, the speed and altitude change rates of the flying object are obtained to reflect the dynamic instability of the flying object. The more dramatic the changes, the higher the potential risk. A virtual flight area division model is established based on the train platform, and the proximity of abnormal flying object data to the dangerous area in the virtual flight area division model is identified to obtain the flying object area proximity; A hierarchical virtual flight zone division model (e.g., safe zone, buffer zone, and danger zone) is established with the train platform as the center. Based on the current coordinate trajectory of the flying object, its spatial distance from the danger zone boundary or the degree of its crossing the boundary is determined, thereby calculating the flying object's regional proximity, that is, its proximity to the danger zone, which is used to measure its spatial threat. An aircraft anomaly scoring matrix is ​​constructed based on the aircraft's speed change rate, altitude change rate, and regional proximity. A weighted calculation is performed on the aircraft anomaly scoring matrix to obtain an aircraft anomaly score. The three key features: speed change rate, altitude change rate, and regional proximity are organized into a three-dimensional aircraft anomaly scoring matrix, with each dimension representing a risk factor. This matrix can be used to further calculate the overall degree of anomaly of an aircraft, preparing for subsequent weighted scoring. A weighted summation is performed on each indicator in the aircraft anomaly scoring matrix to obtain a comprehensive anomaly score for each abnormal aircraft. Weights can be set based on practical experience, expert knowledge, or data-driven methods, and the weighted results reflect the comprehensive anomaly risk of an aircraft in multiple dimensions. Compare the flight object anomaly score with the current risk level threshold to obtain the flight object danger level of the abnormal flight object data; compare the anomaly score of each abnormal flight object with the risk level threshold under the current weather conditions. If the score exceeds the threshold, it is judged as a high danger level; if it is close to but not exceeded, it is a medium danger level; if it is far below the threshold, it is a low danger level. Each abnormal flight object thus obtains a clear danger level label; Traverse the abnormal flying object data set to obtain a comprehensive flying object hazard level; traverse the entire abnormal flying object data set, perform statistical analysis on the hazard levels of all flying objects, calculate the hazard level distribution ratio, and finally obtain a comprehensive flying object hazard level, which can be used for platform safety warning, scheduling intervention and management decision-making.

[0020] The specific steps for anomaly assessment on the abnormal person dataset include: For any abnormal person data in the abnormal person dataset, perform the following operations: Extract the first K frames of abnormal personnel data corresponding to the abnormal personnel data and obtain the personnel behavior unit sequence P, P={P k |k=1, 2, ..., K+1}, K+1 is the total length of the personnel behavior unit sequence P, P k Abnormal personnel behavior data; For each abnormal person data, extract the first K frames of data in the corresponding video frame sequence, and take the current frame as the K+1th frame, forming a behavior segment sequence P. Each abnormal person behavior data Pk Represents a human behavior unit, that is, the human posture, action and other data contained in a frame, which is used to restore the short-term behavior change process and ensure the continuity and contextual perception of human behavior; Abnormal personnel behavior data P k Extract the time series behavior characteristics and obtain the abnormal time series behavior characteristics T k Use action recognition models or human key point detection networks to extract dynamic time series features such as key points, posture vectors, speed changes, and movement amplitudes to obtain corresponding behavioral features. These time series behavioral features reflect the changing trends and abnormal tendencies of the person's actions in each frame and are the core indicators for determining whether the behavior is abnormal. Build a behavioral pattern library that includes normal behavior patterns and abnormal behavior precursor patterns. Build a behavioral pattern library that covers common normal behavior patterns (such as waiting for a bus, walking, waiting, getting on and off a bus, etc.) and abnormal behavior precursor patterns (such as sudden running, climbing over, wandering, walking against the flow, etc.). Each pattern is represented by a template or embedding vector trained with historical behavioral feature data. This can be constructed through manual rule annotation or unsupervised cluster analysis. This library serves as a reference standard for subsequent behavior matching and scoring. The abnormal timing behavior characteristics T k Match with the behavior pattern library and calculate the abnormal personnel behavior score F k Calculate the similarity or probability of each abnormal time series behavior feature with the patterns in the behavior pattern library (such as using cosine similarity) to obtain an abnormal behavior score. This score indicates the similarity between the current frame behavior and the abnormal pattern. The higher the score, the closer it is to potentially dangerous behavior. For the front Abnormal personnel behavior score F k , use the first weight coefficient to perform weighted averaging to obtain the first abnormal personnel behavior average score; for the latter Abnormal personnel behavior score F k ; Use the second weight coefficient to perform weighted averaging on the scores to obtain the second average score of abnormal personnel behavior; represents a floor function; the first weight coefficient and the second weight coefficient are manually set by professional technicians; Abnormal human behavior usually has the characteristics of temporal evolution. In the early stage, it may only deviate from the normal trend, but in the later stage, it may quickly escalate into high-risk behavior. By dividing the behavior sequence into two stages and weighting them separately, the risk characteristics of the behavior evolution process can be captured more accurately. The weighting method can be set according to actual needs. For example, giving higher weights to later behaviors can highlight the response to sudden dangerous behaviors (such as sudden running and collisions). It can also strengthen early detection and prevent risk evolution. The specific setting method is set manually. Compared with directly averaging the behavior scores of all frames, stage-by-stage weighting can more flexibly reflect the changes in risk trends within the time window and improve the discrimination and robustness of the behavior recognition model. Some abnormal behaviors may only appear in the second half and are easily diluted if the overall average is used. Stage-by-stage weighting can highlight the importance of key frames and prevent them from being misjudged as normal. The average of the first abnormal person behavior score and the second abnormal person behavior score is taken as the person anomaly score; the person's danger level is matched based on the person anomaly score; this score is used to quantify the degree to which the person's behavior deviates from the normal pattern in the entire behavior sequence; the person is marked as the corresponding danger level based on the preset danger level standard; the preset danger level standard is set manually; Traverse the abnormal personnel data set to obtain a comprehensive personnel danger level. Traverse the entire abnormal personnel data set, perform the above processing on each data point in turn, and record the danger level of all personnel. By calculating the average level, maximum level, or level distribution ratio, a comprehensive personnel danger level assessment result for the current platform can be formed to provide support for safety scheduling, risk intervention, and early warning decision-making.

[0021] The specific steps for abnormality assessment of abnormal item datasets include: For any abnormal item data in the abnormal item dataset, perform the following operations: Identify the location of abnormal items based on abnormal item data. Image analysis technology is used to identify the specific location of the item on the train platform. This location can be mapped to the platform's structured area division map. The relative distance and location of the item to key areas are then calculated to determine whether the item is in an abnormal or high-risk area. Evaluate the item anomaly score based on the abnormal item data and the abnormal item placement location; identify the item's danger level based on the item anomaly score; based on the location, comprehensively analyze the abnormal item's type, volume, occlusion, and placement behavior to score each item, and obtain an item anomaly score that indicates its potential threat to the security environment; Traverse the item danger levels to obtain a comprehensive item danger level; compare the evaluated item anomaly score with the set level threshold and classify it into different danger levels; traverse the entire abnormal item data set and perform statistics and analysis on the danger levels of all items, such as calculating the proportion of each level to form the current comprehensive item danger level of the platform; Based on the comprehensive flying object hazard level, comprehensive personnel hazard level, and comprehensive object hazard level, corresponding safety intervention operations are performed to complete a railway platform safety monitoring; For example, if the overall flying object hazard level is high, the drone or high-altitude foreign object monitoring plan should be immediately activated, with air safety warnings broadcast and cameras linked for automatic tracking. Security personnel should also be notified to patrol areas where flying objects may intrude. If necessary, trains may be halted or sections of the platform blocked to prevent flying objects from interfering with train operations or causing personal injury. When the overall personnel hazard level reaches medium to high, the voice broadcast system should be automatically activated to issue behavioral warnings (such as "Do not run, Do not climb") to the target area, and security personnel should be dispatched to the target area to provide persuasion, intervention, or removal. Furthermore, real-time behavioral risk information can be pushed to duty officers via large screens or mobile devices, enabling precise control and rapid response. If the overall object hazard level increases, the automatic unclaimed object identification and timed tracking mechanism will be triggered, initiating a broadcast reminder stating "Please take care of your belongings" and simultaneously notifying the security terminal of the location of any suspicious items. For high-risk items (such as those that have been unclaimed for an extended period or have a suspicious shape), dedicated personnel will be immediately dispatched to inspect or remove them, and, if appropriate, the public security and bomb disposal departments will be requested to intervene.

[0022] Example 2, a railway platform safety monitoring system, such as Figure 1 As shown, it includes the following modules: The environmental data acquisition module is used to assume that the train platform includes several sensing devices, collect platform environmental data in real time based on the several sensing devices, and obtain several pieces of platform environmental data each time; combine the several pieces of platform environmental data to obtain an initial platform environmental data set; The platform target screening module is used to screen the initial platform environment data set to obtain a safe target detection data set; the safe target detection data set is divided into safe detection data subsets Z according to the target type i , i=1, 2, 3; wherein, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the item data subset; The abnormal target identification module is used to screen the abnormal features of the security detection data subset Zi respectively to obtain the abnormal flying object data set, abnormal personnel data set and abnormal object data set; perform abnormal assessment on the abnormal flying object data set, abnormal personnel data set and abnormal object data set respectively to obtain the comprehensive flying object danger level, comprehensive personnel danger level and comprehensive object danger level; The safety monitoring intervention module is used to perform corresponding safety intervention operations based on the comprehensive flying object hazard level, the comprehensive personnel hazard level and the comprehensive object hazard level to complete a railway station platform safety monitoring.

[0023] It should be understood that those skilled in the art may make improvements or modifications based on the above description, and all such improvements and modifications shall fall within the scope of protection of the appended claims. Any portion of this specification not described in detail is prior art known to those skilled in the art.

Claims

1. A railway platform safety monitoring method, characterized in that: The following steps are involved: Assume that a train platform includes several sensing devices, based on which platform environment data is collected in real time, and several pieces of platform environment data are obtained each time; the several pieces of platform environment data are combined to obtain an initial platform environment data set; The initial platform environment data set is screened to obtain a safe target detection data set; the safe target detection data set is divided into safe detection data subsets Z according to target type. i , i=1, 2, 3; wherein, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the item data subset; For the security detection data subset Z i Abnormal feature screening is performed separately to obtain abnormal flying object dataset, abnormal personnel dataset and abnormal item dataset; Anomaly assessments are performed on the abnormal flying object dataset, abnormal personnel dataset, and abnormal object dataset to obtain the comprehensive flying object hazard level, comprehensive personnel hazard level, and comprehensive object hazard level; Based on the comprehensive flying object hazard level, comprehensive personnel hazard level and comprehensive object hazard level, corresponding safety intervention operations are performed to complete a railway station platform safety monitoring.

2. A railway platform safety monitoring method according to claim 1, characterized in that: The security target detection dataset is divided into security detection data subsets Z according to target type. i The specific steps include: Perform target detection on the platform environment data in the initial platform environment dataset to obtain target identification labels; retain all platform environment data with target identification labels as yes to form a safe target detection dataset; Traverse the platform environment data in the safety target detection data set and segment it by target to obtain several platform target data; Target types are set to flying objects, personnel, and objects; Divide all station target data by target type to obtain the security detection data subset Z i , i=1, 2, 3; among them, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the item data subset.

3. A railway platform safety monitoring method according to claim 2, characterized in that: For the security detection data subset Z i The specific steps for screening abnormal features include: For the security detection data subset Z i Extract the features of the platform target data in the dataset to obtain the features of the platform target data; The platform target data features are identified using the pre-trained abnormal target recognition model; if the platform target data features are abnormal, the platform target data is identified according to the security detection data subset Z to which the platform target data belongs. i , update the platform target data to abnormal flying object data or abnormal personnel data or abnormal item data, add the abnormal flying object data to the abnormal flying object data set; add the abnormal personnel data to the abnormal personnel data set; add the abnormal item data to the abnormal item data set; if the platform target data feature is normal, delete the corresponding platform target data; Traverse all security detection data subsets Z i , we obtain the abnormal flying object dataset, abnormal personnel dataset and abnormal item dataset after abnormal feature screening.

4. A railway platform safety monitoring method according to claim 3, characterized in that: The specific steps for anomaly assessment of the abnormal flying object dataset include: For any abnormal flying object data in the abnormal flying object dataset, perform the following operations: Obtain the current train platform weather data and identify the current risk level threshold based on the train platform weather data; Analyze the trajectory change rate of abnormal flying object data to obtain the speed change rate and altitude change rate of the flying object; A virtual flight area division model is established based on the train platform, and the proximity of abnormal flying object data to the dangerous area in the virtual flight area division model is identified to obtain the flying object area proximity; Construct an aircraft anomaly scoring matrix based on the aircraft speed change rate, aircraft altitude change rate and aircraft area proximity; Perform weighted calculation on the flight object anomaly scoring matrix to obtain the flight object anomaly score; Compare the flight object anomaly score with the current risk level threshold to obtain the flight object hazard level of the abnormal flight object data; Traverse the abnormal flying object data set to obtain the comprehensive flying object hazard level.

5. A railway platform safety monitoring method according to claim 4, characterized in that: The specific steps for anomaly assessment on the abnormal person dataset include: For any abnormal person data in the abnormal person dataset, perform the following operations: Extract the first K frames of abnormal personnel data corresponding to the abnormal personnel data and obtain the personnel behavior unit sequence P, P={P k |k=1, 2, ..., K+1}, K+1 is the total length of the personnel behavior unit sequence P, P k Abnormal personnel behavior data; Abnormal personnel behavior data P k Extract the time series behavior characteristics and obtain the abnormal time series behavior characteristics T k ; Build a behavior pattern library that includes normal behavior patterns and abnormal behavior precursor patterns; The abnormal timing behavior characteristics T k Match with the behavior pattern library and calculate the abnormal personnel behavior score F k ; For the front Abnormal personnel behavior score F k , use the first weight coefficient to perform weighted averaging to obtain the first abnormal personnel behavior average score; for the latter Abnormal personnel behavior score F k ; Use the second weight coefficient to perform weighted averaging on the scores to obtain the second average score of abnormal personnel behavior; represents the floor function; The average of the first abnormal person behavior score and the second abnormal person behavior score is taken as the person abnormality score; the person danger level is matched according to the person abnormality score; Traverse the abnormal personnel data set to obtain the comprehensive personnel danger level.

6. A railway platform safety monitoring method according to claim 5, characterized in that: The specific steps for abnormality assessment of abnormal item datasets include: For any abnormal item data in the abnormal item dataset, perform the following operations: Identify the location of abnormal items based on abnormal item data; Evaluate the item anomaly score based on the abnormal item data and the abnormal item placement location; identify the item's danger level based on the item anomaly score; Traverse the item hazard levels to obtain the comprehensive item hazard level.

7. A railway platform safety monitoring system, characterized in that: The system applies a railway platform safety monitoring method according to any one of claims 1 to 6, including: The environmental data acquisition module is used to assume that the train platform includes several sensing devices, collect platform environmental data in real time based on the several sensing devices, and obtain several pieces of platform environmental data each time; combine the several pieces of platform environmental data to obtain an initial platform environmental data set; The platform target screening module is used to screen the initial platform environment data set to obtain a safe target detection data set; the safe target detection data set is divided into safe detection data subsets Z according to the target type i , i=1, 2, 3; wherein, the safety detection data subset Z1 represents the flying object data subset, the safety detection data subset Z2 represents the personnel data subset, and the safety detection data subset Z3 represents the item data subset; Abnormal target recognition module, used to detect the security data subset Z i Abnormal feature screening is performed separately to obtain abnormal flying object datasets, abnormal personnel datasets, and abnormal object datasets; abnormal evaluation is performed on the abnormal flying object datasets, abnormal personnel datasets, and abnormal object datasets respectively to obtain comprehensive flying object hazard levels, comprehensive personnel hazard levels, and comprehensive object hazard levels; The safety monitoring intervention module is used to perform corresponding safety intervention operations based on the comprehensive flying object hazard level, the comprehensive personnel hazard level and the comprehensive object hazard level to complete a railway station platform safety monitoring.

Images