System and method for pre-warning personnel inclined line during train coming at loading station

Abstract

The invention provides an early warning system and method for personnel line inclination during train coming at a loading station, and the method comprises the steps: extracting a dynamic behavior feature sequence of personnel from a real-time monitoring video stream of the personnel; determining initial line-crossing risk values of the personnel at different moments in the vehicle coming process of the loading station according to static distance characteristics between the personnel in the operation area and a preset inclined line boundary; determining a cross-dimension correlation deviation degree between all the preliminary line-crossing risk values and the driving track data of the coming vehicle according to the space-time correlation between all the preliminary line-crossing risk values and the driving track data of the coming vehicle in the driving risk period; according to the correlation deviation degree, all the preliminary line crossing risk values are subjected to self-adaptive weighting adjustment, and then comprehensive line inclination risk features of the personnel are determined; and sending a personnel line inclination early warning signal to a loading station monitoring center based on the comprehensive line inclination risk characteristics. By adopting the scheme of the invention, the accuracy of early warning of line inclination of the personnel when the vehicle comes at the loading station can be improved based on dynamic risk assessment of time-space association between the personnel in the loading station and the coming vehicle.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of personnel line intrusion early warning, and more particularly to a personnel line intrusion early warning system and method for a train arriving at a loading station. BACKGROUND

[0002] Personnel line intrusion refers to the behavior of a part of the body of an operating personnel exceeding a safety control line (such as a loading yellow line or a train approaching line) and entering a dangerous area. When a vehicle (such as a train or a freight vehicle) is about to enter or has entered the operating area of a loading station, by monitoring the position, posture and activity track of personnel in the operating area of the loading station, it is automatically identified whether there is a risk of personnel exceeding the prescribed safety line (i.e., line intrusion) or approaching the running track of the vehicle, and when an abnormal situation is detected, an early warning prompt is sent to the operating personnel or the management system in time to prevent personnel from mistakenly entering a dangerous area of train operation and to protect personal safety.

[0003] Personnel line intrusion early warning when a train arrives at a loading station refers to, when a train or a vehicle enters the platform, through sensing the arrival state of the train and identifying the behavior of personnel crossing the line, realizing real-time early warning of personnel in the dangerous operating area of the loading station, protecting personal safety and improving the safety automation level of the loading site. However, in the prior art, the personnel line intrusion early warning method relies on single-dimensional monitoring, such as determining whether personnel cross the line only through a camera or issuing a fixed early warning only according to the distance of the arriving train, and determining the risk only according to the Euclidean distance between personnel and the boundary, ignoring the amplification effect of dynamic behaviors such as personnel running and falling on the risk level, resulting in insufficient consideration of the spatio-temporal correlation of personnel dynamic behaviors (such as rapid approach and body forward leaning) and the movement state of the arriving train (such as speed and real-time distance from the line), causing one-sided risk assessment, late warning opportunity or high false alarm rate and other problems. For example, when personnel are not crossing the line but are rapidly approaching the line, and the arriving train is also accelerating to approach the line, the prior art cannot comprehensively identify the high risk in time based on the dynamic relationship between the two. SUMMARY

[0004] The present application provides a personnel line intrusion early warning system and method for a train arriving at a loading station, which can realize dynamic risk assessment based on the spatio-temporal correlation of personnel and the arriving train in the loading station to improve the accuracy of personnel line intrusion early warning when the train arrives at the loading station.

[0005] In a first aspect, the present application provides a personnel line intrusion early warning method for a train arriving at a loading station, comprising the following steps: synchronously collecting real-time driving track data of the arriving train at the loading station and real-time monitoring video stream of personnel in the operating area; extract a dynamic behavior feature sequence of the personnel from the real-time monitoring video stream, and further determine a static distance feature between the personnel in the work area and the preset tilt line boundary, and determine a preliminary overline risk value of the personnel at different time points in the train loading station train arrival process through the static distance feature in combination with the dynamic behavior feature sequence; perform outlier point elimination on the real-time running track data of the train, and further extract a running risk period of the train reaching the preset tilt line boundary from the eliminated train running track data, and determine a cross-dimension correlation deviation between all preliminary overline risk values and the train running track data through the spatio-temporal correlation of all preliminary overline risk values and the train running track data in the running risk period; perform adaptive weighted adjustment on all preliminary overline risk values in the running risk period according to the correlation deviation, and further determine a comprehensive tilt line risk feature of the personnel according to all adjusted overline risk values; send a personnel tilt line early warning signal to the train loading station monitoring center based on the comprehensive tilt line risk feature.

[0006] In some embodiments, the extraction of the dynamic behavior feature sequence of the personnel from the real-time monitoring video stream specifically includes: perform personnel identification and positioning on each frame of image in the real-time monitoring video stream to obtain the bounding box position of the personnel in each frame of image; perform time sequence tracking on the personnel based on each bounding box position to obtain the position change of the personnel between different frames; estimate the key skeleton points of the detected personnel through the position change of the personnel between different frames to obtain the dynamic behavior feature sequence of the personnel.

[0007] In some embodiments, the determination of the static distance feature between the personnel in the work area and the preset tilt line boundary specifically includes: determine the actual spatial position of the personnel in the train loading station work area from the personnel bounding box identified in each frame of image in the real-time monitoring video stream; take the spatial distance between the actual position of the personnel in the train loading station work area and the preset tilt line boundary as the static distance feature between the personnel in the work area and the preset tilt line boundary.

[0008] In some embodiments, the determination of the preliminary overline risk value of the personnel at different time points in the train loading station train arrival process through the static distance feature in combination with the dynamic behavior feature sequence specifically includes: determine the overline behavior intention of the personnel at different time points in the train loading station train arrival process based on the static distance feature in combination with the dynamic behavior feature sequence; determine the preliminary overline risk value of the personnel at different time points in the train loading station train arrival process through the overline behavior intention of the personnel at different time points.

[0009] In some embodiments, determining the cross-dimensional correlation deviation between all initial line-crossing risk values ​​and oncoming vehicle trajectory data through the spatiotemporal correlation of all initial line-crossing risk values ​​and oncoming vehicle trajectory data within the driving risk period specifically includes: All preliminary risk values ​​of crossing the line and the trajectory data of oncoming vehicles are spatiotemporally aligned within the specified risk period to obtain spatiotemporally aligned data of personnel and oncoming vehicles within the specified risk period. Based on the spatiotemporal alignment data, the temporal correlation strength between personnel and vehicles during the arrival process at the loading station is determined in the time dimension and the spatial correlation strength in the spatial dimension. The cross-dimensional correlation deviation between all preliminary line-crossing risk values ​​and oncoming vehicle trajectory data is determined by the temporal correlation strength and the spatial correlation strength.

[0010] In some embodiments, adaptively weighting all preliminary lane-crossing risk values ​​within the driving risk period based on the correlation deviation specifically includes: The adjustment coefficients in the adaptive weighted adjustment process are determined based on the aforementioned correlation deviation. The time decay factor is determined based on the starting point of the aforementioned driving risk period; Based on the adjustment coefficient and the time decay factor, all preliminary lane crossing risk values ​​within the driving risk period are adaptively weighted and adjusted to obtain the adjusted lane crossing risk values.

[0011] In some embodiments, determining an individual's overall tilt risk characteristics based on all adjusted overshoot risk values ​​specifically includes: Statistical features were extracted from all adjusted risk values ​​of crossing the line to obtain the statistical features of the risk of personnel leaning over the line when a train arrives at the loading station; The overall tilt profile of an individual is determined by the aforementioned risk statistical characteristics.

[0012] Secondly, this application provides a passenger tilt warning system when a train approaches at a loading station, comprising: The data acquisition module is used to simultaneously collect real-time driving trajectory data of vehicles arriving at the loading station and real-time monitoring video streams of personnel in the work area; The processing module is used to extract the dynamic behavior feature sequence of personnel from the real-time monitoring video stream, and then determine the static distance feature between personnel and the preset tilt line boundary in the work area. The static distance feature is combined with the dynamic behavior feature sequence to determine the preliminary risk value of personnel crossing the line at different times during the arrival of the loading station. The processing module is also used to remove anomalies from the real-time driving trajectory data of the approaching vehicle, and then extract the driving risk period when the approaching vehicle reaches the preset inclined line boundary from the removed driving trajectory data. The cross-dimensional correlation deviation between all preliminary cross-line risk values ​​and the approaching vehicle driving trajectory data is determined by the spatiotemporal correlation between all preliminary cross-line risk values ​​and the approaching vehicle driving trajectory data during the driving risk period. The processing module is also used to adaptively weight and adjust all preliminary lane-crossing risk values ​​within the driving risk period based on the correlation deviation, and then determine the comprehensive lane-crossing risk characteristics of personnel based on all adjusted lane-crossing risk values. The execution module is used to send a personnel tilt warning signal to the loading station monitoring center based on the comprehensive tilt risk characteristics.

[0013] Thirdly, this application provides a computer device, which includes a memory and a processor. The memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the computer device performs the above-described method for warning of personnel tilting when a train arrives at the loading station.

[0014] Fourthly, this application provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the aforementioned method for warning of personnel tilting when a train approaches at a loading station.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In this application, real-time driving trajectory data of trains arriving at the loading station and real-time monitoring video streams of personnel within the work area are collected simultaneously. Dynamic behavioral feature sequences of personnel are extracted from the real-time monitoring video streams to determine the static distance characteristics between personnel and the preset tilt line boundary within the work area. The static distance characteristics, combined with the dynamic behavioral feature sequences, determine the initial tilt line risk values ​​of personnel at different times during the train's arrival at the loading station. Anomalies are removed from the real-time driving trajectory data of the arriving trains, and the driving risk period for the trains reaching the preset tilt line boundary is extracted from the removed train trajectory data. The spatiotemporal correlation between all initial tilt line risk values ​​and the train trajectory data within the driving risk period is used to determine the cross-dimensional correlation deviation between all initial tilt line risk values ​​and the train trajectory data. Based on the correlation deviation, all initial tilt line risk values ​​within the driving risk period are adaptively weighted and adjusted, and the comprehensive tilt line risk characteristics of personnel are determined based on all adjusted tilt line risk values. Based on the comprehensive tilt line risk characteristics, a personnel tilt line warning signal is sent to the loading station monitoring center.

[0016] Therefore, this application demonstrates that, firstly, by combining the static distance features with the dynamic behavioral feature sequence to determine the initial risk values ​​of personnel crossing the line at different times during the arrival of a vehicle at the loading station, it effectively addresses the one-sidedness of traditional methods that rely solely on Euclidean distance to judge the risk of personnel leaning towards the line. This allows for dynamic response to changes in personnel behavior, improving the sensitivity of risk perception, especially in early identification when personnel have not yet crossed the line but exhibit a high-risk behavioral trend, laying the foundation for subsequent accurate early warning and enhancing the foresight of risk assessment. Secondly, by determining the cross-dimensional correlation deviation between all initial risk values ​​and the vehicle's trajectory data within the risk period, and by using spatiotemporal alignment to identify whether personnel and vehicles overlap in the same time window or spatially adjacent areas, the accuracy of personnel leaning towards the line risk prediction can be improved. Furthermore, by quantifying the danger level of human-vehicle interaction based on the correlation strength in time and space dimensions, it is possible to dynamically distinguish between high-risk approach and irrelevant contact. The system employs a near-realistic behavioral scenario to effectively mitigate false alarms and missed alarms, thereby achieving dynamic tilt risk perception in a multi-dimensional interactive context. Then, based on the correlation deviation, all preliminary tilt risk values ​​within the driving risk period are adaptively weighted and adjusted. Finally, based on all adjusted tilt risk values, the comprehensive tilt risk characteristics of personnel are determined, improving the effectiveness and scenario relevance of personnel tilt risk values. This makes risk judgments more consistent with the temporal evolution of real-world scenarios, providing support for the formation of highly reliable behavioral early warning data. Finally, based on the comprehensive tilt risk characteristics, a personnel tilt warning signal is sent to the loading station monitoring center, enabling on-site management personnel within the loading station to react quickly, shifting from post-event response to pre-event intervention, significantly improving the level of automated safety control and risk response efficiency at the loading station. In summary, this solution can achieve dynamic risk assessment based on the spatiotemporal correlation between personnel within the loading station and approaching vehicles, improving the accuracy of personnel tilt warnings when vehicles arrive at the loading station. Attached Figure Description

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

[0018] Figure 1 This is an exemplary flowchart of a method for warning of personnel tilting when a vehicle approaches at a loading station, according to some embodiments of this application; Figure 2 This is an exemplary flowchart illustrating the extraction of dynamic behavioral feature sequences according to some embodiments of this application; Figure 3This is an exemplary flowchart illustrating the determination of an initial risk value for crossing the line, according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a passenger tilt warning system when a train approaches at a loading station, according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a method for warning of personnel tilting when a train approaches at a loading station, according to some embodiments of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] refer to Figure 1 The figure is an exemplary flowchart of a method for warning of passenger tilting when a train arrives at a loading station, according to some embodiments of this application. The method mainly includes the following steps: In step 101, real-time driving trajectory data of vehicles arriving at the loading station and real-time monitoring video streams of personnel in the work area are collected simultaneously.

[0021] In practice, the simultaneous acquisition of real-time driving trajectory data of trains arriving at the loading station and real-time monitoring video streams of personnel within the work area can be achieved in the following way: In the train arrival lane area of ​​the loading station, high-precision positioning terminals of Global Navigation Satellite System (GNSS) (such as Real-Time Kinematic-Global Positioning System) can be deployed above or on both sides of the train path. A system (RTK-GPS) or lidar acquisition device is used to capture the spatial position changes of the vehicle arriving at the loading station throughout its journey, generating a real-time sequence of timestamped trajectory points. Each trajectory point includes the vehicle's X, Y (or X, Y, Z) coordinates at a specific moment and the corresponding timestamp, thus obtaining the real-time driving trajectory data of the vehicle arriving at the loading station. Simultaneously, high-definition network video cameras are deployed above the work area to provide full coverage monitoring. These cameras continuously acquire video image streams within the work area at a frame rate of 25 to 30 frames per second, obtaining real-time monitoring video streams of personnel within the work area. Furthermore, to ensure consistency in the time dimension between the two types of data, all GNSS terminals and video acquisition devices are connected to a unified Network Time Protocol (NTP). A Protocol (NTP) server, or a GPS timing module, can be configured to uniformly align the timestamps of the collected data (i.e., synchronize the data, which means unifying the time reference of data collected by different devices to the same standard time base). After the above steps, the real-time driving trajectory data of the train arriving at the loading station and the real-time monitoring video stream of personnel in the work area with the same time can be obtained synchronously, thus providing basic spatiotemporal input data for subsequent fusion analysis. Other methods can also be used to obtain the data in other embodiments, which are not specifically limited here.

[0022] It should be noted that the real-time driving trajectory data of the approaching vehicle in this application represents a sequence of spatial location points continuously recorded over time during the actual driving process of the vehicle in the loading station, which includes the coordinate position and timestamp at each moment; the real-time monitoring video stream in this application refers to a sequence of image frames that are captured in real time and continuously output by video monitoring equipment deployed above the loading station's operating area, which has a temporal order and can reflect the dynamic changes on-site in the loading station's operating area.

[0023] In step 102, the dynamic behavior feature sequence of personnel is extracted from the real-time monitoring video stream, and then the static distance feature between personnel and the preset tilt line boundary in the work area is determined. The static distance feature is combined with the dynamic behavior feature sequence to determine the preliminary risk value of personnel crossing the line at different times during the arrival of the loading station.

[0024] In some embodiments, reference Figure 2As shown, this figure is an exemplary flowchart of extracting dynamic behavioral feature sequences in some embodiments of this application. In this embodiment, the extraction of dynamic behavioral feature sequences of personnel from the real-time monitoring video stream can be achieved by the following steps: First, in step 1021, personnel identification and localization are performed on each frame of the real-time monitoring video stream to obtain the bounding box position of the personnel in each frame of the image. Secondly, in step 1022, the personnel are time-series tracked based on the positions of each bounding box to obtain the positional changes of the personnel between different frames; Finally, in step 1023, the key skeletal points of the detected person are estimated by the positional changes of the person between different frames to obtain the dynamic behavior feature sequence of the person.

[0025] In specific implementation, personnel identification and localization are performed on each frame of the real-time monitoring video stream to obtain the bounding box position of the personnel in each frame. This can be achieved in the following way: a deep convolutional neural network model (e.g., object detection algorithm (You Only Look Once Version 8, YOLOv8)) can be used to identify and locate personnel in each frame of the real-time monitoring video stream to obtain the pixel position of the detected personnel in each frame. The convolutional neural network model performs multi-scale feature extraction and region prediction on each input frame, outputting the bounding box coordinates of all personnel in each frame as the bounding box position of the personnel in each frame, thus accurately identifying all personnel targets within the work area. Based on the bounding box positions, personnel are tracked temporally to obtain the positional changes of personnel between different frames. This can be achieved in the following way: a multi-target tracking algorithm (e.g., Deep Simple Online and Realtime) can be used. DeepSORT tracks and matches the positions of people at each bounding box location in a real-time video stream, performing identity preservation and temporal matching between consecutive frames to obtain the positional changes of people across different frames. DeepSORT integrates Kalman filter prediction, Hungarian algorithm for inter-frame matching, and introduces Person Re-Identification (ReID) features for identity preservation to accurately obtain the continuous motion trajectory of each person. The key skeleton points of the detected people are estimated based on the positional changes of the people across different frames, yielding a sequence of dynamic behavioral features. This can be achieved using human pose estimation algorithms (such as OpenPose or MediaPipe). The Pose function extracts the key skeletal point positions of the detected personnel. The human pose estimation algorithm predicts feature maps based on convolutional neural networks and locates two-dimensional or three-dimensional key points through heatmaps and skeleton connections. Combined with the positional changes of the personnel between different frames, the algorithm further calculates the changing trend of the personnel's key skeletal points over time to extract dynamic behavioral features such as limb orientation (based on head-torso vector) and movement speed vector (based on foot or pelvic key point displacement). All the extracted dynamic behavioral features are then arranged into a dynamic behavioral feature sequence of the personnel according to their corresponding time sequence. Other methods may be used to determine the position in other embodiments, which are not limited here.

[0026] It should be noted that, in this application, the bounding box position refers to the spatial region in each frame of the real-time monitoring video stream that represents the location of personnel within the loading station operation area in the form of a rectangular box, which is used for subsequent personnel target tracking and attitude estimation; in this application, the positional changes of personnel between different frames represent the continuous movement trajectory of personnel within the loading station operation area; in this application, key skeleton points refer to representative positional nodes on the human body, such as the head, shoulders, knees, and feet, which are used to characterize the body posture and behavioral trends of personnel within the loading station operation area; in this application, the dynamic behavioral feature sequence refers to the ordered set of behavioral characteristics exhibited by personnel within the loading station operation area as their actions change over a continuous time period, which is used to characterize the behavioral intentions and risk trends of personnel within the operation area.

[0027] In some embodiments, determining the static distance characteristics between personnel within the work area and a preset tilt line boundary can be achieved through the following steps: The actual spatial location of personnel within the loading station's work area is determined by the personnel bounding boxes identified in each frame of the real-time monitoring video stream. The spatial distance between the actual position of the personnel in the loading station operation area and the preset tilt line boundary is used as the static distance feature between the personnel in the operation area and the preset tilt line boundary.

[0028] It should be noted that the inclined line boundary in this application refers to the safety control line set artificially in the loading station operation area to delineate the area that personnel cannot approach or enter. The inclined line boundary is a continuous set of line segments in the form of several spatial coordinate points, which serves as a safety warning line that prohibits personnel from crossing within the loading station operation area.

[0029] In practice, determining the actual position of personnel within the loading station's work area from the personnel bounding boxes identified in each frame of the real-time monitoring video stream can be achieved in the following way: First, using common techniques such as the Zhang Zhengyou calibration method, the intrinsic parameters (e.g., focal length, principal point) and extrinsic parameters (e.g., shooting angle, relative position) of the high-definition network video cameras deployed above the work area are calibrated, establishing a homography matrix or perspective projection transformation model between the image coordinate system and the ground world coordinate system; then, the center point or foot pixel position of the personnel bounding boxes identified in each frame of the real-time monitoring video stream is converted into two-dimensional actual position coordinates on the ground plane within the work area using this transformation model, and this coordinate serves as the actual position of the personnel within the loading station's work area. The spatial distance between the actual spatial position of personnel in the loading station operation area and the preset inclined line boundary can be used as the static distance feature between personnel in the operation area and the preset inclined line boundary. This can be achieved by calculating the geometric shortest distance between the actual spatial position of personnel in the loading station operation area and all line segments of the preset inclined line boundary (e.g., using the point-to-line segment shortest distance formula). This will yield the spatial distance between the actual spatial position of personnel in the loading station operation area and the preset inclined line boundary at the corresponding time of each frame in the real-time monitoring video stream. The obtained spatial distance can then be used as the static distance feature between personnel in the operation area and the preset inclined line boundary. Other methods can also be used to determine this distance in other embodiments, which are not limited here.

[0030] It should be noted that the actual spatial location in this application refers to the real two-dimensional coordinates of the pixel position of the personnel in the real-time monitoring video stream in the ground space within the loading station operation area, which is used to reflect the specific physical location of the personnel within the loading station operation area; the static distance feature in this application refers to the minimum geometric distance between the personnel and the boundary line of the loading station operation area on the ground plane, which is a key value for measuring whether the personnel are close to the danger zone.

[0031] In some embodiments, reference Figure 3 As shown, this figure is an exemplary flowchart for determining the initial risk value of crossing the line in some embodiments of this application. The determination of the initial risk value of personnel crossing the line at different times during the arrival of a train at the loading station by combining the static distance features with the dynamic behavioral feature sequence can be achieved by the following steps: Based on the static distance features combined with the dynamic behavioral feature sequence, the intention of personnel to cross the line at different times during the arrival of a vehicle at the loading station is determined. The initial risk value of personnel crossing the line at different times is determined by observing their intentional actions of crossing the line during the process of a vehicle arriving at the loading station.

[0032] In specific implementation, determining the intention of personnel to cross the line at different times during the arrival of a vehicle at the loading station based on the static distance features and the dynamic behavior feature sequence can be achieved in the following way: the static distance features and dynamic behavior feature sequence at the corresponding time of each frame in the real-time monitoring video stream can be fused. This feature fusion can be achieved by constructing a multi-dimensional feature vector, where the feature vector at each time moment consists of [spatial distance, orientation angle, velocity magnitude, and the angle between the velocity direction and the boundary]. The temporal feature sequence can then be input into a lightweight temporal classification model (such as a Long Short-Term Memory network or a Multilayer Perceptron). (Perceptron, LSTM+MLP structure) can output the probability of crossing the line at each time step, thus using the output as the crossing behavior intention of personnel at different times during the arrival of a train at the loading station, reflecting whether the personnel are showing an intention to move towards the inclined line boundary at the corresponding time step. The preliminary crossing risk value of personnel at different times during the arrival of a train at the loading station can be determined by the following method: First, the crossing behavior intention of personnel at different times can be normalized and the approach speed weight of the personnel can be added (e.g., the greater the speed at which the personnel move towards the preset inclined line boundary, the higher the crossing risk). (High), and then, based on the above processing results, the evaluation algorithm scores the risk of personnel crossing the line at the corresponding time of each frame in the real-time monitoring video stream, and uses the obtained score as the preliminary risk value of personnel crossing the line at different times during the arrival of the loading station. The preliminary risk value of crossing the line fluctuates between 0 and 1, which comprehensively reflects the contribution of the personnel's position and action trend at different times to the possibility of crossing the line. It is an important input indicator for subsequent risk aggregation and early warning judgment. In addition, the evaluation algorithm is, for example, genetic algorithm, ensemble learning and reinforcement learning, etc. Other methods can also be used in other embodiments, which are not limited here.

[0033] It should be noted that the intention to cross the line in this application refers to the tendency of personnel to cross the boundary of the inclined line at a given moment, which is a predictive output at the behavioral level; the preliminary risk value of crossing the line in this application represents the probability of personnel crossing the line during the process of a vehicle approaching the loading station in the work area. It is used to reflect the degree of potential danger of personnel crossing the line in the work area and is the core quantitative result of the early assessment of the risk of crossing the line.

[0034] In step 103, anomalies are removed from the real-time driving trajectory data of the approaching vehicle. Then, the driving risk period when the approaching vehicle reaches the preset inclined line boundary is extracted from the removed driving trajectory data. The cross-dimensional correlation deviation between all preliminary crossing risk values ​​and the approaching vehicle driving trajectory data within the driving risk period is determined by the spatiotemporal correlation between all preliminary crossing risk values ​​and the approaching vehicle driving trajectory data.

[0035] It should be noted that, in this application, outliers refer to trajectory points in the real-time driving trajectory data of an approaching vehicle that deviate from the normal movement trend of the approaching vehicle due to abnormal interference factors (such as errors, GNSS signal interference, equipment failure, etc.). Specifically, the outlier removal from the real-time driving trajectory data of the approaching vehicle can be achieved in the following way: speed consistency detection and position jump detection are performed on the real-time driving trajectory data of the approaching vehicle to detect and remove jump anomalies caused by positioning errors or signal interference. For example: firstly, the position change between every two adjacent moments in the real-time driving trajectory data and its corresponding instantaneous speed are calculated. The speed is then filtered by a sliding window mean to determine if there is any abnormal acceleration that deviates significantly from the mean (e.g., exceeding the threshold 3σ). A trajectory direction consistency judgment is further introduced. If the movement direction between the current point and the points before and after it changes drastically or jumps (with an angle close to 180°), the point is marked as abnormal. All marked abnormal points are then smoothly replaced by normal trajectory points before and after them using time interpolation (e.g., spline interpolation or linear interpolation). Finally, the continuous and smooth vehicle trajectory data after removing abnormal points and correction is output. Other methods can also be used in other embodiments, which are not limited here.

[0036] In specific implementation, extracting the driving risk period of an oncoming vehicle reaching the preset inclination boundary from the eliminated oncoming vehicle trajectory data can be achieved in the following way: First, based on the vehicle spatial coordinates at each moment in the eliminated oncoming vehicle trajectory data, combined with the spatial coordinates of the preset inclination boundary, calculate the Euclidean distance between the current position of the oncoming vehicle and the inclination boundary, and record the trend of this distance changing over time; next, perform differential analysis on the oncoming vehicle trajectory data to obtain the vehicle's speed and direction at each moment; then, based on the vehicle's current position, speed, and acceleration information, apply a constant velocity model or a constant acceleration model (this model refers to using...) The vehicle's current state information, based on the assumption of constant speed or constant acceleration, predicts the trajectory of the approaching vehicle over a future period using mathematical models. This predicts when the approaching vehicle will enter the adjacent buffer zone (e.g., within 3 meters) of the inclined boundary at the earliest. Finally, the time period between several seconds (e.g., 5 seconds) before the predicted arrival time and several seconds (e.g., 2 seconds) after the approaching vehicle actually passes the inclined boundary is defined as the driving risk period for the approaching vehicle to reach the preset inclined boundary. This driving risk period reflects the time range in which the approaching vehicle is in a high-probability intrusion or near-intrusion state. Other methods can also be used in other embodiments, which are not limited here.

[0037] It should be noted that the driving risk period in this application refers to the time interval from when an oncoming vehicle enters the risk distance range to when it completely leaves. During this driving risk period, the risk of people approaching the edge of the slope is significantly increased.

[0038] In some embodiments, determining the cross-dimensional correlation deviation between all initial line-crossing risk values ​​and oncoming vehicle trajectory data by assessing the spatiotemporal correlation between all initial line-crossing risk values ​​and oncoming vehicle trajectory data within the driving risk period can be achieved through the following steps: All preliminary risk values ​​of crossing the line and the trajectory data of oncoming vehicles are spatiotemporally aligned within the specified risk period to obtain spatiotemporally aligned data of personnel and oncoming vehicles within the specified risk period. Based on the spatiotemporal alignment data, the temporal correlation strength between personnel and vehicles during the arrival process at the loading station is determined in the time dimension and the spatial correlation strength in the spatial dimension. The cross-dimensional correlation deviation between all preliminary line-crossing risk values ​​and oncoming vehicle trajectory data is determined by the temporal correlation strength and the spatial correlation strength.

[0039] In specific implementation, all preliminary line-crossing risk values ​​and oncoming vehicle trajectory data are spatiotemporally aligned within the stated driving risk period to obtain the spatiotemporally aligned data of personnel and oncoming vehicles within the stated driving risk period. This can be achieved by aligning all preliminary line-crossing risk values ​​and oncoming vehicle trajectory data according to a unified timestamp, and using the aligned data as the spatiotemporally aligned data of personnel and oncoming vehicles within the stated driving risk period. This alignment can be achieved through linear interpolation or a time synchronization mechanism to ensure that the two dimensions of data are comparable at the same time. Based on the spatiotemporally aligned data, the temporal correlation strength between personnel and oncoming vehicles in the time dimension and the spatial correlation strength in the spatial dimension during the arrival process at the loading station can be determined by using a sliding window mechanism or a Kalman filter tracker to perform inter-frame segmentation on the spatiotemporally aligned data. The analysis involves calculating the correlation between the gradient change of the initial risk value of crossing the line at each moment and the speed / acceleration of the oncoming vehicle, as well as the rate of change of the distance between the oncoming vehicle and the slope line (this correlation can be quantified using Pearson correlation coefficient, mutual information, or sequence similarity score). The calculated correlation is used as the temporal association strength between personnel and oncoming vehicles during the arrival process at the loading station. In addition, in the spatial dimension, the directional consistency between the personnel-oncoming vehicle spatial vector and the change vector of the risk value of crossing the line is calculated based on the Euclidean distance, relative angle, and velocity vector angle between the personnel and oncoming vehicle positions in each frame (this directional consistency can be quantified using cosine similarity). The calculated directional consistency is used as the spatial association strength between personnel and oncoming vehicles during the arrival process at the loading station. Other methods can also be used in other embodiments, which are not limited here.

[0040] It should be noted that the spatiotemporal correlation in this application refers to the degree of mutual influence between personnel crossing risk data and oncoming vehicle trajectory data in terms of temporal trends and spatial location relationships. Furthermore, this application characterizes the spatiotemporal correlation between all preliminary crossing risk values ​​and oncoming vehicle trajectory data during the risk period by assessing the temporal correlation strength and spatial correlation strength between personnel and oncoming vehicles during the loading station's oncoming process. Specifically, determining the cross-dimensional correlation deviation between all preliminary crossing risk values ​​and oncoming vehicle trajectory data through the temporal and spatial correlation strengths can be achieved in the following way: First, an ideal correlation model can be constructed based on historical safety data, such as a theoretical curve indicating that "the faster the oncoming vehicle speed and the closer it is to the preset inclination boundary, the higher the personnel crossing risk value should be." Then, a 3-second sliding window can be used to calculate the temporal correlation. The strength and spatial correlation strength are the deviation values ​​(such as residual sum of squares) between the actual correlation strength and the ideal correlation model. Finally, the deviation values ​​of all windows are weighted and summed (where the weight of temporal correlation deviation can be set to 0.6 and the weight of spatial correlation deviation to 0.4, and the weight can be dynamically adjusted based on Bayesian optimization or grid search). The final result is used as the cross-dimensional correlation deviation degree between all preliminary cross-line risk values ​​and oncoming vehicle trajectory data. The correlation deviation degree ranges from 0 to 1. The larger the value, the more significant the deviation between the actual correlation and the ideal correlation, indicating that the risk level is more detached from the real dangerous situation. This application uses the correlation deviation degree to reflect the reasonableness of the coupling between the preliminary cross-line risk value and the oncoming vehicle status during the driving risk period, and assists the subsequent personnel in determining the tilt warning. Other methods can also be used to determine this in other embodiments, which are not limited here.

[0041] It should be noted that the spatiotemporal alignment data in this application refers to the structured information sequence formed by pairing personnel crossing risk values ​​from different sources with oncoming vehicle trajectory data under the same time reference; the temporal correlation strength in this application refers to the degree of similar rhythm or trend between changes in personnel behavior risk and the driving status of oncoming vehicles in time, which is used to determine whether oncoming vehicles and personnel behavior are responsive or coupled; the spatial correlation strength in this application refers to the degree of dangerous proximity between personnel and oncoming vehicles in space, which is used to reflect the dynamic evolution of the possibility of spatial conflict between oncoming vehicles and personnel; the cross-dimensionality in this application refers to the fact that the personnel crossing risk data and the oncoming vehicle trajectory data belong to two different monitoring object dimensions: personnel behavior and oncoming vehicle trajectory; the correlation deviation degree in this application refers to the indicator that measures the degree of unreasonable deviation between the initial personnel crossing risk value and the oncoming vehicle behavior data, which is used to measure the rationality of the initial crossing risk value of loading station personnel during the oncoming vehicle process, and is a key intermediate variable for achieving highly reliable personnel tilt warning.

[0042] In step 104, all preliminary lane-crossing risk values ​​within the driving risk period are adaptively weighted and adjusted based on the correlation deviation, and then the comprehensive lane-crossing risk characteristics of the personnel are determined based on all the adjusted lane-crossing risk values.

[0043] In some embodiments, adaptively weighting all preliminary lane-crossing risk values ​​within the driving risk period based on the correlation deviation can be achieved using the following steps: The adjustment coefficients in the adaptive weighted adjustment process are determined based on the aforementioned correlation deviation. The time decay factor is determined based on the starting point of the aforementioned driving risk period; Based on the adjustment coefficient and the time decay factor, all preliminary lane crossing risk values ​​within the driving risk period are adaptively weighted and adjusted to obtain the adjusted lane crossing risk values.

[0044] In specific implementation, the adjustment coefficient in the adaptive weighted adjustment process based on the correlation deviation can be determined in the following way: the correlation deviation can be used as input to generate the adjustment coefficient in the adaptive weighted adjustment process through Sigmoid curve mapping. For example, if the correlation deviation is 0, the adjustment coefficient is 1.0 (i.e., no adjustment is needed); if the correlation deviation is 1, the adjustment coefficient is 2.0 (i.e., significantly amplifying the risk). In addition, if the correlation deviation is an intermediate value, it will smoothly transition according to the curve to ensure a dynamic mapping where the higher the correlation deviation, the greater the adjustment force. The time decay factor can be determined based on the starting point of the driving risk period, in the following way: based on the starting point of the driving risk period, the adjustment coefficient is decayed by 0.1 over time (e.g., every 3 seconds) (not less than 1.0), to ensure that the adjustment is more sensitive during critical periods when the approaching vehicle is close to the preset inclination boundary (e.g., within 10 meters of the inclination boundary), thereby introducing the time decay factor. The adjustment coefficient and the time decay factor are used to adjust the driving risk period... All initial risk values ​​for crossing the line within a segment are adaptively weighted and adjusted to obtain the adjusted risk value. This can be achieved by multiplying the initial risk value for crossing the line at each moment within the driving risk period by its corresponding adjustment coefficient and time decay factor, while setting upper and lower limits (e.g., 0-1.2) to prevent the adjusted value from exceeding a reasonable range. In some embodiments, the coefficient of variation (i.e., standard deviation / mean) of the adjusted risk value can be calculated using a 5-second sliding window. If the coefficient of variation is >0.3, the adjustment coefficient within the window is smoothed a second time (e.g., the average of adjacent windows is taken: the average of the adjustment coefficients at corresponding moments in the current window and the preceding and following adjacent windows is taken to replace the original adjustment coefficient) to avoid unreasonable fluctuations in the adjusted risk value due to sudden changes in correlation deviation. Finally, the adjusted risk value is output to more accurately express the degree of danger of personnel crossing the line in a dynamic environment and can serve as a direct basis for subsequent personnel tilt warning output. Other methods can also be used in other embodiments, which are not limited here.

[0045] It should be noted that the adjustment coefficient in this application refers to the weighting parameter used to amplify or reduce the initial risk value of exceeding the limit; the time decay factor in this application refers to the parameter that reduces the adjustment coefficient over time, which is used to reduce the influence weight of the risk value at a time far from the danger time, so as to highlight the risk during the critical period of the loading station's arrival process; the adaptive weighted adjustment in this application refers to the correction method that automatically changes the adjustment intensity according to the correlation deviation degree and time factor.

[0046] In some embodiments, determining an individual's overall tilt risk characteristics based on all adjusted tilt risk values ​​can be achieved using the following steps: Statistical features were extracted from all adjusted risk values ​​of crossing the line to obtain the statistical features of the risk of personnel leaning over the line when a train arrives at the loading station; The overall tilt profile of an individual is determined by the aforementioned risk statistical characteristics.

[0047] In specific implementation, statistical features are extracted from all adjusted risk values ​​for crossing the line. The statistical features of the risk of personnel leaning towards the loading station when a train approaches can be obtained in the following way: A 5-second sliding window (in some embodiments, a grid search or Bayesian optimization can be used to dynamically select the window (e.g., 3-10 seconds)) can be used to calculate the mean (reflecting the average risk level of that period), maximum (reflecting the instantaneous highest risk), duration (e.g., the continuous time during which the risk value exceeds 0.6, where 0.6 is a preset moderate risk threshold), and volatility (i.e., the ratio of the standard deviation to the mean, reflecting the degree of volatility of the risk of crossing the line) of the risk values ​​within each window. All the obtained statistical features are then combined. As a statistical feature of personnel tilting towards the loading station when a train arrives, the comprehensive tilting risk feature of personnel can be determined by the following method: a risk assessment model (such as support vector regression, logistic regression, or a simple rule engine) can be used to assess the risk of personnel's tendency to cross the line when a train arrives at the loading station based on the statistical feature. Finally, the comprehensive risk score and dynamic change assessment result of personnel in the loading station's operating area during the risk period of the train arrival can be used as the comprehensive tilting risk feature of personnel. The comprehensive risk score ranges from 0 to 1, and the higher the score, the higher the tilting risk. Other methods can also be used in other embodiments, which are not limited here.

[0048] It should be noted that the risk statistical characteristics in this application refer to the quantitative indicators extracted from the adjusted overshoot risk values ​​that can reflect the risk patterns of personnel tilting to the edge; the comprehensive tilt risk characteristics in this application refer to the comprehensive description of the risk level, fluctuations, and trends of personnel during the risk period of travel when the train arrives at the loading station, which is used to reflect the risk level and changing trend of personnel tilting to the edge during the process of the train arriving at the loading station.

[0049] In step 105, a personnel tilt warning signal is sent to the loading station monitoring center based on the comprehensive tilt risk characteristics.

[0050] In specific implementation, sending personnel tilt warning signals to the loading station monitoring center based on the comprehensive tilt risk characteristics can be achieved in the following way: First, the loading station safety regulations preset three-level warning thresholds corresponding to the comprehensive tilt risk characteristics, for example: low risk: comprehensive risk score ≤ 0.4; medium risk: 0.4 < comprehensive risk score ≤ 0.7; high risk: comprehensive risk score > 0.7, and associated response rules, that is: low risk is only recorded by the system, medium risk triggers on-site audio and visual prompts, and high risk is simultaneously pushed to the monitoring center; then, the comprehensive tilt risk characteristics of personnel are compared with the warning thresholds to determine the current personnel tilt risk level; subsequently, standardized warning messages (in JSON format, including key content such as warning level, trigger time, personnel location coordinates, real-time status of approaching vehicles, and dynamic changes) are generated for medium and high risks as personnel tilt warning signals; then, the warning signals are pushed to the monitoring server of the loading station monitoring center via industrial Ethernet (such as based on transmission control protocols and Internet protocols), and simultaneously stored in the database (such as Structured Query Language). QueryLanguage (MySQL) is used for historical tracing and analysis. After receiving the personnel tilt warning signal, the monitoring center highlights the location of the warned personnel and the real-time trajectory of the approaching vehicle on the electronic map, triggering an audible and visual alarm. Among them, high risk is accompanied by a voice prompt "Personnel are approaching the tilt line, and the approaching vehicle is about to arrive" and displays a warning signal containing risk details in a pop-up window on the monitoring terminal. At the same time, the sending time, receiving status and processing result of the warning signal are recorded to form a closed-loop management record. Other methods can also be used to implement this in other embodiments, which are not specifically limited here.

[0051] It should be noted that the warning threshold in this application represents the critical value used to classify the risk level of personnel tilting when a train arrives at the loading station.

[0052] In another aspect, in some embodiments, this application provides a passenger tilt warning system when a train approaches at a loading station, with reference to... Figure 4 The figure is a schematic diagram of a passenger tilt warning system for loading stations when a train arrives, according to some embodiments of this application. The passenger tilt warning system 400 for loading stations includes: a data acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to synchronously acquire real-time driving trajectory data of vehicles arriving at the loading station and real-time monitoring video streams of personnel in the work area; Processing module 402 in this application is mainly used to extract the dynamic behavior feature sequence of personnel from the real-time monitoring video stream, and then determine the static distance feature between personnel and the preset tilt line boundary in the work area. The static distance feature is combined with the dynamic behavior feature sequence to determine the preliminary risk value of personnel crossing the line at different times during the arrival of the loading station. The processing module 402 described in this application is also used to remove anomalies from the real-time driving trajectory data of the approaching vehicle, and then extract the driving risk period when the approaching vehicle reaches the preset tilt line boundary from the removed driving trajectory data. The cross-dimensional correlation deviation between all preliminary cross-line risk values ​​and the approaching vehicle driving trajectory data is determined by the spatiotemporal correlation between all preliminary cross-line risk values ​​and the approaching vehicle driving trajectory data during the driving risk period. The processing module 402 described in this application is also used to adaptively weight and adjust all preliminary lane crossing risk values ​​within the driving risk period based on the correlation deviation degree, and then determine the comprehensive lane crossing risk characteristics of the personnel based on all the adjusted lane crossing risk values; The execution module 403 in this application is mainly used to send a personnel tilt warning signal to the loading station monitoring center based on the comprehensive tilt risk characteristics.

[0053] The foregoing has detailed examples of the passenger tilt warning system and method for loading stations provided in this application. It is understood that the corresponding apparatus, in order to achieve the above functions, includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0054] In some embodiments, this application also provides a computer device, the computer device including a memory and a processor, the memory for storing a computer program, and the processor for calling and running the computer program from the memory, so that the computer device performs the above-described method for warning of personnel tilting when a train arrives at a loading station.

[0055] In some embodiments, reference Figure 5 The dashed lines in the figure indicate that the unit or module is optional. This figure is a schematic diagram of the structure of the computer equipment implementing the passenger tilt warning method for loading station arrival as described in this application. The passenger tilt warning method for loading station arrival in the above embodiments can be achieved through… Figure 5The computer device 500 shown is used to implement this, and the computer device 500 includes at least one processor 501, a memory 502 and at least one communication unit 505. The computer device 500 may be a terminal device, a server or a chip.

[0056] The processor 501 can be a general-purpose processor or a special-purpose processor. For example, the processor 501 can be a central processing unit (CPU). The CPU can be used to control the computer device 500, execute software programs, and process data from the software programs. The computer device 500 may also include a communication unit 505 for inputting (receiving) and outputting (transmitting) signals.

[0057] For example, computer device 500 may be a chip, communication unit 505 may be the input and / or output circuit of the chip, or communication unit 505 may be the communication interface of the chip, and the chip may be a component of terminal device, network device or other device.

[0058] For example, computer device 500 may be a terminal device or a server, and communication unit 505 may be a transceiver of the terminal device or the server, or communication unit 505 may be a transceiver circuit of the terminal device or the server.

[0059] The computer device 500 may include one or more memories 502 storing a program 504. The program 504 can be executed by a processor 501 to generate instructions 503, causing the processor 501 to perform the methods described in the above method embodiments according to the instructions 503. Optionally, the memory 502 may also store data (such as a target audit model). Optionally, the processor 501 may also read data stored in the memory 502, which may be stored at the same storage address as the program 504, or the data may be stored at a different storage address than the program 504.

[0060] The processor 501 and memory 502 can be configured separately or integrated together, for example, integrated on the system-on-chip (SOC) of the terminal device.

[0061] It should be understood that each step of the above method embodiment can be completed by hardware logic circuits or software instructions in the processor 501. The processor 501 can be a CPU, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, such as discrete gates, transistor logic devices, or discrete hardware components.

[0062] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] For example, in some embodiments, this application also provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the above-described method for warning of personnel tilting when a train approaches at a loading station.

[0064] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0065] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for early warning of personnel leaning to the line when a train approaches at a loading station, characterized in that, Includes the following steps: Simultaneously collect real-time driving trajectory data of vehicles arriving at the loading station and real-time monitoring video streams of personnel within the work area; The dynamic behavior feature sequence of personnel is extracted from the real-time monitoring video stream, and then the static distance feature between personnel and the preset tilt line boundary in the work area is determined. The static distance feature is combined with the dynamic behavior feature sequence to determine the initial risk value of personnel crossing the line at different times during the arrival of the loading station. The real-time driving trajectory data of the approaching vehicle is subjected to anomaly removal, and then the driving risk period when the approaching vehicle reaches the preset inclined line boundary is extracted from the removed approaching vehicle driving trajectory data. The cross-dimensional correlation deviation between all preliminary cross-line risk values ​​and approaching vehicle driving trajectory data is determined by the spatiotemporal correlation between all preliminary cross-line risk values ​​and approaching vehicle driving trajectory data within the driving risk period. Based on the correlation deviation, all preliminary lane-crossing risk values ​​within the driving risk period are adaptively weighted and adjusted, and then the comprehensive lane-crossing risk characteristics of personnel are determined based on all the adjusted lane-crossing risk values. Based on the comprehensive tilt risk characteristics, a personnel tilt warning signal is sent to the loading station monitoring center.

2. The method as described in claim 1, characterized in that, Extracting the dynamic behavioral feature sequence of personnel from the real-time monitoring video stream specifically includes: Personnel identification and localization are performed on each frame of the real-time monitoring video stream to obtain the bounding box position of the personnel in each frame of the image. Based on the location of each bounding box, the personnel are tracked in time to obtain the positional changes of the personnel between different frames; By estimating the key skeletal points of the detected person through the positional changes of the person between different frames, a dynamic behavioral feature sequence of the person is obtained.

3. The method as described in claim 1, characterized in that, Determining the static distance characteristics between personnel and the preset tilt line boundary within the work area specifically includes: The actual spatial location of personnel within the loading station's work area is determined by the personnel bounding boxes identified in each frame of the real-time monitoring video stream. The spatial distance between the actual position of the personnel in the loading station operation area and the preset tilt line boundary is used as the static distance feature between the personnel in the operation area and the preset tilt line boundary.

4. The method as described in claim 1, characterized in that, The determination of the initial risk value of personnel crossing the line at different times during the arrival of a train at the loading station by combining the static distance features with the dynamic behavioral feature sequence specifically includes: Based on the static distance features combined with the dynamic behavioral feature sequence, the intention of personnel to cross the line at different times during the arrival of a vehicle at the loading station is determined. The initial risk value of personnel crossing the line at different times is determined by observing their intentional actions of crossing the line during the process of a vehicle arriving at the loading station.

5. The method as described in claim 1, characterized in that, The cross-dimensional correlation deviation between all initial line-crossing risk values ​​and oncoming vehicle trajectory data is determined by analyzing their spatiotemporal correlation within the stated risk period. Specifically, this includes: All preliminary risk values ​​of crossing the line and the trajectory data of oncoming vehicles are spatiotemporally aligned within the specified risk period to obtain spatiotemporally aligned data of personnel and oncoming vehicles within the specified risk period. Based on the spatiotemporal alignment data, the temporal correlation strength between personnel and vehicles during the arrival process at the loading station is determined in the time dimension and the spatial correlation strength in the spatial dimension. The cross-dimensional correlation deviation between all preliminary line-crossing risk values ​​and oncoming vehicle trajectory data is determined by the temporal correlation strength and the spatial correlation strength.

6. The method as described in claim 1, characterized in that, The adaptive weighted adjustment of all preliminary lane-crossing risk values ​​within the driving risk period based on the aforementioned correlation deviation specifically includes: The adjustment coefficients in the adaptive weighted adjustment process are determined based on the aforementioned correlation deviation. The time decay factor is determined based on the starting point of the aforementioned driving risk period; Based on the adjustment coefficient and the time decay factor, all preliminary lane crossing risk values ​​within the driving risk period are adaptively weighted and adjusted to obtain the adjusted lane crossing risk values.

7. The method as described in claim 1, characterized in that, The comprehensive tilt risk characteristics of personnel, determined based on all adjusted overshoot risk values, specifically include: Statistical features were extracted from all adjusted risk values ​​of crossing the line to obtain the statistical features of the risk of personnel leaning over the line when a train arrives at the loading station; The overall tilt profile of an individual is determined by the aforementioned risk statistical characteristics.

8. A passenger tilt warning system for when a train approaches at a loading station, characterized in that, include: The data acquisition module is used to simultaneously collect real-time driving trajectory data of vehicles arriving at the loading station and real-time monitoring video streams of personnel in the work area; The processing module is used to extract the dynamic behavior feature sequence of personnel from the real-time monitoring video stream, and then determine the static distance feature between personnel and the preset tilt line boundary in the work area. The static distance feature is combined with the dynamic behavior feature sequence to determine the preliminary risk value of personnel crossing the line at different times during the arrival of the loading station. The processing module is also used to remove anomalies from the real-time driving trajectory data of the approaching vehicle, and then extract the driving risk period when the approaching vehicle reaches the preset inclined line boundary from the removed driving trajectory data. The cross-dimensional correlation deviation between all preliminary cross-line risk values ​​and the approaching vehicle driving trajectory data is determined by the spatiotemporal correlation between all preliminary cross-line risk values ​​and the approaching vehicle driving trajectory data during the driving risk period. The processing module is also used to adaptively weight and adjust all preliminary lane-crossing risk values ​​within the driving risk period based on the correlation deviation, and then determine the comprehensive lane-crossing risk characteristics of personnel based on all adjusted lane-crossing risk values. The execution module is used to send a personnel tilt warning signal to the loading station monitoring center based on the comprehensive tilt risk characteristics.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store computer programs, and the processor is used to call and run the computer programs from the memory, so that the computer device performs the personnel tilting warning method for the arrival of a vehicle at the loading station as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions or code that, when executed on a computer, cause the computer to implement the personnel tilting warning method for the arrival of a train at the loading station as described in any one of claims 1 to 7.

Images