Railway construction protection method and system

Abstract

The invention discloses a railway construction protection method and system, and the method comprises the steps: firstly, setting hyper-parameters based on a construction operation position; secondly, acquiring video stream data of a construction scene in real time based on an unmanned aerial vehicle; secondly, segmenting a railway in the obtained video stream data by adopting an instance segmentation algorithm, and obtaining pixel point coordinates of an uplink track and pixel point coordinates of a downlink track; then, based on the pixel point coordinates of the uplink track and the pixel point coordinates of the downlink track, a fitting straight line of the uplink track and the downlink track is calculated; then, based on the fitting straight line of the uplink track and the downlink track, an electronic fence is automatically generated; and finally, based on the coordinate position of the constructor and the electronic fence, whether the constructor breaks into the dangerous area is judged. According to the method, the instance segmentation algorithm is adopted to segment the railway, the segmentation effect is good, manual drawing of an electronic fence is not needed, and only the unmanned aerial vehicle needs to be carried to a designated construction site without external cooperation.

Description

Technical Field

[0001] The present invention belongs to the technical field of rail transportation, and in particular relates to a railway construction protection method and system. Background Art

[0002] With the increasing density of railway transportation, railway construction, electrical and engineering personnel need to construct, inspect and maintain equipment related to railway lines. In the case of heavy construction and transportation tasks, trains often run on the line during construction and inspection and maintenance. Railways are generally divided into up and down tracks. When workers are working on the up track, the down track is normally open. Many workers lack safety awareness. When they take a break, they will run to another track for rest, which will cause safety hazards. In addition, the railway construction process has the characteristics of many construction points, long lines, and poor working environment. Personnel face various risks, especially most railway construction sites are constructed in the field, which makes personnel management difficult and personnel safety cannot be guaranteed. Therefore, when the staff are working on site, they need to enclose the on-site work site and dangerous areas. At present, the safety protection methods for construction site personnel are still traditional methods such as setting up warning lines and relying on personnel work experience. Therefore, the safety management of railway construction personnel is relatively rough, lacking effective safety management methods, and cannot effectively guarantee the personal safety of personnel.

[0003] Under the complex and harsh conditions of railway construction, personnel safety is greatly threatened. Figure 1 As shown, an existing management method is introduced, which is divided into two steps. One is to obtain the GIS database along the line, manually mark the dangerous area in the GIS map library, and delineate the electronic fence information. The other is to install the foundation enhancement system along the line, which requires construction workers to wear positioning devices and install positioning devices on vehicles. Then the location information of the vehicle and construction workers can be displayed on the map, and finally the operation is determined whether it is in violation of the rules based on the area of ​​the electronic fence. However, the above method is cumbersome. When constructing in different places, it is necessary to update the GIS map information and manually delineate the electronic fence. Furthermore, it is necessary to install positioning devices on construction workers and engineering machinery vehicles. Construction workers are prone to lose positioning devices or do not wear positioning devices. There are many uncontrollable factors, and when construction workers enter dangerous areas, they cannot be prompted with danger signals.

[0004] How to avoid safety accidents among construction workers and how to better ensure the personal safety of workers has become a practical problem that needs to be solved urgently. Summary of the invention

[0005] In view of the above problems, the present invention provides a railway construction protection method and system, which have strong timeliness, safety and reliability.

[0006] The object of the present invention is to provide a railway construction protection method, comprising: Set hyperparameters based on the location of the construction work; Obtain video stream data of construction scenes in real time based on drones; The railway in the acquired video stream data is segmented by using an instance segmentation algorithm to obtain the pixel coordinates of the uplink track and the pixel coordinates of the downlink track; Calculating the fitting straight lines of the uplink track and the downlink track based on the pixel coordinates of the uplink track and the pixel coordinates of the downlink track respectively; Automatically generate electronic fence based on the fitting straight line of the up track and down track; Based on the coordinate position of the construction workers and the electronic fence, it is determined whether the construction workers have entered the dangerous area.

[0007] Further, based on the construction work location, setting the hyper parameters includes determining the dangerous area and the safe area of ​​the railway based on the construction work location, including, Determine whether the construction work location is in the up track area or the down track area, where: If the construction work is located in the up track area, the down track area is the dangerous area and the up track area is the safe area; If the construction work location is in the down track area, the up track area is the dangerous area and the down track area is the safe area.

[0008] Furthermore, the railway in the acquired video stream data is segmented using an instance segmentation algorithm, and the pixel coordinates of the uplink track and the pixel coordinates of the downlink track are obtained, including: The acquired video stream data is input into the instance segmentation algorithm yolov8-seg model, the uplink track and the downlink track are classified, and the uplink track mask and the downlink track mask are output; wherein the uplink track mask includes the coordinate positions of all pixel points of the uplink track, and the downlink track mask includes the coordinate positions of all pixel points of the downlink track.

[0009] Further, based on the pixel coordinates of the uplink track and the pixel coordinates of the downlink track, respectively, the fitting straight lines of the uplink track and the downlink track are calculated, including: The least square method is used to perform straight line fitting on the pixel coordinates of each corresponding track to obtain the fitting straight lines of the uplink track and the downlink track.

[0010] Furthermore, the least square method is used to perform straight line fitting on the pixel coordinates of each corresponding track, including: Prepare data sets of pixel point coordinates of the uplink track and the downlink track respectively; Establish a fitted straight line model; Based on the fitted straight line model, the objective function of the fitted straight line is established; Based on the established fitting straight line objective function, the target equation group is obtained by differentiating the slope and intercept in the fitting straight line model; Substitute the data in the data set of pixel point coordinates of the uplink track and the downlink track into the target equation group to obtain the slope and intercept of the fitting line of the uplink track and the downlink track; The slope and intercept of the upward track fitting line and the downward track fitting line are respectively substituted into the fitting line model to obtain the upward track fitting line and the downward track fitting line.

[0011] Furthermore, based on the fitting straight line of the up track and the down track, the electronic fence is automatically generated including: Determine the slope and intercept of the electronic fence, wherein the slope of the electronic fence is obtained according to a preset algorithm based on the slope of the fitting straight line of the upward track and the slope of the fitting straight line of the downward track, and the intercept of the electronic fence is obtained according to a preset algorithm based on the intercept of the fitting straight line of the upward track and the intercept of the fitting straight line of the downward track; Based on the determined slope and intercept of the electronic fence, a first straight line is formed, and the first straight line is the electronic fence.

[0012] Furthermore, the method further includes obtaining the coordinate position of the construction workers in the video stream data and each construction worker, including: The trained neural network-based target detection model is used to identify the construction workers in the video image, and the coordinate positions of the construction workers in the video image are obtained.

[0013] Furthermore, the training process of the neural network-based target detection model includes: Collect construction scene video stream data at the construction site; Label the construction workers in all the collected video stream data and create a data set; The selected neural network model is iteratively trained using the data set, the selected neural network model is optimized, and finally the optimized neural network model is used as the neural network-based target detection model.

[0014] Furthermore, it also includes determining the track area where the construction personnel are located, including, The Y axis in the video image in the video stream data to be detected is preset to have the downward direction in the image as the positive direction; When the Y coordinate in any construction worker's coordinate position is smaller than the Y coordinate on the electronic fence corresponding to the corresponding horizontal coordinate, it means that the construction worker is in the up track area; When the Y coordinate in any construction worker's coordinate position is greater than the Y coordinate on the electronic fence corresponding to the corresponding horizontal coordinate, it means that the construction worker is in the down track area.

[0015] Further, judging whether the construction personnel have entered the dangerous area includes: If the up-track area is a dangerous area, when the coordinate position of any construction worker is in the up-track area, it means that the construction worker has entered the dangerous area; If the down track area is a dangerous area, when the coordinate position of any construction worker is in the down track area, it means that the construction worker has entered the dangerous area; When construction workers enter a dangerous area, drones are used to issue warnings.

[0016] Another object of the present invention is to provide a railway construction protection system, including a drone, wherein the drone includes: A setting module, used to set hyper parameters based on the construction work location; The acquisition module is used to obtain video stream data of the construction scene in real time based on the drone; A processing module, used for segmenting the railway in the acquired video stream data by using an instance segmentation algorithm, and obtaining pixel coordinates of an up track and pixel coordinates of a down track; A calculation module, used to calculate the fitting straight lines of the upward track and the downward track based on the pixel coordinates of the upward track and the pixel coordinates of the downward track respectively; A generation module, used to automatically generate an electronic fence based on the fitted straight line of the up track and the down track; The monitoring module is used to determine whether construction workers have entered a dangerous area based on the construction workers' coordinate positions and electronic fences.

[0017] Furthermore, the processing module is also used to obtain the construction workers in the video stream data and the coordinate position of each construction worker, including: The trained neural network-based target detection model is used to identify the construction workers in the video image, and the coordinate positions of the construction workers in the video image are obtained.

[0018] The protection method of the present invention uses a drone to monitor the track construction scene from a high altitude, and segments the railway through an instance segmentation algorithm, which has a good segmentation effect and realizes accurate identification of the rails. Further, an electronic fence is generated in real time, so that the above method can be applied to any construction location beside the railway, without manually drawing an electronic fence. When the construction site is changed, it only needs to carry the drone to the designated construction site, and no external cooperation is required.

[0019] In addition, the safety and reliability of railway construction are guaranteed by simultaneously obtaining the real-time location of construction workers and monitoring whether they enter dangerous areas.

[0020] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 A flow chart of a railway construction safety protection method in the prior art is shown; Figure 2 A schematic diagram of a process flow of a railway construction protection method in an embodiment of the present invention is shown; Figure 3 A schematic flow chart of another railway construction protection method in an embodiment of the present invention is shown; Figure 4 An image of a railway construction scene acquired in real time based on a drone in an embodiment of the present invention is shown; Figure 5 A schematic diagram of a Yolov8-seg algorithm network structure in an embodiment of the present invention is shown; Figure 6 A railway schematic diagram segmented by an example segmentation algorithm in an embodiment of the present invention is shown; Figure 7 A schematic diagram of fitting straight lines of an upward track and a downward track in an embodiment of the present invention is shown; Figure 8 A schematic diagram of an electronic fence in an embodiment of the present invention is shown; Fig. 9 A flowchart of a training process of a target detection model based on a neural network in an embodiment of the present invention is shown; Fig.10 A structural schematic diagram of a railway construction protection system in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] like Figure 2 As shown, a railway construction protection method is introduced in an embodiment of the present invention, and the method includes: first, setting hyperparameters based on the construction operation position; second, obtaining video stream data of the construction scene in real time based on the drone; then, using the instance segmentation algorithm to segment the railway in the obtained video stream data, and obtaining the pixel coordinates of the up track and the pixel coordinates of the down track; then, respectively calculating the fitting straight lines of the up track and the down track based on the pixel coordinates of the up track and the pixel coordinates of the down track; then, automatically generating an electronic fence based on the fitting straight lines of the up track and the down track; finally, judging whether the construction personnel have entered the dangerous area based on the coordinate position of the construction personnel and the electronic fence. The construction scene is monitored by the drone at high altitude, and the electronic fence is generated in real time. The construction scene of the track is monitored by the drone at high altitude, and the railway is segmented by the instance segmentation algorithm, and the segmentation effect is good, and the accurate identification of the rail is realized. Further, the electronic fence is generated in real time, so that the above method can be applied to any construction location beside the railway, and there is no need to manually draw the electronic fence. When the construction site is changed, it only needs to carry the drone to the designated construction site, and no external cooperation is required. In addition, the safety and reliability of railway construction are guaranteed by simultaneously obtaining the real-time location of construction workers and monitoring whether they enter dangerous areas.

[0025] Specifically, based on the construction work location, setting the hyperparameters includes determining the dangerous area and safe area of ​​the railway based on the construction work location, including determining whether the construction work location is in the up track area or the down track area, wherein if the construction work location is in the up track area, the down track area is the dangerous area, and the up track area is the safe area; if the construction work location is in the down track area, the up track area is the dangerous area, and the down track area is the safe area. Wherein, the construction work location is known location data, and the hyperparameters are determined according to the on-site construction personnel. If the construction personnel are constructing in the up track area, then the down track is a dangerous area. Similarly, if the construction personnel are constructing in the down track, then the up track is set as a dangerous area. For example, Figure 4 In the construction scenario shown, the construction work location is in the down track area, so it is necessary to set the down track area as a safe area and the up track area as a dangerous area in advance.

[0026] like Figure 3 As shown, the railway construction protection method includes two parts, one part is to generate an electronic fence in real time, and the other part is to monitor whether the construction workers enter the dangerous area, which specifically includes the following steps: Step S1: The drone acquires video stream data in real time, wherein the video stream data includes video images and pictures, wherein the video images are superimposed on the pictures, such as Figure 4 Shown is a video image in the video stream data.

[0027] Step S11: Use an instance segmentation algorithm to segment the railway in the acquired video stream data. Specifically, use the following Figure 5 The Yolov8-seg algorithm structure shown in the figure is an instance segmentation algorithm model based on a convolutional neural network. The network structure is mainly divided into three parts, namely the backbone network, the neck layer network, and the head layer network. The backbone network extracts features from the input image. The backbone network is composed of Conv (convolution layer), C2f (CSPBottleneck with 2 convolutions: CSP bottleneck structure with two convolutions), and SPPF (SpatialPyramid Pooling – Fast: Pyramid Pooling Technology) modules, where Conv is convolution + normalization + activation function. The C2f module mainly divides the feature layer into channels, performs Conv operations respectively, and then reorganizes them using the Connect operation to obtain a new feature layer. The SPPF module mainly performs multi-layer pooling operations on the feature layer, and performs a connect (establishing a connection or association between two entities) operation on the multi-pooled feature layer to reorganize and obtain a new feature layer. The neck layer network is composed of C2f, upsample, and connect modules, where upsample is an upsampling operation, and connect is a merge connection of feature layers of the same size. The head layer network is composed of Conv, C2f, and connect modules. The model is trained using the instance segmentation algorithm yolov8-seg and the uplink and downlink railways in the image are segmented. The input video image is a construction scene shot by a drone. Through instance segmentation, we obtain Figure 6The output image effect diagram in , where the output data of segment (segmentation module) is to segment the up and down tracks and classify the tracks. The railway includes the up track and the down track, that is, directly obtain the up track mask and the down track mask, where the up track mask includes the coordinate positions of all pixel points on the up track (also called the pixel values ​​corresponding to the pixel points of the down track), and the down track mask includes the coordinate positions of all pixel points on the down track (also called the pixel values ​​corresponding to the pixel points of the down track). Instance segmentation uses the results of target detection and semantic segmentation at the same time, and extracts the Mask corresponding to the target in the semantic segmentation through the index of the target's highest confidence category provided by target detection. In short, it is to segment specific objects (specific instances) in a category, making the acquisition of the track more accurate and reliable.

[0028] Step S12: Obtain pixel coordinate positions of the uplink track mask and the downlink track mask.

[0029] Step S13: Based on the obtained pixel coordinate positions of the uplink track mask and the downlink track mask, a least square method is used to perform straight line fitting on the corresponding pixel coordinates on each track to obtain the fitting straight lines of the uplink track and the downlink track. Specifically, The pixel point coordinate data sets of the uplink track and the downlink track are respectively prepared. For example, the pixel point coordinate data sets are shown in Table 1. Table 1 Dataset of pixel coordinates

[0030] Establish a fitted straight line model. Specifically, assume that the fitted straight line models of the upward and downward tracks are . Still using the data in Table 1 as an example, there are n data in the pixel point coordinate data set of any track. , where n represents one or more data, but is not limited to this. In certain scenarios, n can also be 0, and i satisfies , then for each , the straight line model of the track is established as:

[0031] Based on the fitted straight line model, the objective function of the fitted straight line is established, where the objective function satisfies:

[0032] Based on the established fitting straight line objective function, the target equation group is obtained by differentiating the slope and intercept in the fitting straight line model, which includes: first, based on formula (2) , respectively, for the slope and the intercept Derivation obtains formulas (3) and (4):

[0033] Then, let the partial derivatives in formulas (3) and (4) be 0, and we get the following system of linear equations with two variables:

[0034] The set of two-variable linear equations shown in formulas (5) and (6) is the target set of equations. By solving the corresponding data in the pixel point coordinate data set of the uplink track and the pixel point coordinate data of the downlink track respectively through formulas (5) and (6), the slope k1 and intercept b1 of the fitting line of the uplink track and the slope k2 and intercept b2 of the fitting line of the downlink track can be obtained respectively, thereby fitting two straight lines as follows: Figure 7 As shown, one of them represents the running direction of the up track and the other represents the running direction of the down track. Further, the formulas for the fitting straight lines of the up track and the down track are obtained as formula (7) and formula (8):

[0035] Step S14, automatically generating an electronic fence based on the fitting straight line of the upward track and the downward track includes: Determine the slope and intercept of the electronic fence, wherein the slope of the electronic fence is obtained according to a preset algorithm based on the slope of the fitting line of the upward track and the slope of the fitting line of the downward track, and the intercept of the electronic fence is obtained according to a preset algorithm based on the intercept of the fitting line of the upward track and the intercept of the fitting line of the downward track. Specifically, the preset algorithm may be that the slope of the electronic fence is the average of the slope of the fitting line of the upward track and the slope of the fitting line of the downward track, and the intercept of the electronic fence is the average of the intercept of the fitting line of the upward track and the intercept of the fitting line of the downward track. Still using formulas (7) and (8) for exemplary explanation, the slope of the electronic fence is and the intercept satisfy:

[0036] But not limited to this, preferably, the preset algorithm can also be set according to the actual scene a value between the uplink and downlink track intercepts, preferably, the preset algorithm includes setting a ratio , the slope of the electronic fence and the intercept satisfy:

[0037] Therefore, the average value in the above formulas (9) and (10) can be considered as the ratio The value of is 0.5.

[0038] Based on the determined slope and intercept of the electronic fence, a first straight line is formed, and the first straight line is the electronic fence, wherein the formula of the electronic fence satisfies:

[0039] Specifically, the electronic fence obtained is as follows: Figure 8 As shown, , They are the pixel coordinates on the uplink and downlink tracks respectively.

[0040] like Figure 3 In the process, while obtaining the electronic fence, the construction workers in the video stream data and the coordinate position of each construction worker are also obtained, which specifically includes the following steps: Step S21, calling the trained neural network-based target detection model; wherein the coordinate position of the construction personnel is obtained through a target detection algorithm such as a yolo series neural network model. Specifically, it is necessary to obtain a detection model through offline training of a target detection algorithm such as a yolo (you only look once, a convolution-based neural network target detection network structure) series neural network model, so as to Fig. 9 As shown, the training process of the neural network-based target detection model includes: Collect construction scene video stream data at the construction site; Label the construction workers in all the collected video stream data and create a data set; The selected neural network model is iteratively trained using the data set, the selected neural network model is optimized, and finally the optimized neural network model is used as the target detection model based on the neural network. Among them, the selected neural network module is optimized to obtain a neural network model that meets the requirements of detection accuracy, detection speed, etc.

[0041] Step S22, obtaining the construction workers in the video image and the coordinate position of each construction worker, includes using a trained neural network-based target detection model to identify the construction workers in the video image and obtaining the coordinate position of the construction workers in the video image.

[0042] In the embodiment of the present invention, the coordinate positions of n construction workers in the image are obtained according to the target detection model based on the neural network. , and z=0, 1, ..., n-1. Among them, and It can be any position in the worker detection frame. Preferably, and is the center point of the detection box.

[0043] Further, the method also includes determining the track area where the construction personnel are located, including, for example, Figure 8 As shown. First, the Y-axis in the video image in the preset detected video stream data takes the downward direction in the image as the positive direction; then, when the Y coordinate in any construction worker's coordinate position is less than the Y coordinate on the electronic fence corresponding to the corresponding horizontal coordinate, it indicates that the construction worker is in the upward track area; when the Y coordinate in any construction worker's coordinate position is greater than the Y coordinate on the electronic fence corresponding to the corresponding horizontal coordinate, it indicates that the construction worker is in the downward track area. That is, with the electronic fence as the dividing line, the area above the electronic fence is the upward track area, and the area below is the downward track area, so that the coordinate positions of the construction workers in the upward and downward track areas respectively satisfy:

[0044] Formula (14) indicates that the coordinate position of the construction personnel is in the up-track area, and formula (15) indicates that the coordinate position of the construction personnel is in the down-track area.

[0045] Step S23, determine whether n is greater than z. If it is greater than z, there are still construction workers, and it is necessary to determine whether the construction workers have entered the dangerous area, and execute step S24; if it is less than z, end the determination.

[0046] Step S24: Based on the obtained electronic fence, determine y z > x z + Is it true, and at the same time let z accumulate 1, and return to step S23, wherein if it is true, execute step S2411, if not, execute step S2421; S2411, prove that the coordinates of the construction personnel are in the down track area; S2412: If the downlink track area is set as a dangerous area, the drone will shout a warning; S2421, prove that the coordinates of the construction personnel are in the up track area; S2422: If the uplink track area is set as a dangerous area, the drone will issue a warning.

[0047] The above method is applied to the track construction site, and can automatically generate an electronic fence without manual animation, and is applicable to all railway operation scenarios. Especially for the construction of one side of the track, it is easy to deploy, does not require external conditions, and does not require construction personnel to install electronic equipment. It only needs to move the drone, and finally the drone can sound an alarm to remind construction personnel in the dangerous area in time. In addition, by simultaneously identifying and obtaining the construction personnel and the location of the construction personnel, and real-time monitoring of whether the construction personnel enter the dangerous area, the above method improves the reliability of railway safety protection and the safety of construction.

[0048] like Fig.10 As shown, the embodiment of the present invention also introduces a railway construction protection system capable of executing the above method, the system includes a drone, the drone includes a setting module, an acquisition module, a processing module, a calculation module, a generation module and a monitoring module, wherein the setting module is used to set hyperparameters based on the construction operation position; the acquisition module is used to obtain video stream data of the construction scene in real time based on the drone; the processing module is used to segment the railway in the acquired video stream data using an instance segmentation algorithm to obtain the pixel coordinates of the up track and the pixel coordinates of the down track; the calculation module is used to calculate the fitting straight line of the up track and the down track based on the pixel coordinates of the up track and the pixel coordinates of the down track respectively; the generation module is used to automatically generate an electronic fence based on the fitting straight line of the up track and the down track; the monitoring module is used to determine whether the construction personnel have entered the dangerous area based on the coordinate position of the construction personnel and the electronic fence. The above system uses a drone to monitor the construction scene from a high altitude and generates an electronic fence in real time. The above system can be applied to any construction location beside the railway, without manually drawing the electronic fence, and changing the construction site. It only needs to carry the drone to the designated construction site, and no external cooperation is required, which greatly improves the safety and reliability of railway safety protection.

[0049] In an embodiment of the present invention, the processing module is also used to obtain the construction workers in the video stream data and the coordinate position of each construction worker, including using a trained neural network-based target detection model to identify the construction workers in the video image, and obtaining the coordinate position of the construction workers in the video image.

[0050] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A railway construction protection method, characterized in that: include, Set hyperparameters based on the location of the construction work; Obtain video stream data of construction scenes in real time based on drones; The railway in the acquired video stream data is segmented by using an instance segmentation algorithm to obtain the pixel coordinates of the uplink track and the pixel coordinates of the downlink track; Calculating the fitting straight lines of the uplink track and the downlink track based on the pixel coordinates of the uplink track and the pixel coordinates of the downlink track respectively; Automatically generate electronic fence based on the fitting straight line of the up track and down track; Based on the coordinate position of the construction workers and the electronic fence, it is determined whether the construction workers have entered the dangerous area.

2. The railway construction protection method according to claim 1, characterized in that: Based on the location of the construction work, setting hyper parameters includes determining the dangerous area and safe area of ​​the railway based on the location of the construction work, including, Determine whether the construction work location is in the up track area or the down track area, where: If the construction work is located in the up track area, the down track area is the dangerous area and the up track area is the safe area; If the construction work location is in the down track area, the up track area is the dangerous area and the down track area is the safe area.

3. The railway construction protection method according to claim 2, characterized in that: The instance segmentation algorithm is used to segment the railway in the acquired video stream data, and the pixel coordinates of the uplink track and the pixel coordinates of the downlink track are obtained, including: The acquired video stream data is input into the instance segmentation algorithm yolov8-seg model, the uplink track and the downlink track are classified, and the uplink track mask and the downlink track mask are output; wherein the uplink track mask includes the coordinate positions of all pixel points of the uplink track, and the downlink track mask includes the coordinate positions of all pixel points of the downlink track.

4. The railway construction protection method according to claim 3, characterized in that: Based on the pixel coordinates of the uplink track and the pixel coordinates of the downlink track, the fitting straight lines of the uplink track and the downlink track are calculated, including: The least square method is used to perform straight line fitting on the pixel coordinates of each corresponding track to obtain the fitting straight lines of the uplink track and the downlink track.

5. The railway construction protection method according to claim 4, characterized in that: The least square method is used to fit the pixel coordinates of each corresponding track into a straight line, including: Prepare data sets of pixel point coordinates of the uplink track and the downlink track respectively; Establish a fitted straight line model; Based on the fitted straight line model, the objective function of the fitted straight line is established; Based on the established fitting straight line objective function, the target equation group is obtained by differentiating the slope and intercept in the fitting straight line model; Substitute the data in the data set of pixel point coordinates of the uplink track and the downlink track into the target equation group to obtain the slope and intercept of the fitting line of the uplink track and the downlink track; The slope and intercept of the upward track fitting line and the downward track fitting line are respectively substituted into the fitting line model to obtain the upward track fitting line and the downward track fitting line.

6. The railway construction protection method according to claim 5, characterized in that: Based on the fitting straight line of the up and down tracks, the electronic fence is automatically generated including: Determine the slope and intercept of the electronic fence, wherein the slope of the electronic fence is obtained according to a preset algorithm based on the slope of the fitting straight line of the upward track and the slope of the fitting straight line of the downward track, and the intercept of the electronic fence is obtained according to a preset algorithm based on the intercept of the fitting straight line of the upward track and the intercept of the fitting straight line of the downward track; Based on the determined slope and intercept of the electronic fence, a first straight line is formed, and the first straight line is the electronic fence.

7. The railway construction protection method according to claim 1, characterized in that: It also includes obtaining the construction workers in the video stream data and the coordinate position of each construction worker, including: The trained neural network-based target detection model is used to identify the construction workers in the video image, and the coordinate positions of the construction workers in the video image are obtained.

8. The railway construction protection method according to claim 7, characterized in that: The training process of the neural network-based target detection model includes: Collect construction scene video stream data at the construction site; Label the construction workers in all the collected video stream data and create a data set; The selected neural network model is iteratively trained using the data set, the selected neural network model is optimized, and finally the optimized neural network model is used as the neural network-based target detection model.

9. The railway construction protection method according to any one of claims 1 to 8, characterized in that: It also includes determining the track area where the construction personnel are located, including, The Y axis in the video image in the video stream data to be detected is preset to have the downward direction in the image as the positive direction; When the Y coordinate in any construction worker's coordinate position is smaller than the Y coordinate on the electronic fence corresponding to the corresponding horizontal coordinate, it means that the construction worker is in the up track area; When the Y coordinate in any construction worker's coordinate position is greater than the Y coordinate on the electronic fence corresponding to the corresponding horizontal coordinate, it means that the construction worker is in the down track area.

10. The railway construction protection method according to claim 9, characterized in that: Determine whether construction workers have entered the dangerous area, including: If the up-track area is a dangerous area, when the coordinate position of any construction worker is in the up-track area, it means that the construction worker has entered the dangerous area; If the down track area is a dangerous area, when the coordinate position of any construction worker is in the down track area, it means that the construction worker has entered the dangerous area; When construction workers enter a dangerous area, drones are used to issue warnings.

11. A railway construction protection system, characterized in that: Comprising a drone, the drone comprising, A setting module, used to set hyper parameters based on the construction work location; The acquisition module is used to obtain video stream data of the construction scene in real time based on the drone; A processing module, used for segmenting the railway in the acquired video stream data by using an instance segmentation algorithm, and obtaining pixel coordinates of an up track and pixel coordinates of a down track; A calculation module, used to calculate the fitting straight lines of the upward track and the downward track based on the pixel coordinates of the upward track and the pixel coordinates of the downward track respectively; A generation module, used to automatically generate an electronic fence based on the fitted straight line of the up track and the down track; The monitoring module is used to determine whether construction workers have entered a dangerous area based on the construction workers' coordinate positions and electronic fences.

12. The railway construction protection system according to claim 11, characterized in that: The processing module is also used to obtain the construction personnel in the video stream data and the coordinate position of each construction personnel, including: The trained neural network-based target detection model is used to identify the construction workers in the video image, and the coordinate positions of the construction workers in the video image are obtained.

Images