Station area safety risk card control system

Abstract

The invention discloses a station area safety risk card control system, and belongs to the technical field of railway station equipment, and the system comprises a video monitoring module which is used for continuously obtaining a plurality of station area images with different visual angles and shooting time information; the image processing and transmitting module is used for processing the station area image to obtain a to-be-recognized image; the electronic work card module is used for being worn on a person in a station area and generating positioning data of an electronic work card; the Beidou positioning module is used for constructing a station area virtual fence and positioning the electronic work cards in real time; and the security control alarm platform is used for carrying out image recognition based on an image recognition network according to the to-be-recognized image, carrying out positioning detection according to the positioning data of the electronic work card, and carrying out foreign matter invasion early warning, work card leaving early warning and personnel abnormal invasion early warning according to the fence abnormal invasion data and the image recognition result. The problem that station safety risk card control is not accurate and comprehensive enough is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of railway station equipment, and in particular relates to a station area safety risk card control system. Background Art

[0002] Traditionally, railway station areas use video surveillance systems to monitor security risks, but with the rapid development of railway transportation systems, higher requirements have been placed on railway station area security control, active defense, and hidden danger inspection. Traditional video surveillance systems require dedicated personnel to monitor the images sent back by video surveillance, which wastes manpower and material resources. In addition, if sudden illegal operations are not handled in a timely manner due to human negligence, real-time alarms cannot be achieved, and it is even more difficult to achieve pre-warning and in-process alarms. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a station area safety risk card control system, which solves the problem that station area safety risk card control is not accurate and comprehensive enough.

[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0005] The present invention provides a station area safety risk card control system, comprising:

[0006] The video monitoring module is used to shoot the station area at all times and continuously obtain station area images with shooting time information from several different perspectives;

[0007] Image processing and transmission module, used to process station area images and transmit the processed images to be identified to the security control platform;

[0008] The electronic badge module is used to be worn by personnel in the station area. It generates the positioning data of the electronic badge by receiving differential data and satellite signals;

[0009] Beidou positioning module is used to build a virtual fence in the station area, and to perform real-time positioning of the electronic work badge by playing differential data through the Beidou differential base station, and to transmit the positioning data of the virtual fence and electronic work badge in the station area to the security control platform;

[0010] The security control alarm platform is used to perform image recognition based on the image recognition network according to the image to be recognized, and to perform positioning detection according to the positioning data of the electronic work badge. In addition, based on the fence abnormal intrusion data and image recognition results, it can also issue foreign object intrusion warnings, work badge absence warnings, and abnormal personnel intrusion warnings.

[0011] The beneficial effects of the present invention are as follows: a station area safety risk card control system provided by the present invention uses a video monitoring module to shoot areas in the station area from different perspectives to obtain station area images with shooting time information, so as to achieve all-round and all-time monitoring coverage of the station area; the image processing and transmission module is used to enhance the contrast and retain details of the station area image, which can effectively improve the effect of image feature extraction; the electronic work badge module can receive the differential data and satellite signals of the Beidou positioning module to generate real-time positioning data of the electronic work badge, thereby reflecting the real-time position of the wearer, facilitating timely management warnings; The Beidou positioning module can provide differential data for high-precision positioning of electronic work badges. It can also set up virtual fences in the station area based on Beidou high-precision maps to prevent people from entering preset restricted areas, thereby ensuring the safety and control of the station area. The image recognition network is trained according to the images to be identified through the security control alarm platform, and positioning detection can be performed through the positioning data of the electronic work badge and the virtual fence of the station area. According to the abnormal intrusion data of the fence and the image recognition results, foreign object intrusion warning, work badge leaving the person warning and abnormal personnel intrusion warning can be realized, thereby realizing comprehensive, efficient, accurate and fast station area safety risk card control warning.

[0012] Furthermore, the image processing and transmission module includes:

[0013] The image preprocessing submodule is used to preprocess the station area image to obtain the image to be recognized;

[0014] The image transmission submodule is used to transmit the image to be identified to the security control platform through the wireless access point and the 5G network.

[0015] Furthermore, the image preprocessing module includes:

[0016] The first image processing unit is used to convert the original station area image into a grayscale image to obtain a station area grayscale image;

[0017] A second image processing unit, used for calculating the local contrast of the grayscale image of the station area through a Laplace operator;

[0018] The calculation expression of the local contrast of the grayscale image of the station area is as follows:

[0019]

[0020] Among them, C(x,y) represents the local contrast of the grayscale image of the station area, represents the result of convolution between the grayscale image of the station area and the Laplace filter, and |·| represents the absolute value operation;

[0021] A third image processing unit, used for calculating a nonlinear gain factor according to a local contrast of the grayscale image of the station area;

[0022] The calculation expression of the nonlinear gain factor is as follows:

[0023]

[0024] Among them, α represents the nonlinear gain factor, C represents the local contrast of the station area image, and Q represents the preset contrast threshold;

[0025] A fourth image processing unit, configured to multiply the grayscale value of each pixel in the original station area image by the nonlinear gain factor corresponding to the pixel position according to the nonlinear gain, to obtain a grayscale image of the station area after gain;

[0026] A fifth image processing unit, used for adjusting the grayscale image of the station area after gain to obtain a dark primary color station area image;

[0027] The calculation expression of the dark primary color station area image is as follows:

[0028]

[0029] I(x,y)=c(x,y)×l(x,y)+A×(1-I(x,y)),

[0030] Among them, I dark (x,y) represents the dark original color station area image, It means taking the minimum value based on the pixel color value of the original station area image, c dark (x, y) represents the pixel dark original color value of the dark original color station area image, I(x, y) represents the original station area image, c(x, y) represents the pixel color value, l(x, y) represents the light transmittance of the pixel, and A represents the atmospheric light value;

[0031] A sixth image processing unit, used for obtaining an image to be identified based on the dark primary color station area image, a Laplace filter and an atmospheric light value;

[0032] The calculation expression of the image to be recognized is as follows:

[0033]

[0034] Among them, I′(x, y) represents the image to be identified, λ′(x, y) represents the propagation map of the second station area, λ(x, y) represents the propagation map of the first station area, ε represents the propagation processing factor, [L] represents the Laplace filter, and χ represents the propagation auxiliary parameter.

[0035] The beneficial effect of adopting the above further scheme is: the present invention achieves the purpose of retaining image details while improving image contrast through the image preprocessing module, thereby obtaining a high-quality image to be identified, and effectively improving the efficiency of image feature extraction and image recognition of the image recognition network.

[0036] Furthermore, the image transmission submodule includes:

[0037] An image switch unit, for receiving an image to be identified and wirelessly transmitting the image to be identified to a wireless access point;

[0038] A plurality of wireless access point units, used for wirelessly receiving the image to be identified transmitted by the corresponding image switch, and transmitting the image to be identified to the main wireless access point through a 5G signal;

[0039] The main wireless access point unit is used to transmit the image to be identified to the security control platform via the network cable.

[0040] The beneficial effect of adopting the above further scheme is: the present invention can quickly and accurately transmit the images to be identified from different perspectives to the security control platform through the wireless communication connection between several wireless access point units and the main wireless access point unit, which can not only achieve low-latency image signal transmission and reduce system delay, but also reduce redundant wiring and avoid image transmission problems caused by line damage.

[0041] Furthermore, the Beidou positioning module includes:

[0042] The virtual fence submodule is used to set up virtual fences for preset restricted areas in the station area based on high-precision Beidou maps;

[0043] Several Beidou differential base stations are used to generate differential data, broadcast the differential data to the electronic work badges within the coverage area of ​​each Beidou differential base station, and transmit the positioning data of the electronic work badges in the area to the security control platform.

[0044] The beneficial effects of adopting the above-mentioned further scheme are as follows: the present invention realizes real-time and accurate positioning of electronic work badges through high-precision Beidou maps and Beidou differential base stations, and effectively sets up virtual fences within the station area, providing a basis for generating abnormal fence intrusion data when the electronic work badge enters the virtual fence, and providing protection for security card control of preset restricted areas in the station area.

[0045] Furthermore, the positioning data of the electronic badge includes real-time positioning time information, real-time positioning position information and basic information of the person corresponding to the electronic badge.

[0046] The beneficial effects of adopting the above further scheme are: the positioning data of the electronic work badge is set to include the real-time positioning time information, real-time positioning position information and basic information of the personnel corresponding to the electronic work badge. Based on the real-time positioning information of the electronic work badge, the activity trajectory of the personnel in the station can be located with high precision; through the real-time positioning time information and the basic information of the personnel corresponding to the electronic work badge, the shooting time information of the station area image can be matched, thereby achieving double confirmation of abnormal intrusion of personnel into the preset restricted area, and realizing accurate safety risk control of personnel in the station.

[0047] Furthermore, the security control alarm platform includes:

[0048] The positioning detection submodule is used to judge the abnormal fence intrusion according to the positioning data of the electronic work badge and the virtual fence of the station area. If the positioning data of the electronic work badge is within the range of the virtual fence of the station area, the abnormal fence intrusion data is generated;

[0049] The image recognition submodule is used to train the image recognition network based on the image to be recognized, and perform work badge recognition, foreign object intrusion recognition and personnel recognition on the enhanced image to be recognized corresponding to the newly acquired station area image, to obtain an image recognition result, wherein the image recognition result is a combination of the work badge recognition result, foreign object intrusion recognition result and personnel recognition result obtained after performing work badge recognition, foreign object intrusion recognition and personnel recognition on the same enhanced image to be recognized;

[0050] The foreign object intrusion warning submodule is used to identify the foreign object intrusion on the railway track according to the foreign object intrusion recognition result in the image recognition result. If there is foreign object intrusion on the railway track, an intrusion alarm is issued;

[0051] The badge-away warning submodule is used to determine if the badge-away person is present based on the positioning data of the electronic badge, the badge recognition result in the image recognition result, and the person recognition result. If the badge-away person is present, an alarm is triggered;

[0052] The abnormal intrusion warning submodule is used to combine the abnormal intrusion data of fences with the personnel recognition results to judge the abnormal intrusion of personnel. If there is an abnormal intrusion of personnel, an abnormal intrusion warning will be issued.

[0053] The beneficial effects of adopting the above further scheme are as follows: the security control alarm platform realizes positioning detection, image processing, badge recognition, foreign object intrusion recognition and personnel recognition based on the positioning data of the virtual fence and electronic badge in the station area and the image to be recognized, and combines the fence abnormal intrusion data, badge recognition results and personnel recognition results to realize foreign object intrusion warning, badge leaving person warning and abnormal personnel intrusion warning, thereby realizing accurate and comprehensive station safety risk control.

[0054] Furthermore, the image recognition submodule includes:

[0055] A data set unit, used to construct a data set of images to be identified based on the received images to be identified;

[0056] A data enhancement unit, used for randomly scaling, randomly cropping, randomly arranging and splicing the images to be identified in the image data set to be identified, to obtain enhanced images to be identified;

[0057] The adaptive anchor frame unit is used to annotate the enhanced image to be identified by using the roLabelImg image annotation tool to obtain an annotated image to be identified, wherein the annotated objects in the image to be identified include personnel, work badges, railway tracks, trains and foreign objects on the railway tracks in the station area;

[0058] The recognition network training unit is used to perform feature extraction, badge recognition, foreign object intrusion recognition and personnel recognition training on the annotated image to be recognized based on the image recognition network, until the image recognition prediction result and the annotated information of the image to be recognized reach a preset accuracy threshold or the training rounds reach a preset training number threshold, thereby obtaining a trained image recognition network;

[0059] The image recognition unit is used to use the trained image recognition network to perform work badge recognition, foreign object intrusion recognition and personnel recognition on the enhanced image to be recognized corresponding to the newly acquired station area image to obtain an image recognition result.

[0060] The beneficial effect of adopting the above further scheme is as follows: in the present invention, image enhancement and adaptive anchor frames are performed on the images to be identified from each perspective in the image data set to be identified, so as to facilitate multi-classification identification of personnel, work badges, railway tracks, trains and foreign objects on railway tracks in the station area, and the image recognition network is trained, which provides a basis for obtaining the image recognition results corresponding to the station area images taken at any time, thereby effectively ensuring the effect of foreign object intrusion warning, work badge leaving the person warning and abnormal personnel intrusion warning.

[0061] Furthermore, the image recognition network uses the backbone feature extraction network and feature fusion network of the YOLOv5s neural network model to extract the labeled features, and performs work badge recognition, foreign object intrusion recognition and personnel recognition through the prediction network of the YOLOv5s neural network model to obtain image recognition results.

[0062] Furthermore, the image recognition network is trained for work badge recognition, foreign object intrusion recognition and personnel recognition according to the image prediction loss function;

[0063] The calculation expressions of the image prediction loss function are as follows:

[0064] Loss = λ cls L cls +λ reg L reg ,

[0065]

[0066] Among them, Loss represents the image prediction loss function, λ cls Represents the weight coefficient of classification loss, L cls represents the classification loss function, λ reg represents the weight coefficient of bounding box regression loss, L reg represents the bounding box regression loss function, i represents the i-th annotated image to be recognized, N represents the total number of annotated images to be recognized, j represents the j-th recognition category, C represents the total number of recognition categories, and y ij represents the prediction result of the i-th labeled image to be identified of the j-th category, log represents the logarithmic function, represents the label information of the work badge on the labeled image to be identified, x represents the parameter difference between the predicted bounding box and the true bounding box, |·| represents the absolute value operation, otherwise represents other cases, and t i represents the predicted bounding box parameters of the i-th annotated image to be recognized, Represents the true bounding box parameters of the i-th annotated image to be recognized.

[0067] The beneficial effect of adopting the above further scheme is: the image prediction loss function provided by the present invention enables the image recognition network to achieve better performance in both classification and positioning, better enhances the robustness of the image recognition network, helps to optimize the training process, and converges to the optimal solution more quickly.

[0068] Other advantages of the present invention will be analyzed in more detail in subsequent embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0070] Figure 1 The present invention is a flowchart of the steps of a station area safety risk card control system in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0072] like Figure 1 As shown, in one embodiment of the present invention, the present invention provides a station area security risk card control system, comprising:

[0073] The video monitoring module is used to shoot the station area at all times and continuously obtain station area images with shooting time information from several different perspectives;

[0074] In this embodiment, the station area is photographed from different angles by the regional ball camera and the optical array camera to obtain a station area image with shooting time information; the optical array camera uses multiple small lenses instead of the traditional large lens to achieve a wider angle of view and a larger shooting range. Each small lens shoots the scene from a slightly different angle, and can capture multi-angle image data, thereby obtaining more complete and richer light field information and realizing digital refocusing. The regional ball camera is composed of a panoramic camera and a detail camera, and can choose to output color and black and white images. It has a strong panoramic fill light capability and can also achieve 7×24 hours full coverage shooting. Based on the optical array camera and the regional ball camera, a complete, accurate and rich station area image can be obtained. The station area image with shooting time information is convenient for combining the positioning data of the electronic work card with real-time positioning time information and real-time positioning position information to perform dual confirmation of image and positioning, ensuring the comprehensive and accurate control of the safety risk card in the station area, wherein the shooting time information carried by the station area monitoring image as the time attribute of the image always accompanies the processing and recognition process of the image, providing a basis for dual confirmation of the personnel monitoring positioning information in the station.

[0075] Image processing and transmission module, used to process station area images and transmit the processed images to be identified to the security control platform;

[0076] The image processing and transmission module comprises:

[0077] The image preprocessing submodule is used to preprocess the station area image to obtain the image to be recognized;

[0078] The image preprocessing module comprises:

[0079] The first image processing unit is used to convert the original station area image into a grayscale image to obtain a station area grayscale image; the conversion into a grayscale image facilitates single-channel image processing;

[0080] A second image processing unit, used for calculating the local contrast of the grayscale image of the station area through a Laplace operator;

[0081] The calculation expression of the local contrast of the grayscale image of the station area is as follows:

[0082]

[0083] Among them, C(x,y) represents the local contrast of the grayscale image of the station area, represents the result of convolution between the grayscale image of the station area and the Laplace filter, and |·| represents the absolute value operation;

[0084] A third image processing unit, used for calculating a nonlinear gain factor according to a local contrast of the grayscale image of the station area;

[0085] The calculation expression of the nonlinear gain factor is as follows:

[0086]

[0087] Among them, α represents the nonlinear gain factor, C represents the local contrast of the station area image, and Q represents the preset contrast threshold;

[0088] A fourth image processing unit, configured to multiply the grayscale value of each pixel in the original station area image by the nonlinear gain factor corresponding to the pixel position according to the nonlinear gain, to obtain a grayscale image of the station area after gain;

[0089] A fifth image processing unit, used for adjusting the grayscale image of the station area after gain to obtain a dark primary color station area image;

[0090] The calculation expression of the dark primary color station area image is as follows:

[0091]

[0092] I(x,y)=c(x,y)×l(x,y)+A×(1-I(x,y)),

[0093] Among them, I dark (x,y) represents the dark original color station area image, It means taking the minimum value based on the pixel color value of the original station area image, c dark(x, y) represents the pixel dark original color value of the dark original color station area image, I(x, y) represents the original station area image, c(x, y) represents the pixel color value, l(x, y) represents the light transmittance of the pixel, and A represents the atmospheric light value;

[0094] A sixth image processing unit, used for obtaining an image to be identified based on the dark primary color station area image, a Laplace filter and an atmospheric light value;

[0095] The calculation expression of the image to be recognized is as follows:

[0096]

[0097] Among them, I′(x, y) represents the image to be identified, λ′(x, y) represents the propagation map of the second station area, λ(x, y) represents the propagation map of the first station area, ε represents the propagation processing factor, [L] represents the Laplace filter, and χ represents the propagation auxiliary parameter.

[0098] The image transmission submodule is used to transmit the image to be identified to the security control platform through the wireless access point and the 5G network.

[0099] The image transmission submodule comprises:

[0100] An image switch unit, for receiving an image to be identified and wirelessly transmitting the image to be identified to a wireless access point;

[0101] In this embodiment, the wireless access point unit is installed at a high place in the station area without obvious obstruction, so as to effectively reduce the shielding and obstruction of the wireless signal.

[0102] A plurality of wireless access point units, used for wirelessly receiving the image to be identified transmitted by the corresponding image switch, and transmitting the image to be identified to the main wireless access point through a 5G signal;

[0103] The main wireless access point unit is used to transmit the image to be identified to the security control platform via the network cable.

[0104] In this embodiment, the main wireless access point is set at a high place on the roof in the station area, so as to timely receive the image to be identified transmitted by each wireless access node unit, so as to timely extract and identify the features of the image to be identified.

[0105] The electronic badge module is used to be worn by personnel in the station area. It generates the positioning data of the electronic badge by receiving differential data and satellite signals;

[0106] The positioning data of the electronic badge includes the real-time positioning time information, the real-time positioning position information and the basic information of the person corresponding to the electronic badge.

[0107] In this embodiment, through the real-time positioning time information of the electronic work badge and the time information of the workshop area image, it is possible to achieve dual confirmation of the time and space of the personnel wearing the electronic work badge based on the image and Beidou positioning, thereby ensuring the comprehensive and accurate control of safety risks in the station area; the basic information of the personnel corresponding to the electronic work badge includes work number, ID number, name, gender and age, which is convenient for quickly and accurately confirming the specific information of the corresponding monitored and located personnel, and implementing continuous tracking and monitoring.

[0108] Beidou positioning module is used to build a virtual fence in the station area, and to perform real-time positioning of the electronic work badge by playing differential data through the Beidou differential base station, and to transmit the positioning data of the virtual fence and electronic work badge in the station area to the security control platform;

[0109] The Beidou positioning module includes:

[0110] The virtual fence submodule is used to set up virtual fences for preset restricted areas in the station area based on high-precision Beidou maps;

[0111] In this embodiment, the preset prohibited areas in the station area are the safety yellow line areas, the driving dispatching area, the machinery room, the crew room, the emergency brake device area and other areas where there are safety hazards.

[0112] Several Beidou differential base stations are used to generate differential data, broadcast the differential data to the electronic badges in the coverage area of ​​each Beidou differential base station, and transmit the positioning data of the electronic badges in the area to the security control platform. In this embodiment, by combining the virtual fence and the positioning of the Beidou differential base station, it is possible to monitor the activities of people in the station area through real-time high-precision positioning and prevent people without the right of passage from entering the preset prohibited area.

[0113] The security control alarm platform is used to perform image recognition based on the image recognition network according to the image to be recognized, and to perform positioning detection according to the positioning data of the electronic work badge. In addition, based on the fence abnormal intrusion data and image recognition results, it can also issue foreign object intrusion warnings, work badge absence warnings, and abnormal personnel intrusion warnings.

[0114] The security control alarm platform includes:

[0115] The positioning detection submodule is used to judge the abnormal fence intrusion according to the positioning data of the electronic work badge and the virtual fence of the station area. If the positioning data of the electronic work badge is within the range of the virtual fence of the station area, the abnormal fence intrusion data is generated;

[0116] The image recognition submodule is used to train the image recognition network based on the image to be recognized, and perform work badge recognition, foreign object intrusion recognition and personnel recognition on the enhanced image to be recognized corresponding to the newly acquired station area image, to obtain an image recognition result, wherein the image recognition result is a combination of the work badge recognition result, foreign object intrusion recognition result and personnel recognition result obtained after performing work badge recognition, foreign object intrusion recognition and personnel recognition on the same enhanced image to be recognized;

[0117] The image recognition submodule comprises:

[0118] A data set unit, used to construct a data set of images to be identified based on the received images to be identified;

[0119] A data enhancement unit is used to randomly scale, randomly crop, randomly arrange and splice the images to be identified in the image data set to be identified, so as to obtain enhanced images to be identified;

[0120] In this embodiment, the enhanced images to be identified effectively increase the diversity of the data set, improve the robustness and generalization ability of the model, avoid independent data enhancement operations on each image block, thereby greatly reducing the computational complexity of the network model and improving the accuracy and robustness of target inspection.

[0121] The adaptive anchor frame unit is used to annotate the enhanced image to be identified by using the roLabelImg image annotation tool to obtain an annotated image to be identified, wherein the annotated objects in the image to be identified include personnel, work badges, railway tracks, trains and foreign objects on the railway tracks in the station area;

[0122] In this embodiment, the roLabelImg image annotation tool supports annotation of targets in the image such as rectangular boxes, polygons, points, lines, etc. Its biggest advantage is the newly added rotated rectangle annotation function. Compared with traditional rectangular box annotation, rotated rectangle annotation can more accurately describe the shape and position of the object, and can express the rotation angle of the rotated rectangle by adding an angle parameter.

[0123] The recognition network training unit is used to perform feature extraction, badge recognition, foreign object intrusion recognition and personnel recognition training on the annotated image to be recognized based on the image recognition network, until the image recognition prediction result and the annotated information of the image to be recognized reach a preset accuracy threshold or the training rounds reach a preset training number threshold, thereby obtaining a trained image recognition network;

[0124] The image recognition network uses the backbone feature extraction network and feature fusion network of the YOLOv5s neural network model to extract the labeled features, and performs work badge recognition, foreign object intrusion recognition and personnel recognition through the prediction network of the YOLOv5s neural network model to obtain image recognition results.

[0125] In this embodiment, the YOLOv5s neural network model is an existing neural network model, and is also the network model with the smallest depth and the smallest feature map width in the YOLOv5 neural network model series. It only needs to perform a forward pass on the entire image once to complete the object detection task, so it has very high real-time performance; the YOLOv5s neural network model can detect multiple objects in an image at the same time, and can also perform better detection and positioning for overlapping objects; the YOLOv5s neural network model is not only suitable for target detection in static images, but also for applications such as video target tracking. At the same time, its lightweight characteristics enable it to run effectively in resource-limited environments (such as embedded devices or mobile applications). In this embodiment, the annotation information of the image to be identified is the content annotated by the roLabelImg image annotation tool to the annotated object in the enhanced image to be identified.

[0126] The image recognition network is trained to recognize work badges, foreign objects and personnel according to the image prediction loss function;

[0127] The calculation expressions of the image prediction loss function are as follows:

[0128] Loss = λ cls L cls +λ reg L reg ,

[0129]

[0130] Among them, Loss represents the image prediction loss function, λ cls Represents the weight coefficient of classification loss, L cls represents the classification loss function, λ reg represents the weight coefficient of bounding box regression loss, L reg represents the bounding box regression loss function, i represents the i-th annotated image to be recognized, N represents the total number of annotated images to be recognized, j represents the j-th recognition category, C represents the total number of recognition categories, and y ij represents the prediction result of the i-th labeled image to be identified of the j-th category, log represents the logarithmic function, represents the label information of the work badge on the labeled image to be identified, x represents the parameter difference between the predicted bounding box and the true bounding box, |·| represents the absolute value operation, otherwise represents other cases, and ti represents the predicted bounding box parameters of the i-th annotated image to be recognized, Represents the parameters of the real bounding box of the i-th annotated image to be identified. In this embodiment, the identification categories include personnel, work badges, railway tracks, trains and foreign objects on railway tracks in the station area, a total of 5 categories, that is, C=5; the parameter differences between the predicted bounding box and the real bounding box include the differences in center coordinates, width, and height.

[0131] The image recognition unit is used to use the trained image recognition network to perform work badge recognition, foreign object intrusion recognition and personnel recognition on the enhanced image to be recognized corresponding to the newly acquired station area image to obtain an image recognition result.

[0132] The foreign object intrusion warning submodule is used to determine the foreign object intrusion on the railway track based on the foreign object intrusion recognition result in the image recognition result. If there is foreign object intrusion on the railway track, a foreign object intrusion alarm is issued; in this embodiment, the foreign object intrusion on the railway track refers to the presence of foreign objects on the railway track.

[0133] The work badge away from the person warning submodule is used to judge the work badge away from the person based on the positioning data of the electronic work badge, the work badge recognition result in the image recognition result and the personnel recognition result. If the work badge is away from the person, an alarm will be issued. In this embodiment, the work badge away from the person refers to the situation that the positioning data of the electronic work badge has no position update within the preset displacement monitoring time, and the work badge recognition result and the personnel recognition result are separated in the image, wherein the pixel distance between the center points of the bounding boxes of the work badge recognition result and the personnel recognition result in the image can be used as a basis for judging their separation in the image. When the pixel distance between the center point of the bounding box of the work badge recognition result and the center point of the bounding box of the personnel recognition result is greater than the width of the bounding box of the personnel recognition result, it means that the work badge recognition result and the personnel recognition result are separated in the image.

[0134] The abnormal intrusion warning submodule is used to combine the abnormal intrusion data of the fence with the personnel recognition results to judge the abnormal intrusion of the personnel. If there is an abnormal intrusion of the personnel, an abnormal intrusion warning is issued. In this embodiment, the abnormal intrusion of the personnel refers to the situation where the shooting time information corresponding to the personnel recognition result of the preset prohibited area and the real-time positioning time information of the corresponding electronic work card are within the preset recognition time error range.

[0135] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A station area security risk card control system, characterized in that: include: The video monitoring module is used to shoot the station area at all times and continuously obtain station area images with shooting time information from several different perspectives; Image processing and transmission module, used to process station area images and transmit the processed images to be identified to the security control platform; The electronic badge module is used to be worn by personnel in the station area. It generates the positioning data of the electronic badge by receiving differential data and satellite signals; Beidou positioning module is used to build a virtual fence in the station area, and to perform real-time positioning of the electronic work badge by playing differential data through the Beidou differential base station, and to transmit the positioning data of the virtual fence and electronic work badge in the station area to the security control platform; The security control alarm platform is used to perform image recognition based on the image recognition network according to the image to be recognized, and to perform positioning detection based on the positioning data of the electronic work badge and the virtual fence of the station area. It also issues foreign object intrusion warnings, work badge absence warnings, and abnormal personnel intrusion warnings based on the fence abnormal intrusion data and image recognition results.

2. The station area security risk card control system according to claim 1 is characterized in that: The image processing and transmission module comprises: The image preprocessing submodule is used to preprocess the station area image to obtain the image to be recognized; The image transmission submodule is used to transmit the image to be identified to the security control platform through the wireless access point and the 5G network.

3. The station area security risk card control system according to claim 2 is characterized in that: The image preprocessing module comprises: The first image processing unit is used to convert the original station area image into a grayscale image to obtain a station area grayscale image; A second image processing unit, used for calculating the local contrast of the grayscale image of the station area through a Laplace operator; The calculation expression of the local contrast of the grayscale image of the station area is as follows: C(x,y)=|▽ 2 f(x,y)|, Among them, C(x,y) represents the local contrast of the grayscale image of the station area, ▽ 2 f(x,y) represents the result of convolution between the grayscale image of the station area and the Laplacian filter, and |·| represents the absolute value operation; A third image processing unit, used for calculating a nonlinear gain factor according to a local contrast of the grayscale image of the station area; The calculation expression of the nonlinear gain factor is as follows: Among them, α represents the nonlinear gain factor, C represents the local contrast of the station area image, and Q represents the preset contrast threshold; A fourth image processing unit, configured to multiply the grayscale value of each pixel in the original station area image by the nonlinear gain factor corresponding to the pixel position according to the nonlinear gain, to obtain a grayscale image of the station area after gain; A fifth image processing unit, used for adjusting the grayscale image of the station area after gain to obtain a dark primary color station area image; The calculation expression of the dark primary color station area image is as follows: I(x,y)=c(x,y)×l(x,y)+A×(1-I(x,y)), Among them, I dark (x,y) represents the dark original color station area image, It means taking the minimum value based on the pixel color value of the original station area image, c dark (x, y) represents the pixel dark original color value of the dark original color station area image, I(x, y) represents the original station area image, c(x, y) represents the pixel color value, l(x, y) represents the light transmittance of the pixel, and A represents the atmospheric light value; A sixth image processing unit, used for obtaining an image to be identified based on the dark primary color station area image, a Laplace filter and an atmospheric light value; The calculation expression of the image to be recognized is as follows: Among them, I′(x, y) represents the image to be identified, λ′(x, y) represents the propagation map of the second station area, λ(x, y) represents the propagation map of the first station area, ε represents the propagation processing factor, [L] represents the Laplace filter, and χ represents the propagation auxiliary parameter.

4. The station area security risk card control system according to claim 3 is characterized in that: The image transmission submodule comprises: An image switch unit, for receiving an image to be identified and wirelessly transmitting the image to be identified to a wireless access point; A plurality of wireless access point units, used for wirelessly receiving the image to be identified transmitted by the corresponding image switch, and transmitting the image to be identified to the main wireless access point through a 5G signal; The main wireless access point unit is used to transmit the image to be identified to the security control platform via the network cable.

5. The station area security risk card control system according to claim 4 is characterized in that: The Beidou positioning module includes: The virtual fence submodule is used to set up virtual fences for preset restricted areas in the station area based on high-precision Beidou maps; Several Beidou differential base stations are used to generate differential data, broadcast the differential data to the electronic work badges within the coverage area of ​​each Beidou differential base station, and transmit the positioning data of the electronic work badges in the area to the security control platform.

6. The station area security risk card control system according to claim 5 is characterized in that: The positioning data of the electronic badge includes the real-time positioning time information, the real-time positioning position information and the basic information of the person corresponding to the electronic badge.

7. The station area security risk card control system according to claim 5 is characterized in that: The security control alarm platform includes: The positioning detection submodule is used to judge the abnormal fence intrusion according to the positioning data of the electronic work badge and the virtual fence of the station area. If the positioning data of the electronic work badge is within the range of the virtual fence of the station area, the abnormal fence intrusion data is generated; The image recognition submodule is used to train the image recognition network based on the image to be recognized, and perform work badge recognition, foreign object intrusion recognition and personnel recognition on the enhanced image to be recognized corresponding to the newly acquired station area image, to obtain an image recognition result, wherein the image recognition result is a combination of the work badge recognition result, foreign object intrusion recognition result and personnel recognition result obtained after performing work badge recognition, foreign object intrusion recognition and personnel recognition on the same enhanced image to be recognized; The foreign object intrusion warning submodule is used to identify the foreign object intrusion on the railway track according to the foreign object intrusion recognition result in the image recognition result. If there is foreign object intrusion on the railway track, an intrusion alarm is issued; The badge-away warning submodule is used to determine if the badge-away person is present based on the positioning data of the electronic badge, the badge recognition result in the image recognition result, and the person recognition result. If the badge-away person is present, an alarm is triggered; The abnormal intrusion warning submodule is used to combine the abnormal intrusion data of fences with the personnel recognition results to judge the abnormal intrusion of personnel. If there is an abnormal intrusion of personnel, an abnormal intrusion warning will be issued.

8. The station area security risk card control system according to claim 7 is characterized in that: The image recognition submodule comprises: A data set unit, used to construct a data set of images to be identified based on the received images to be identified; A data enhancement unit is used to randomly scale, randomly crop, randomly arrange and splice the images to be identified in the image data set to be identified, so as to obtain enhanced images to be identified; The adaptive anchor frame unit is used to annotate the enhanced image to be identified by using the roLabelImg image annotation tool to obtain an annotated image to be identified, wherein the annotated objects in the image to be identified include personnel, work badges, railway tracks, trains and foreign objects on the railway tracks in the station area; The recognition network training unit is used to perform feature extraction, badge recognition, foreign object intrusion recognition and personnel recognition training on the annotated image to be recognized based on the image recognition network, until the image recognition prediction result and the annotated information of the image to be recognized reach a preset accuracy threshold or the training rounds reach a preset training number threshold, thereby obtaining a trained image recognition network; The image recognition unit is used to use the trained image recognition network to perform work badge recognition, foreign object intrusion recognition and personnel recognition on the enhanced image to be recognized corresponding to the newly acquired station area image to obtain an image recognition result.

9. The station area security risk card control system according to claim 8 is characterized in that: The image recognition network uses the backbone feature extraction network and feature fusion network of the YOLOv5s neural network model to extract the labeled features, and performs work badge recognition, foreign object intrusion recognition and personnel recognition through the prediction network of the YOLOv5s neural network model to obtain image recognition results.

10. The station area security risk card control system according to claim 9, characterized in that: The image recognition network is trained to recognize work badges, foreign objects and personnel according to the image prediction loss function; The calculation expressions of the image prediction loss function are as follows: Loss=λclsLcls+λregLreg, Among them, Loss represents the image prediction loss function, λcls represents the weight coefficient of classification loss, Lcls represents the classification loss function, λreg represents the weight coefficient of bounding box regression loss, Lreg represents the bounding box regression loss function, i represents the i-th annotated image to be recognized, N represents the total number of annotated images to be recognized, j represents the j-th recognition category, C represents the total number of recognition categories, yij represents the prediction result of the i-th annotated image to be recognized of the j-th recognition category, log represents the logarithmic function, represents the labeling information of the work badge on the labeled image to be identified, x represents the parameter difference between the predicted bounding box and the true bounding box, |·| represents the absolute value operation, otherwise represents other situations, ti represents the predicted bounding box parameters of the i-th labeled image to be identified, Represents the true bounding box parameters of the i-th annotated image to be recognized.

Images