Intrusion detection method, detection device and subway detection vehicle

Abstract

The application provides an intrusion detection method, a detection device and a subway detection vehicle. The method comprises the following steps: when a vehicle runs at a normal driving speed in a platform tunnel area, obtaining a plurality of first point cloud coordinates and a plurality of first video data of the platform tunnel area; comparing the plurality of first point cloud coordinates with preconfigured standard point cloud coordinates to determine a potential intrusion point in the platform tunnel area; obtaining second video data corresponding to the potential intrusion point from the plurality of first video data, and comparing the second video data with preconfigured standard video data to obtain a loss value corresponding to the second video data; and determining that an intrusion failure occurs at the potential intrusion point when it is determined that the loss value is greater than a loss threshold. The problem that the prior art cannot accurately detect whether an intrusion failure occurs at the platform tunnel area of the platform door is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rail transit, and in particular to an intrusion detection method, a detection device and a subway detection vehicle. BACKGROUND

[0002] In order to ensure the safe driving of subway vehicles, the standard limit profile of the subway line is usually calibrated as the limit of the subway line, and any building, equipment and facility cannot exceed the limit, otherwise it is intrusion. In the prior art, three-dimensional point cloud data of the subway tunnel section is collected by a line laser or the like, and if the coordinates of the collected point cloud data are within the standard limit profile, it is determined that there is an intrusion component at the position corresponding to the point cloud data, and an intrusion failure occurs.

[0003] However, there are transparent doors, light-absorbing rubber strips and other components at the platform tunnel area of the platform screen door, which can easily cause problems such as laser projection and laser absorption, so that the line laser or the like cannot accurately obtain the three-dimensional point cloud data at the platform screen door, and it is difficult to accurately detect whether an intrusion failure has occurred at the platform screen door in the platform tunnel area. SUMMARY

[0004] The present application provides an intrusion detection method, a detection device and a subway detection vehicle, which are used to solve the problem that the prior art cannot accurately detect whether an intrusion failure has occurred at the platform screen door in the platform tunnel area.

[0005] In a first aspect, the present application provides an intrusion detection method, comprising: acquiring a plurality of first point cloud coordinates and a plurality of first video data of a platform tunnel area when a vehicle is running at a normal driving speed in the platform tunnel area; comparing the plurality of first point cloud coordinates with preconfigured standard point cloud coordinates to determine potential intrusion points in the platform tunnel area; acquiring second video data corresponding to the potential intrusion points from the plurality of first video data, and comparing the second video data with preconfigured standard video data to obtain a loss value corresponding to the second video data; and determining that an intrusion failure has occurred at the potential intrusion point when it is determined that the loss value is greater than a loss threshold.

[0006] In a specific embodiment, the preconfigured standard point cloud coordinates are acquired by triggering the vehicle to run at least once in the platform tunnel area at a test speed, and the plurality of second point cloud coordinates of the platform tunnel area acquired are used as the preconfigured standard point cloud coordinates; wherein the test speed is less than the normal driving speed.

[0007] In one specific embodiment, the intrusion detection method further includes: classifying the standard point cloud coordinates according to kilometer markers to obtain a first subset formed by multiple standard point cloud coordinates corresponding to each kilometer marker in the platform tunnel area; then comparing the multiple first point cloud coordinates with pre-configured standard point cloud coordinates to determine potential intrusion points in the platform tunnel area includes: classifying the multiple first point cloud coordinates according to kilometer markers to obtain a second subset formed by multiple first point cloud coordinates corresponding to each kilometer marker in the platform tunnel area; for each first point cloud coordinate in the second subset corresponding to each kilometer marker in the platform tunnel area, obtaining at least one standard point cloud coordinate from the first subset corresponding to the kilometer marker that is less than a preset distance from the first point cloud coordinate, and obtaining the average distance between the first point cloud coordinate and the at least one standard point cloud coordinate; when it is determined that the average distance is greater than a preset distance threshold, and the first point cloud coordinate is located in the platform door area within the platform tunnel area, the point corresponding to the first point cloud coordinate is determined to be a potential intrusion point.

[0008] In one specific embodiment, the encroachment detection method further includes: classifying the standard point cloud coordinates according to kilometer markers to obtain a first subset formed by multiple standard point cloud coordinates corresponding to each kilometer marker within the platform tunnel area; then comparing the multiple first point cloud coordinates with pre-configured standard point cloud coordinates to determine potential encroachment points within the platform tunnel area includes: classifying the multiple first point cloud coordinates according to kilometer markers to obtain a second subset formed by multiple first point cloud coordinates corresponding to each kilometer marker within the platform tunnel area; for each kilometer marker within the platform tunnel area, the second subset {V k，s For each first point cloud coordinate in}, traverse the first subset. Get The minimum multiple standard point cloud coordinates are determined, and the average distance between the first point cloud coordinates and the multiple standard point cloud coordinates is obtained; when the average distance is determined to be greater than a preset distance threshold, and the first point cloud coordinates are located in the platform tunnel area within the platform screen door area, the point corresponding to the first point cloud coordinates is determined to be a potential intrusion point; wherein, V k，s Let the first point be the cloud coordinates. Here, m represents the standard point cloud coordinates, s represents the kilometer marker, and k represents the first point cloud coordinate sequence.

[0009] In an embodiment, the preconfigured standard video data is obtained by triggering the vehicle to run in the station tunnel area at a test speed at least once, and the obtained third video data of the station tunnel area is taken as the preconfigured standard video data, wherein the test speed is less than the normal driving speed.

[0010] In an embodiment, the intrusion detection method further comprises: training the standard video data by using a deep learning network to obtain a standard feature vector; and comparing the second video data with the preconfigured standard video data to obtain a loss value corresponding to the second video data, comprising: extracting features of the second video data by using a multi-layer convolutional neural network to obtain a feature vector of the second video data; obtaining channel weights and spatial pixel weights of the feature vector respectively, and performing weighted processing on the feature vector by using the channel weights and the spatial pixel weights respectively to obtain a channel-weighted feature vector and a spatial-weighted feature vector; multiplying the channel-weighted feature vector and the spatial-weighted feature vector to obtain a weighted feature vector, and performing similarity calculation on the weighted feature vector and the standard feature vector to obtain a similarity weight of the weighted feature vector; performing weighted processing on the standard feature vector by using the similarity weight to obtain a weighted standard feature vector; performing video reconstruction on the weighted standard feature vector, and comparing the reconstructed video data with the standard video data to obtain the loss value corresponding to the second video data.

[0011] In a second aspect, the application provides a detection device, comprising: a line laser scanner configured to obtain a plurality of first point cloud coordinates of a station tunnel area when a vehicle runs in the station tunnel area at a normal driving speed; a video acquisition module configured to obtain a plurality of first video data of the station tunnel area when the vehicle runs in the station tunnel area at the normal driving speed; and a processing module configured to compare the plurality of first point cloud coordinates with preconfigured standard point cloud coordinates to determine potential intrusion points in the station tunnel area, and configured to obtain second video data corresponding to the potential intrusion points from the plurality of first video data, and compare the second video data with preconfigured standard video data to obtain a loss value corresponding to the second video data, and configured to determine that an intrusion fault occurs at the potential intrusion point when the loss value is greater than a loss threshold.

[0012] In an embodiment, the processing module is further configured to trigger the vehicle to run in the station tunnel area at a test speed at least once, and obtain a plurality of second point cloud coordinates of the station tunnel area as the preconfigured standard point cloud coordinates respectively; wherein the test speed is less than the normal driving speed.

[0013] In an embodiment, the processing module is further configured to classify the standard point cloud coordinates according to the kilometer markers to obtain a first subset of a plurality of standard point cloud coordinates corresponding to each kilometer marker in the station tunnel area; and the processing module is specifically configured to classify the plurality of first point cloud coordinates according to the kilometer markers to obtain a second subset of a plurality of first point cloud coordinates corresponding to each kilometer marker in the station tunnel area; for each first point cloud coordinate in the second subset corresponding to each kilometer marker in the station tunnel area, obtain at least one standard point cloud coordinate within a preset distance from the first point cloud coordinate from the first subset corresponding to the kilometer marker, and obtain an average distance between the first point cloud coordinate and the at least one standard point cloud coordinate; and determine that a point corresponding to the first point cloud coordinate is a potential intrusion point when it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in a platform door area in the station tunnel area.

[0014] In an embodiment, the processing module is further configured to classify the standard point cloud coordinates according to the kilometer markers to obtain a first subset of a plurality of standard point cloud coordinates corresponding to each kilometer marker in the station tunnel area; and the processing module is specifically configured to classify the plurality of first point cloud coordinates according to the kilometer markers to obtain a second subset of a plurality of first point cloud coordinates corresponding to each kilometer marker in the station tunnel area; for each first point cloud coordinate in the second subset corresponding to each kilometer marker in the station tunnel area, traverse the first subset k，s} corresponding to the kilometer marker to obtain a plurality of standard point cloud coordinates closest to the first point cloud coordinate, and obtain an average distance between the first point cloud coordinate and the plurality of standard point cloud coordinates; and determine that a point corresponding to the first point cloud coordinate is a potential intrusion point when it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in a platform door area in the station tunnel area. In an embodiment, the processing module is further configured to classify the standard point cloud coordinates according to the kilometer markers to obtain a first subset of a plurality of standard point cloud coordinates corresponding to each kilometer marker in the station tunnel area; and the processing module is specifically configured to classify the plurality of first point cloud coordinates according to the kilometer markers to obtain a second subset of a plurality of first point cloud coordinates corresponding to each kilometer marker in the station tunnel area; for each first point cloud coordinate in the second subset corresponding to each kilometer marker in the station tunnel area, traverse the first subset k，s} corresponding to the kilometer marker to obtain a plurality of standard point cloud coordinates closest to the first point cloud coordinate, and obtain an average distance between the first point cloud coordinate and the plurality of standard point cloud coordinates; and determine that a point corresponding to the first point cloud coordinate is a potential intrusion point when it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in a platform door area in the station tunnel area.

[0015] ​​In an embodiment, the processing module is further configured to trigger the vehicle to run at a test speed in the station tunnel area at least once, and acquire third video data of the station tunnel area as the preconfigured standard video data; wherein the test speed is less than the normal driving speed.

[0016] In an embodiment, the processing module is further configured to train the standard video data by using a deep learning network to obtain a standard feature vector; the processing module comprises: an encoder submodule configured to extract features of the second video data by using a multi-layer convolutional neural network to obtain a feature vector of the second video data; a self-attention submodule configured to respectively acquire channel weights and spatial pixel weights of the feature vector, and respectively perform weighted processing on the feature vector by using the channel weights and the spatial pixel weights to obtain a channel-weighted feature vector and a spatial-weighted feature vector; a dynamic prototype learning submodule configured to multiply the channel-weighted feature vector and the spatial-weighted feature vector to obtain a weighted feature vector, and perform similarity calculation on the weighted feature vector and the standard feature vector to obtain a similarity weight of the weighted feature vector; the dynamic prototype learning submodule is further configured to perform weighted processing on the standard feature vector by using the similarity weight to obtain a weighted standard feature vector; and a decoder submodule configured to perform video reconstruction on the weighted standard feature vector, and compare the reconstructed video data with the standard video data to obtain a loss value corresponding to the second video data.

[0017] In a third aspect, the application provides a subway detection vehicle, comprising the detection device according to the second aspect and a vehicle body.

[0018] The application provides a limit intrusion detection method, a detection device and a subway detection vehicle. The method comprises the following steps: obtaining a plurality of first point cloud coordinates and a plurality of first video data of a platform tunnel area when a vehicle runs at a normal running speed in the platform tunnel area; comparing the plurality of first point cloud coordinates with preconfigured standard point cloud coordinates to determine a potential limit intrusion point in the platform tunnel area; obtaining second video data corresponding to the potential limit intrusion point from the plurality of first video data, and comparing the second video data with preconfigured standard video data to obtain a loss value corresponding to the second video data; and determining that a limit intrusion fault occurs at the potential limit intrusion point when the loss value is greater than a loss threshold. Compared with the prior art, the application can accurately determine whether a limit intrusion fault occurs at the platform tunnel area of the platform tunnel area. After comparing the plurality of first point cloud coordinates of the platform tunnel area with the preconfigured standard point cloud coordinates to determine the potential limit intrusion point in the platform tunnel area, the application compares the second video data corresponding to the potential limit intrusion point with the preconfigured standard video data to obtain the loss value corresponding to the second video data, thereby determining whether the potential limit intrusion point has a limit intrusion fault, effectively improving the accuracy of limit intrusion detection at the platform tunnel area of the platform tunnel area, and solving the problem that the prior art cannot accurately determine whether a limit intrusion fault occurs at the platform tunnel area of the platform tunnel area. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without any creative labor.

[0020] Figure 1 A schematic diagram of a subway detection vehicle provided by the application applied in a tunnel scene;

[0021] Figure 2 A flowchart of a limit intrusion detection method provided by the application embodiment one;

[0022] Figure 3 A flowchart of a limit intrusion detection method provided by the application embodiment two;

[0023] Figure 4 A flowchart of a limit intrusion detection method provided by the application embodiment three;

[0024] Figure 5a A flowchart of a limit intrusion detection method provided by the application embodiment four;

[0025] Figure 5b A feature vector obtained by weighting a channel a schematic view of the embodiment of the application;

[0026] Figure 5c to obtain a spatially weighted feature vector a schematic view of the embodiment of the application;

[0027] Figure 6 a structural schematic view of an embodiment of a detection device provided by the application;

[0028] Figure 7 a structural schematic view of a processing module of an embodiment of a detection device provided by the application. DETAILED DESCRIPTION

[0029] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments made by those of ordinary skill in the art under the inspiration of the embodiments of the present application belong to the scope of protection of the present application.

[0030] The terms “first”, “second”, “third”, “fourth” and the like (if any) in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms “include” and “have” and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0031] In order to ensure the safe running of the subway vehicle, the standard limit profile of the subway line will be calibrated as the limit of the subway line, and any building, equipment and facility cannot exceed the limit, otherwise it is an intrusion limit. In the prior art, three-dimensional point cloud data of the subway tunnel section is collected by a line laser or the like, and if the coordinates of the collected point cloud data are within the standard limit profile, it is determined that there is an intrusion limit component at the position corresponding to the point cloud data, and an intrusion limit failure occurs.

[0032] However, there are transparent doors, light-absorbing rubber strips and other components at the platform screen door in the subway platform tunnel area, which can easily cause problems such as laser projection and laser absorption, so that the line laser or the like cannot accurately obtain the three-dimensional point cloud data at the platform screen door, and it is difficult to accurately detect whether there is an intrusion limit component at the platform screen door in the platform tunnel area.

[0033] Based on the above technical problems, the technical concept of the present application is as follows: how to accurately detect whether there is an intrusion component at the platform tunnel area of the shield door.

[0034] The intrusion detection scheme of the present application will be described in detail below.

[0035] Figure 1 A schematic diagram of a subway detection vehicle provided by the present application is applied in a tunnel scene. As shown in the figure, Figure 1 The scene includes a subway tunnel, a platform, a shield door between the platform and the tunnel, and a subway detection vehicle running on the subway tunnel. Among them, the subway detection vehicle mainly includes a detection device 11 and a vehicle body 12. Among them, the detection device 11 includes a line laser scanner 111, a video acquisition module 121, and a processing module 131.

[0036] The line laser scanner 111 obtains a plurality of first point cloud coordinates of the platform tunnel area when the vehicle runs at a normal running speed in the platform tunnel area. The platform tunnel area includes a plurality of shield doors 13. The video acquisition module 121 obtains a plurality of first video data of the platform tunnel area when the vehicle runs at a normal running speed in the platform tunnel area. The processing module 131 compares the plurality of first point cloud coordinates with the preconfigured standard point cloud coordinates to determine the potential intrusion point in the platform tunnel area, and obtains the second video data corresponding to the potential intrusion point from the plurality of first video data. The second video data is compared with the preconfigured standard video data to obtain the loss value corresponding to the second video data, and when the loss value is greater than the loss threshold, it is determined that the intrusion failure occurs at the potential intrusion point.

[0037] Next, the technical scheme of the present application will be described in detail through specific embodiments. It should be noted that the following several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0038] Figure 2 A flowchart of an embodiment one of an intrusion detection method provided by the present application is shown in the figure. Figure 2 The intrusion detection method specifically includes the following steps:

[0039] Step S201: When the vehicle runs at a normal running speed in the platform tunnel area, a plurality of first point cloud coordinates and a plurality of first video data of the platform tunnel area are obtained.

[0040] In the embodiment, the platform tunnel area refers to a tunnel area including a platform door area. The detection vehicle runs in the platform tunnel area at a normal driving speed. The detection device obtains a plurality of first point cloud coordinates of the platform tunnel area through a line laser scanner and obtains a plurality of first video data of the platform tunnel area through a video acquisition module. Exemplarily, the normal driving speed can be 30 km / h.

[0041] Step S202: Comparing the plurality of first point cloud coordinates with preconfigured standard point cloud coordinates to determine a potential intrusion point in the platform tunnel area.

[0042] In the embodiment, the preconfigured standard point cloud coordinates can be obtained by the detection device through the line laser scanner when the detection vehicle runs in the platform tunnel area at a test speed multiple times after it is determined that no intrusion fault occurs in the platform tunnel area. Exemplarily, the test speed can be less than or equal to 5 km / h.

[0043] Comparing the plurality of first point cloud coordinates with the preconfigured standard point cloud coordinates can determine the potential intrusion point in the platform tunnel area. At the potential intrusion point, an intrusion fault can occur.

[0044] Step S203: Obtaining second video data corresponding to the potential intrusion point from the plurality of first video data and comparing the second video data with preconfigured standard video data to obtain a loss value corresponding to the second video data.

[0045] In the embodiment, the second video data corresponding to the potential intrusion point is obtained from the plurality of first video data obtained by the video acquisition module to further determine whether an intrusion fault occurs at the potential intrusion point.

[0046] Specifically, the second video data is compared with the preconfigured standard video data to obtain a loss value corresponding to the second video data. The preconfigured standard video data can be obtained by the detection device through the video acquisition module when the detection vehicle runs in the platform tunnel area at a test speed multiple times after it is determined that no intrusion fault occurs in the platform tunnel area.

[0047] Step S204: When it is determined that the loss value is greater than a loss threshold, it is determined that an intrusion fault occurs at the potential intrusion point.

[0048] In the embodiment, when the vehicle runs in the platform tunnel area at a normal driving speed, a plurality of first point cloud coordinates and a plurality of first video data of the platform tunnel area are acquired; the plurality of first point cloud coordinates are compared with the preconfigured standard point cloud coordinates to determine a potential intrusion point in the platform tunnel area; from the plurality of first video data, second video data corresponding to the potential intrusion point is acquired, and the second video data is compared with the preconfigured standard video data to acquire a loss value corresponding to the second video data; when it is determined that the loss value is greater than a loss threshold, it is determined that an intrusion fault occurs at the potential intrusion point. Compared with the prior art which is difficult to accurately detect whether an intrusion fault occurs at the platform tunnel area, the application compares the plurality of first point cloud coordinates of the platform tunnel area acquired with the preconfigured standard point cloud coordinates to determine the potential intrusion point in the platform tunnel area, and then compares the second video data corresponding to the potential intrusion point with the preconfigured standard video data to acquire the loss value corresponding to the second video data, so as to determine whether the potential intrusion point has an intrusion fault, thereby effectively improving the accuracy of intrusion detection at the platform tunnel area, and solving the problem that the prior art is difficult to accurately detect whether an intrusion fault occurs at the platform tunnel area.

[0049] Figure 3 For the flowchart of the intrusion detection method embodiment two provided by the application, on the basis of the above-mentioned Figure 2 embodiment, referring to Figure 3 , before the above-mentioned step S202, the intrusion detection method further comprises:

[0050] Step S301: According to the kilometer marker, the standard point cloud coordinates are classified and processed to acquire a first subset formed by a plurality of standard point cloud coordinates corresponding to each kilometer marker in the platform tunnel area.

[0051] In the embodiment, when it is determined that no intrusion fault occurs in the platform tunnel area, the vehicle can be triggered to run in the platform tunnel area at a test speed at least once, and the plurality of second point cloud coordinates of the platform tunnel area acquired are taken as the preconfigured standard point cloud coordinates; wherein the test speed is less than the normal driving speed. For example, the test speed can be 5 km / h, and the normal driving speed can be 30 km / h.

[0052] In the embodiment, the processing module classifies and processes the standard point cloud coordinates acquired by the line laser scanner according to the kilometer marker.

[0053] Specifically, in order to obtain the accurate position of the detection vehicle in motion, an electronic tag RFID can be arranged at the starting point and the ending point of the platform tunnel area respectively, and combined with the encoder installed on the wheel of the detection vehicle, the accurate position of the detection vehicle can be obtained, and the position of the detection vehicle is expressed in the form of a kilometer marker, for example, 1 meter, 2 meters, …, from the starting point of the platform tunnel area. According to the kilometer marker, the standard point cloud coordinates are classified and processed to obtain a first subset formed by a plurality of standard point cloud coordinates corresponding to each kilometer marker in the platform tunnel area.

[0054] The first subset can be represented as V = {V , wherein m represents the serial number of the standard point cloud coordinates, s represents the kilometer marker, and B represents the standard point cloud. m e N, N is the number of standard point cloud coordinates corresponding to the kilometer marker. Exemplarily, N can be 1000. Exemplarily, is the first subset of the kilometer marker “1 meter from the starting point of the platform tunnel area”.

[0055] In this embodiment, by classifying and processing the standard point cloud coordinates according to the kilometer marker, all the standard point cloud coordinates can be arranged according to the kilometer marker and combined into a standard point cloud cloud chart of the platform tunnel area according to the mileage corresponding to the kilometer marker.

[0056] The standard point cloud cloud chart is subjected to a culling processing, and points more than 5 meters away from the center of the platform tunnel area are removed to obtain a standard point cloud cloud chart after removing irrelevant noise points. The standard point cloud cloud chart after the culling processing can also be subjected to a further filtering processing, and a neighborhood mean filtering is adopted to remove white noise introduced in the detection process of the line laser scanner.

[0057] Before the above step S202, the intrusion detection method further includes a step S301, and the above step S202 specifically includes the following steps:

[0058] Step S302: According to the kilometer marker, the plurality of first point cloud coordinates are classified and processed to obtain a second subset formed by a plurality of first point cloud coordinates corresponding to each kilometer marker in the platform tunnel area.

[0059] In this embodiment, the processing module classifies and processes the plurality of first point cloud coordinates obtained by the line laser scanner according to the kilometer marker.

[0060] Specifically, the position of the detection vehicle can be expressed in the form of a kilometer marker, for example, 1 meter, 2 meters, …, from the starting point of the platform tunnel area. According to the kilometer marker, the plurality of first point cloud coordinates are classified and processed to obtain a second subset formed by a plurality of first point cloud coordinates corresponding to each kilometer marker in the platform tunnel area.

[0061] The second subset can be represented as V = {V k，s} represents, where k represents the serial number of the first point cloud coordinate, and s represents the kilometer marker. k e N, N is the number of standard point cloud coordinates corresponding to the kilometer marker. Exemplarily, N can be 1000. Exemplarily, {V k，1} is a second subset corresponding to the kilometer marker "1 meter away from the starting point of the platform tunnel region".

[0062] Step S303: For each first point cloud coordinate in the second subset corresponding to each kilometer marker in the platform tunnel region, at least one standard point cloud coordinate less than a preset distance from the first point cloud coordinate is obtained from the first subset corresponding to the kilometer marker, and the average distance between the first point cloud coordinate and the at least one standard point cloud coordinate is obtained.

[0063] In this embodiment, for each first point cloud coordinate in the second subset corresponding to each kilometer marker, at least one standard point cloud coordinate less than a preset distance from the first point cloud coordinate is obtained from the first subset corresponding to the kilometer marker. Exemplarily, for the first point cloud coordinate V k，1 in the second subset corresponding to the kilometer marker "1 meter away from the starting point of the platform tunnel region" {V 1，1 , at least one standard point cloud coordinate less than a preset distance from the first point cloud coordinate V 1，1 is obtained from the first subset corresponding to the kilometer marker, and the average distance between the first point cloud coordinate V 1，1 and the at least one standard point cloud coordinate is obtained. Exemplarily, from the first subset , 5 standard point cloud coordinates less than a preset distance from the first point cloud coordinate V 1，1 are obtained, and the average distance between the first point cloud coordinate V 1，1 and the 5 standard point cloud coordinates is obtained.

[0064] Step S304: When it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in the platform tunnel region, it is determined that the point corresponding to the first point cloud coordinate is a potential intrusion point.

[0065] In this embodiment, it is determined whether the average distance is greater than a preset distance threshold, and when it is determined that the average distance is greater than a preset distance threshold, it is determined whether the first point cloud coordinate is located in the platform tunnel region. When it is determined that the first point cloud coordinate is located in the platform tunnel region, it is determined that the point corresponding to the first point cloud coordinate is a potential intrusion point, and at the potential intrusion point, an intrusion fault may have occurred.

[0066] In the embodiment, the standard point cloud coordinates and the first point cloud coordinates are classified according to the milepost, to obtain a first subset and a second subset corresponding to each milepost in the platform tunnel area, and the first point cloud coordinates in the second subset are compared with the standard point cloud coordinates in the first subset to determine the potential intrusion point in the platform tunnel area. By comparing the point cloud coordinates with the point cloud coordinates when no intrusion fault occurs in the platform tunnel area, instead of comparing with the standard limit contour, the accuracy of the intrusion detection at the platform screen door is effectively improved, which provides a prerequisite for further determining the intrusion fault according to the video data, and further solves the problem that the prior art cannot accurately detect whether an intrusion fault occurs at the platform screen door in the platform tunnel area.

[0067] Figure 4 An embodiment of a flowchart of an intrusion detection method provided in the present application is shown in the above Figure 2 Based on the embodiment shown in the above Figure 4 Before the step S202, the intrusion detection method further includes:

[0068] Step S401: According to the milepost, the standard point cloud coordinates are classified to obtain a first subset formed by a plurality of standard point cloud coordinates corresponding to each milepost in the platform tunnel area.

[0069] In the embodiment, when it is determined that no intrusion fault occurs in the platform tunnel area, the vehicle can be triggered to run at least once in the platform tunnel area at a test speed, and a plurality of second point cloud coordinates of the platform tunnel area obtained are taken as preconfigured standard point cloud coordinates; wherein the test speed is less than the normal driving speed. For example, the test speed can be 5km / h, and the normal driving speed can be 30km / h.

[0070] In the embodiment, the processing module classifies the standard point cloud coordinates obtained by the line laser scanner according to the milepost.

[0071] Specifically, in order to obtain the accurate position of the detection vehicle in motion, an electronic tag RFID can be arranged at the starting point and the ending point of the platform tunnel area, and combined with the encoder installed on the wheels of the detection vehicle, the accurate position of the detection vehicle can be obtained, and the position of the detection vehicle is expressed in the form of milepost, such as 1 meter, 2 meters, …, from the starting point of the platform tunnel area. According to the milepost, the standard point cloud coordinates are classified to obtain a first subset formed by a plurality of standard point cloud coordinates corresponding to each milepost in the platform tunnel area.

[0072] The first subset can be used wherein m represents the serial number of the standard point cloud coordinate, s represents the kilometer mark, and B represents the standard point cloud. m e N, N is the number of standard point cloud coordinates corresponding to the kilometer mark. Exemplarily, N can be 1000. Exemplarily, is the first subset for which the kilometer mark is “1 meter from the start of the platform tunnel region”.

[0073] In this embodiment, by classifying the standard point cloud coordinates according to the kilometer mark, all the standard point cloud coordinates can be arranged according to the mileage corresponding to the kilometer mark, and combined into a standard point cloud cloud chart of the platform tunnel region.

[0074] The standard point cloud cloud chart is subjected to culling processing, and points more than 5 meters away from the center of the platform tunnel region are removed to obtain a standard point cloud cloud chart after removing irrelevant points. Further filtering processing can be performed on the standard point cloud cloud chart after culling processing, and a neighborhood mean filter is used to remove white noise introduced in the detection process of the line laser scanner.

[0075] The above step S202 specifically includes the following steps:

[0076] Step S402: classifying the plurality of first point cloud coordinates according to the kilometer mark to obtain a second subset formed by the plurality of first point cloud coordinates corresponding to each kilometer mark in the platform tunnel region.

[0077] In this embodiment, the processing module classifies the plurality of first point cloud coordinates obtained by the line laser scanner according to the kilometer mark.

[0078] Specifically, the position of the detected vehicle can be represented in the form of a kilometer mark, such as 1 meter, 2 meters, … from the start of the platform tunnel region. The plurality of first point cloud coordinates are classified according to the kilometer mark to obtain a second subset formed by the plurality of first point cloud coordinates corresponding to each kilometer mark in the platform tunnel region.

[0079] The second subset can be represented as {V k，s}, wherein k represents the serial number of the first point cloud coordinate, and s represents the kilometer mark. k e N, N is the number of standard point cloud coordinates corresponding to the kilometer mark. Exemplarily, N can be 1000. Exemplarily, {V k，1} is the second subset for which the kilometer mark is “1 meter from the start of the platform tunnel region”.

[0080] Step S403: for each first point cloud coordinate in the second subset {V k，s} corresponding to each kilometer mark in the platform tunnel region, traversing the first subset to obtain the minimum multiple standard point cloud coordinates, and obtain the average distance between the first point cloud coordinate and the multiple standard point cloud coordinates.

[0081] wherein V k，s is the first point cloud coordinate, is the standard point cloud coordinate, m is the serial number of the standard point cloud coordinate, s is the kilometer marker, and k is the serial number of the first point cloud coordinate.

[0082] In this embodiment, for each first point cloud coordinate in the second subset corresponding to each kilometer marker, the first subset corresponding to the kilometer marker is traversed to obtain the multiple standard point cloud coordinates in the first subset corresponding to the kilometer marker. Illustratively, for the first point cloud coordinate V k，1 in the second subset {V 1，1 corresponding to the kilometer marker “1 meter from the start point of the platform tunnel area”, the first subset corresponding to the kilometer marker is traversed to obtain the multiple standard point cloud coordinates in the first subset corresponding to the kilometer marker. Illustratively, the minimum multiple standard point cloud coordinates in the first subset are obtained, and the average distance between the first point cloud coordinate V 1，1 and the multiple standard point cloud coordinates is obtained. Illustratively, the minimum 5 standard point cloud coordinates in the first subset are obtained, and the average distance between the first point cloud coordinate V 1，1 and the 5 standard point cloud coordinates is obtained.

[0083] Step S404: When it is determined that the average distance is greater than the preset distance threshold and the first point cloud coordinate is located in the screen door area within the platform tunnel area, it is determined that the point corresponding to the first point cloud coordinate is a potential intrusion point.

[0084] In this embodiment, it is determined whether the average distance is greater than the preset distance threshold, and when it is determined that the average distance is greater than the preset distance threshold, it is determined whether the first point cloud coordinate is located in the screen door area within the platform tunnel area, and when it is determined that the first point cloud coordinate is located in the screen door area within the platform tunnel area, it is determined that the point corresponding to the first point cloud coordinate is a potential intrusion point, and an intrusion fault may have occurred at the potential intrusion point.

[0085] In the embodiment, the standard point cloud coordinates and the first point cloud coordinates are classified according to the milepost respectively to obtain the first subset and the second subset corresponding to each milepost in the platform tunnel area, and the first point cloud coordinates in the second subset are compared with the standard point cloud coordinates in the first subset to determine the potential intrusion point in the platform tunnel area. By comparing the point cloud coordinates with the point cloud coordinates in the platform tunnel area when no intrusion fault occurs, instead of comparing with the standard limit contour, the accuracy of the intrusion detection at the platform tunnel area is effectively improved, which provides a prerequisite for further determining the intrusion fault according to the video data, and further solves the problem that the prior art cannot accurately detect whether the intrusion fault occurs at the platform tunnel area.

[0086] Figure 5a The flowchart of a fourth embodiment of an intrusion detection method provided in the application is shown in the above Figure 2 to Figure 4 based on the embodiment shown in the above Figure 5a , before the step S202, the intrusion detection method further includes:

[0087] Step S501: training the standard video data by using the deep learning network to obtain a standard feature vector.

[0088] In the embodiment, the standard video data is trained by using the deep learning network to obtain a standard feature vector, and the standard feature vector is stored to form a storage module memow item.

[0089] Then, the step S202 specifically includes the following steps:

[0090] Step S502: extracting features of the second video data by using a multi-layer convolutional neural network to obtain a feature vector of the second video data.

[0091] In the embodiment, the processing module can include an encoder submodule, which extracts features of the second video data to obtain a feature vector z of the second video data, where z = f(x; θ c ).

[0092] Step S503: respectively obtaining channel weights and spatial pixel weights of the feature vector, and respectively weighting the feature vector by using the channel weights and the spatial pixel weights to obtain a channel-weighted feature vector and a spatial-weighted feature vector.

[0093] In this embodiment, the processing module may further include a self-attention submodule, which employs a squeeze-and-excitation (SE) module to learn in both channel and spatial dimensions, thereby obtaining the channel weights and spatial pixel weights of the feature vector.

[0094] Specifically, Figure 5b To obtain the channel-weighted feature vector A schematic diagram, such as Figure 5b As shown, at the channel level, the compression module Squeeze is first used, employing the formula:

[0095]

[0096] The feature vector z of the second video data is compressed from H×W×C to 1×1×C dimension through global average pooling.

[0097] Then, through the excitation module, using the formula:

[0098] s c1 =F ex (z c1 )=σ(W2,δ(z c1 ,W1))

[0099] Get channel weights s c1 The activation module consists of two fully connected layers, a ReLU function δ, and a sigmoid function σ.

[0100] Using channel weights s c1 Using the formula:

[0101]

[0102] For the eigenvector z c1 Perform weighted processing to obtain the channel-weighted feature vector.

[0103] Specifically, Figure 5c To obtain the spatially weighted feature vector A schematic diagram, such as Figure 5c As shown, in the spatial dimension, firstly, the compression module Squeeze is used, employing the formula:

[0104]

[0105] z c2 =F sq (z)=max i∈C (z(i))

[0106] The dimension of z is compressed from HxWxC to HxWx1 by channel global average pooling and channel max pooling. The HxWx1 dimensional channel descriptions obtained by the two kinds of pooling are spliced into HxWx2 dimensional.

[0107] By the excitation module, the formula is used:

[0108] s c2 =F ex (z c2 )=σ(δ(z c2 ,W1))

[0109] The spatial pixel weight s c2 is obtained. The excitation module is constructed by a convolution layer and a sigmoid function σ.

[0110] The spatial pixel weight s c2 is used, and the formula is used:

[0111]

[0112] The feature vector z c2 is weighted to obtain the spatially weighted feature vector

[0113] Step S504: The channel-weighted feature vector and the spatially weighted feature vector are multiplied to obtain a weighted feature vector, and similarity calculation is performed with the standard feature vector to obtain a similarity weight of the weighted feature vector.

[0114] In the embodiment, the processing module can further include a dynamic prototype learning submodule, which multiplies the channel-weighted feature vector and the spatially weighted feature vector to obtain a weighted feature vector

[0115] The self-attention submodule extracts the weighted feature vector into HxW 1x1xC dimensional query modules queryitem, and uses the formula:

[0116]

[0117]

[0118] The weight w i of each query module and the storage module is calculated to obtain a similarity weight of the weighted feature vector, where the d function is a cosine similarity function, and the exp function is an exponential function. For example, in order to avoid overfitting of the convolutional network, the similarity weight A ReLU operation is performed, i.e., values less than a threshold in w are zeroed out.

[0119] Step S505: The standard feature vector is weighted using the similarity weight to obtain a weighted standard feature vector.

[0120] In this embodiment, the formula is used:

[0121]

[0122] The standard feature vector M is weighted to obtain a weighted standard feature vector

[0123] Step S506: The weighted standard feature vector is video reconstructed, and the reconstructed video data is compared with the standard video data to obtain a loss value corresponding to the second video data.

[0124] In this embodiment, the processing module can further include a decoder submodule that video reconstructs the weighted standard feature vector compares the reconstructed video data with the standard video data to obtain a loss value corresponding to the second video data.

[0125] In this embodiment, the standard video data is trained using a deep learning network to obtain a standard feature vector, the second video data is feature extracted, weighted, and similarity calculated with the standard feature vector using a multi-layer convolutional neural network, the video of the weighted standard feature vector is reconstructed, the reconstructed video data is compared with the standard video data to obtain a loss value corresponding to the second video data, and it is determined whether an intrusion fault occurs at a potential intrusion point, effectively improving the accuracy of intrusion detection at a platform screen door, and further solving the problem that the prior art cannot accurately detect whether an intrusion fault occurs at a platform screen door.

[0126] The following is an apparatus embodiment of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the apparatus embodiments of the present application, please refer to the method embodiments of the present application.

[0127] Figure 6 A structural schematic diagram of a detection device embodiment provided by the present application is shown in FIG. 1. Figure 6As shown, the detection device 60 comprises a line laser scanner 61, a video acquisition module 62, and a processing module 63. The line laser scanner 61 is configured to acquire a plurality of first point cloud coordinates of a platform tunnel region when a vehicle runs in the platform tunnel region at a normal driving speed. The video acquisition module 62 is configured to acquire a plurality of first video data of the platform tunnel region when the vehicle runs in the platform tunnel region at the normal driving speed. The processing module 63 is configured to compare the plurality of first point cloud coordinates with preconfigured standard point cloud coordinates, to determine a potential intrusion point in the platform tunnel region. The processing module 63 is further configured to acquire second video data corresponding to the potential intrusion point from the plurality of first video data, and compare the second video data with preconfigured standard video data, to obtain a loss value corresponding to the second video data. The processing module 63 is further configured to determine that an intrusion fault occurs at the potential intrusion point when it is determined that the loss value is greater than a loss threshold.

[0128] The detection device provided by the embodiments of the present application can execute the technical solutions shown in the above method embodiments, and has similar implementation principles and beneficial effects, which will not be described here in detail.

[0129] In a possible implementation, the processing module 63 is further configured to trigger the vehicle to run in the platform tunnel region at least once at a test speed, and acquire a plurality of second point cloud coordinates of the platform tunnel region as the preconfigured standard point cloud coordinates. The test speed is less than the normal driving speed.

[0130] In a possible implementation, the processing module 63 is further configured to classify the standard point cloud coordinates according to mileposts, to obtain a first subset of a plurality of standard point cloud coordinates corresponding to each milepost in the platform tunnel region. The processing module 63 is specifically configured to classify a plurality of first point cloud coordinates according to mileposts, to obtain a second subset of a plurality of first point cloud coordinates corresponding to each milepost in the platform tunnel region. For each first point cloud coordinate in the second subset corresponding to each milepost in the platform tunnel region, at least one standard point cloud coordinate within a preset distance from the first point cloud coordinate is obtained from the first subset corresponding to the milepost, and an average distance between the first point cloud coordinate and the at least one standard point cloud coordinate is obtained. When it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in a platform screen door region in the platform tunnel region, it is determined that a point corresponding to the first point cloud coordinate is a potential intrusion point.

[0131] The detection device provided by the embodiments of the present application can execute the technical solutions shown in the above method embodiments, and has similar implementation principles and beneficial effects, which will not be described here in detail.

[0132] In a possible implementation, the processing module 63 is further configured to classify the standard point cloud coordinates according to the kilometer markers to obtain a first subset of standard point cloud coordinates corresponding to each kilometer marker in the platform tunnel region; and the processing module 63 is specifically configured to classify the first point cloud coordinates according to the kilometer markers to obtain a second subset of first point cloud coordinates corresponding to each kilometer marker in the platform tunnel region; and for each first point cloud coordinate in the second subset {V k，s} corresponding to each kilometer marker in the platform tunnel region, the processing module 63 is configured to traverse the first subset obtain a plurality of standard point cloud coordinates with the smallest distances, and obtain an average distance between the first point cloud coordinate and the plurality of standard point cloud coordinates; and when it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in the platform tunnel region, the processing module 63 is configured to determine that a point corresponding to the first point cloud coordinate is a potential intrusion point; wherein V k，s is the first point cloud coordinate, is the standard point cloud coordinate, m is the serial number of the standard point cloud coordinate, s is the kilometer marker, and k is the serial number of the first point cloud coordinate.

[0133] The detection device provided in the embodiments of the present application can execute the technical solutions shown in the method embodiments, and the implementation principles and beneficial effects are similar, which will not be repeated here.

[0134] In a possible implementation, the processing module 63 is further configured to trigger the vehicle to run in the platform tunnel region at least once at a test speed, and use the obtained third video data of the platform tunnel region as the preconfigured standard video data; wherein the test speed is less than the normal driving speed.

[0135] The detection device provided in the embodiments of the present application can execute the technical solutions shown in the method embodiments, and the implementation principles and beneficial effects are similar, which will not be repeated here.

[0136] Figure 7 is a structural schematic diagram of a processing module of a detection device provided in the present application; the processing module 70 is further configured to train the standard video data by using a deep learning network to obtain a standard feature vector. As Figure 7As shown, the processing module 70 includes an encoder submodule 71, a self-attention submodule 72, a dynamic prototype learning submodule 73, and a decoder submodule 74. The encoder submodule 71 is configured to extract features of the second video data by using a multi-layer convolutional neural network to obtain a feature vector of the second video data. The self-attention submodule 72 is configured to obtain channel weights and spatial pixel weights of the feature vector, respectively, and perform weighting processing on the feature vector by using the channel weights and the spatial pixel weights, respectively, to obtain a channel-weighted feature vector and a spatial-weighted feature vector. The dynamic prototype learning submodule 73 is configured to multiply the channel-weighted feature vector and the spatial-weighted feature vector to obtain a weighted feature vector, and perform similarity calculation on the weighted feature vector and the standard feature vector to obtain a similarity weight of the weighted feature vector. The dynamic prototype learning submodule 73 is further configured to perform weighting processing on the standard feature vector by using the similarity weight to obtain a weighted standard feature vector. The decoder submodule 74 is configured to perform video reconstruction on the weighted standard feature vector, and compare the reconstructed video data with the standard video data to obtain a loss value corresponding to the second video data.

[0137] The embodiment of the present application also provides a subway detection vehicle, which comprises the detection device provided by any one of the foregoing embodiments and a vehicle body.

[0138] Those skilled in the art can understand that all or part of the steps of the foregoing method embodiments can be completed by program instruction-related hardware. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, the steps of the foregoing method embodiments are executed; and the foregoing storage medium includes various program code storage media such as ROM, RAM, magnetic discs, or optical discs.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or equivalently replace some or all of the technical features thereof; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method of intrusion detection, characterized by, The method comprises the following steps: acquiring a plurality of first point cloud coordinates and a plurality of first video data of a platform tunnel area when a vehicle runs at a normal driving speed in the platform tunnel area; comparing the plurality of first point cloud coordinates with preconfigured standard point cloud coordinates to determine potential intrusion points in the platform tunnel area; from the plurality of first video data, acquiring second video data corresponding to the potential intrusion points, and comparing the second video data with preconfigured standard video data to obtain a loss value corresponding to the second video data; when it is determined that the loss value is greater than a loss threshold, it is determined that an intrusion fault has occurred at the potential intrusion point; The method further comprises: training the standard video data using a deep learning network to obtain a standard feature vector; The comparison of the second video data with the preconfigured standard video data to obtain a loss value corresponding to the second video data comprises: using a multi-layer convolutional neural network to extract features of the second video data to obtain a feature vector of the second video data; respectively acquiring channel weights and spatial pixel weights of the feature vector, and respectively weighting the feature vector using the channel weights and the spatial pixel weights to obtain a channel-weighted feature vector and a spatial-weighted feature vector; multiplying the channel-weighted feature vector and the spatial-weighted feature vector to obtain a weighted feature vector, and performing similarity calculation on the weighted feature vector and the standard feature vector to obtain a similarity weight of the weighted feature vector; weighting the standard feature vector using the similarity weight to obtain a weighted standard feature vector; performing video reconstruction on the weighted standard feature vector, and comparing the reconstructed video data with the standard video data to obtain a loss value corresponding to the second video data; The method further comprises: According to the kilometer marker, the standard point cloud coordinates are classified to obtain a first subset of a plurality of standard point cloud coordinates corresponding to each kilometer marker in the platform tunnel area; The comparison of the plurality of first point cloud coordinates with the preconfigured standard point cloud coordinates to determine the potential intrusion points in the platform tunnel area comprises: According to the kilometer marker, the plurality of first point cloud coordinates are classified to obtain a second subset of a plurality of first point cloud coordinates corresponding to each kilometer marker in the platform tunnel area; For each first point cloud coordinate in the second subset corresponding to each kilometer marker in the platform tunnel area, at least one standard point cloud coordinate within a preset distance from the first point cloud coordinate is obtained from the first subset corresponding to the kilometer marker, and an average distance between the first point cloud coordinate and the at least one standard point cloud coordinate is obtained; When it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in a platform screen door area in the platform tunnel area, it is determined that the point corresponding to the first point cloud coordinate is a potential intrusion point.

2. The intrusion detection method of claim 1, wherein, The preconfigured standard point cloud coordinates are obtained in the following manner: trigger the vehicle to run in the station tunnel area at a test speed at least once, and acquire a plurality of second point cloud coordinates of the station tunnel area as the preconfigured standard point cloud coordinates respectively; wherein the test speed is less than the normal driving speed.

3. The intrusion detection method of claim 2, wherein, Also includes: According to the kilometer mark, the standard point cloud coordinates are classified and processed to obtain a first subset formed by a plurality of standard point cloud coordinates corresponding to each kilometer mark in the station tunnel area; Then the comparison of the plurality of first point cloud coordinates with the preconfigured standard point cloud coordinates to determine the potential intrusion points in the station tunnel area includes: According to the kilometer mark, the plurality of first point cloud coordinates are classified and processed to obtain a second subset formed by a plurality of first point cloud coordinates corresponding to each kilometer mark in the station tunnel area; for each kilometer marker corresponding to the second subset in the station tunnel area , traverse the first subset , obtain the minimum standard point cloud coordinates, and obtain the average distance between the first point cloud coordinates and the standard point cloud coordinates; When it is determined that the average distance is greater than the preset distance threshold and the first point cloud coordinate is located in the screen door area of the station tunnel area, it is determined that the point corresponding to the first point cloud coordinate is a potential intrusion point; wherein, is the first point cloud coordinate, is the standard point cloud coordinate, m is the serial number of the standard point cloud coordinate, s is the kilometer marker, and k is the serial number of the first point cloud coordinate.

4. The intrusion detection method according to any one of claims 1 to 3, characterized in that, The acquisition method of the preconfigured standard video data is: trigger the vehicle to run in the station tunnel area at a test speed at least once, and acquire a plurality of third video data of the station tunnel area as the preconfigured standard video data; wherein the test speed is less than the normal driving speed.

5. A detection device, characterized in that includes: a line laser scanner, configured to acquire a plurality of first point cloud coordinates of the station tunnel area when the vehicle runs in the station tunnel area at a normal driving speed; a video acquisition module, configured to acquire a plurality of first video data of the station tunnel area when the vehicle runs in the station tunnel area at a normal driving speed; a processing module, configured to compare the plurality of first point cloud coordinates with preconfigured standard point cloud coordinates to determine potential intrusion points in the station tunnel area; The processing module is further configured to acquire second video data corresponding to the potential intrusion point from the plurality of first video data, and compare the second video data with preconfigured standard video data to obtain a loss value corresponding to the second video data; The processing module is further configured to determine that an intrusion fault has occurred at the potential intrusion point when it is determined that the loss value is greater than a loss threshold; The processing module is further configured to: train the standard video data using a deep learning network to obtain a standard feature vector; Then the processing module includes: an encoder submodule, configured to extract features of the second video data using a multi-layer convolutional neural network to obtain a feature vector of the second video data; a self-attention submodule, configured to acquire channel weights and spatial pixel weights of the feature vector respectively, and perform weighted processing on the feature vector using the channel weights and the spatial pixel weights respectively to obtain a channel-weighted feature vector and a spatial-weighted feature vector; The dynamic prototype learning submodule is configured to multiply the channel-weighted feature vector and the space-weighted feature vector, obtain a weighted feature vector, and perform similarity calculation on the weighted feature vector and the standard feature vector to obtain a similarity weight of the weighted feature vector; The dynamic prototype learning submodule is further configured to perform weighting processing on the standard feature vector by using the similarity weight to obtain a weighted standard feature vector; The decoder submodule is configured to perform video reconstruction on the weighted standard feature vector, and compare the reconstructed video data with the standard video data to obtain a loss value corresponding to the second video data. The processing module is further configured to: According to the kilometer marker, the standard point cloud coordinates are classified to obtain a first subset formed by a plurality of standard point cloud coordinates corresponding to each kilometer marker in the station tunnel region. The processing module is further configured to: According to the kilometer marker, the plurality of first point cloud coordinates are classified to obtain a second subset formed by a plurality of first point cloud coordinates corresponding to each kilometer marker in the station tunnel region. For each first point cloud coordinate in the second subset corresponding to each kilometer marker in the station tunnel region, at least one standard point cloud coordinate with a distance less than a preset distance from the first point cloud coordinate is obtained from the first subset corresponding to the kilometer marker, and an average distance between the first point cloud coordinate and the at least one standard point cloud coordinate is obtained. When it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in the screen door region in the station tunnel region, it is determined that a point corresponding to the first point cloud coordinate is a potential intrusion point.

6. The detection device of claim 5, wherein, The processing module is further configured to: Trigger the vehicle to run in the station tunnel region at least once at a test speed, and obtain a plurality of second point cloud coordinates of the station tunnel region as the preconfigured standard point cloud coordinates; the test speed is less than the normal driving speed.

7. The detection device of claim 6, wherein, The processing module is further configured to: According to the kilometer marker, the standard point cloud coordinates are classified to obtain a first subset formed by a plurality of standard point cloud coordinates corresponding to each kilometer marker in the station tunnel region. The processing module is further configured to: According to the kilometer marker, the plurality of first point cloud coordinates are classified to obtain a second subset formed by a plurality of first point cloud coordinates corresponding to each kilometer marker in the station tunnel region. for each kilometer marker corresponding to the second subset in the station tunnel area , traverse the first subset , obtain a plurality of standard point cloud coordinates with the smallest value, and obtain the average distance between the first point cloud coordinate and the plurality of standard point cloud coordinates; When it is determined that the average distance is greater than a preset distance threshold and the first point cloud coordinate is located in the screen door region in the station tunnel region, it is determined that a point corresponding to the first point cloud coordinate is a potential intrusion point. wherein, is the first point cloud coordinate, is the standard point cloud coordinate, m is the serial number of the standard point cloud coordinate, s is the kilometer marker, and k is the serial number of the first point cloud coordinate.

8. The detection device according to any one of claims 5 to 7, characterized in that, The processing module is further configured to: Trigger the vehicle to run in the station tunnel region at least once at a test speed, and obtain a plurality of third video data of the station tunnel region as the preconfigured standard video data; the test speed is less than the normal driving speed.

9. A subway inspection car characterized by, The detection device and the vehicle body according to any one of claims 5 to 8. ​

Images