Railway track detection method, system and railway track detection vehicle

Abstract

The present invention belongs to the technical field of railway track detection, and discloses a railway track detection method, system and railway track detection vehicle, wherein the method comprises the following steps: constructing a railway track defect detection model and a railway track digital world three-dimensional model of a railway network; collecting railway track image data and corresponding railway track position data of the railway track in real time, and inputting the railway track image data of the railway track into the defect detection model for defect detection; rendering the railway track digital world three-dimensional model according to the railway track position data of the current railway track; obtaining the railway track digital world three-dimensional model with the defect target highlighted according to the railway track image data corresponding to the defect detection result. The present invention solves the problems of large human cost investment, low safety, low automation, low efficiency, weak scene connection, difficult data transmission and low practicality in the existing railway inspection technology.

Description

Technical Field

[0001] The present invention belongs to the technical field of railway track detection, and in particular relates to a railway track detection method, system and railway track detection vehicle. Background Art

[0002] Railway transportation occupies a very important position in my country's transportation industry. Due to its advantages such as fast transportation speed, large transportation capacity and less impact from environmental factors, in addition to transporting goods, railways are also one of the preferred means of transportation for people to travel.

[0003] Under the influence of the natural environment and rolling stock over the years, railway lines will experience various problems such as changes in track geometry, track surface abrasions and internal damage, roadbed deformation, and fastener loss or displacement.

[0004] Track defects not only affect vehicle operations, but in serious cases can also endanger driving safety.

[0005] Track inspection is one of the important means to detect defects in a timely manner and ensure the safety and reliability of railway transportation.

[0006] At present, my country's railway engineering departments mainly use manual inspections and vehicle inspections to conduct line inspections.

[0007] The prior art has the following problems:

[0008] The manual inspection method has high labor costs, heavy workload and long time for inspectors. Some defects and faults are missed due to subjective judgment, and safety accidents are prone to occur in areas with complex environments, resulting in low safety.

[0009] Most of the existing railway track inspection methods rely on railway track inspection vehicles to collect on-site data of railway tracks in real time, and experts analyze the on-site data to obtain the defects or faults of the railway tracks. The methods have low automation and efficiency, and the obtained defects or faults are weakly related to the on-site data and the actual scenes of the railway tracks, and cannot provide guidance for railway track maintenance work based on the defects or faults.

[0010] The railway track inspection vehicle in the prior art has a simple structure. When the railway track inspection vehicle operates in a signal shielded area or in the wild, data transmission failure is easily caused by transmission signal loss, and the practicality is low. Summary of the invention

[0011] In order to solve the problems of high manpower cost investment, low safety, low degree of automation, low efficiency, weak scene connection, difficult data transmission and low practicality in the prior art of railway inspection, the purpose of the present invention is to provide a railway track detection method, system and railway track inspection vehicle.

[0012] The technical solution adopted by the present invention is:

[0013] A railway track detection method based on digital twin and image recognition comprises the following steps:

[0014] Constructing a railway track defect detection model and a railway track digital world 3D model of the railway network, and storing model metadata of the railway track digital world 3D model of the railway network;

[0015] Collecting railway track image data and corresponding railway track position data of the railway track in real time, and inputting the railway track image data of the railway track into a defect detection model for defect detection to obtain corresponding defect detection results;

[0016] If the defect detection result of the current railway track is that there are defects, then according to the railway track position data of the current railway track, the model metadata of the corresponding railway track digital world 3D model is called for rendering to obtain the rendered railway track digital world 3D model, otherwise, the defect detection is continued;

[0017] According to the defect detection result, which is the railway track image data corresponding to the defect, the rendered railway track digital world 3D model is corrected to obtain the railway track digital world 3D model with the defect target highlighted.

[0018] Furthermore, a railway track defect detection model is constructed, including the following steps:

[0019] Obtain historical railway track image datasets;

[0020] Based on the YOLOv6 algorithm, an initial defect detection model for railway tracks was constructed;

[0021] The historical railway track image dataset is input into the initial defect detection model for optimization training to obtain the optimal defect detection model for railway tracks.

[0022] Furthermore, the historical railway track image data includes historical rail surface image data, fastener image data, overhead wire suspension device component image data, rail waist nail image data, transponder image data and leaky cable fixture image data.

[0023] Furthermore, constructing a three-dimensional digital world model of railway tracks of the railway network and storing model metadata of the three-dimensional digital world model of railway tracks of the railway network includes the following steps:

[0024] Obtain railway track route drawings of the railway network, video data of each railway track in the railway network, ledger information, ledger models and geographic information system GIS information;

[0025] According to the railway track route drawings of the railway network, as well as the ledger information and ledger models of each railway track, the initial railway track digital world 3D model of the railway network is constructed;

[0026] According to the video data of each railway track, the model appearance of the initial railway track digital world 3D model is corrected to obtain a corrected railway track digital world 3D model of the railway network;

[0027] According to the GIS information of each railway track, the track route of the revised railway track digital world 3D model is corrected to obtain the final railway track digital world 3D model of the railway network;

[0028] According to the railway track location data in the GIS information of each railway track, the model metadata of different areas of the final railway track digital world 3D model are stored separately, and the railway track location data is used as an indicator label of the model metadata of the railway track digital world 3D model of each area.

[0029] Furthermore, real-time acquisition of railway track image data and corresponding railway track position data of the railway track, and inputting the railway track image data of the railway track into a defect detection model for defect detection to obtain corresponding defect detection results, includes the following steps:

[0030] The railway track video data of the railway track collected in real time is input into the defect detection model for frame interception to obtain continuous frame railway track image data;

[0031] Preprocessing the railway track image data of the continuous frames to obtain the preprocessed railway track image data of the continuous frames;

[0032] Performing grid division and prior frame setting on each frame of preprocessed railway track image data to obtain continuous frames of railway track image data with prior frames;

[0033] Obtaining the offsets of all the a priori frames and the corresponding preset prediction frames in the railway track image data with a priori frames of continuous frames, and obtaining the initial prediction frames corresponding to all the grids according to the offsets, to obtain the railway track image data with the initial prediction frames of continuous frames;

[0034] According to a preset intersection-over-union ratio and a preset confidence level, non-maximum suppression screening is performed on all initial prediction frames in the railway track image data with initial prediction frames of the continuous frames to obtain the railway track image data with final prediction frames of the continuous frames;

[0035] According to the position of the final prediction frame, target detection is performed on each frame of railway track image data with the final prediction frame, so as to obtain a number of detection targets of each frame of railway track image data with the final prediction frame;

[0036] The target tracking algorithm is used to perform target tracking and defect detection on several detection targets of the railway track image data with the final prediction box in consecutive frames to obtain the defect detection results corresponding to the same detection target.

[0037] Further, according to the defect detection result, the image data of the railway track corresponding to the defect is corrected to obtain the digital world 3D model of the railway track with the defect target highlighted, including the following steps:

[0038] According to the defect detection results, the detection targets corresponding to the defects are matched with the corresponding model components in the rendered three-dimensional model of the railway track in the digital world;

[0039] According to the railway track image data corresponding to the detection target, the corresponding defect target in the rendered railway track digital world three-dimensional model is corrected to obtain the railway track digital world three-dimensional model after the defect target is corrected;

[0040] The defect target is highlighted, and the corresponding defect detection result is added to the digital world 3D model of the railway track after the defect target is corrected, so as to obtain the digital world 3D model of the railway track with the defect target highlighted.

[0041] A railway track detection system based on digital twins and image recognition is used to implement a railway track detection method. The system includes a railway track detection center, a cloud data center, several network base stations and several railway track detection vehicles. The railway track detection center is connected to the cloud data center and the several network base stations respectively. Each network base station is arranged one-to-one at a key point of data transmission in the railway network, and each network base station is connected to several railway track detection vehicles in a corresponding area respectively. Each railway track detection vehicle is arranged one-to-one at a railway track of the railway network.

[0042] The railway track inspection center is used to construct a defect detection model of the railway track and a railway track digital world three-dimensional model of the railway network, send the model metadata of the railway track defect detection model and the model metadata of the railway track digital world three-dimensional model of the railway network to the cloud data center for storage, receive and store the railway track image data of the railway track collected in real time by the railway track inspection vehicle transmitted by the network base station and the corresponding railway track position data, input the railway track image data of the railway track into the defect detection model for defect detection, obtain the corresponding defect detection result, send the defect detection result to the cloud data center for storage, call the model metadata of the corresponding railway track digital world three-dimensional model stored in the cloud data center for rendering according to the current railway track position data of the railway track, obtain the rendered railway track digital world three-dimensional model, and use the railway track image data to correct the rendered railway track digital world three-dimensional model according to the current railway track defect detection result, so as to obtain the railway track digital world three-dimensional model with the defect target highlighted;

[0043] A cloud data center, used to store model metadata of a defect detection model of a railway track, model metadata of a railway track digital world three-dimensional model of a railway network, railway track image data of the railway track and corresponding railway track position data, and defect detection results obtained by performing defect detection with a defect detection model;

[0044] A network base station, used to receive the railway track image data and corresponding railway track position data of the railway track uploaded by all railway track inspection vehicles in the area, and transmit the railway track image data and corresponding railway track position data of the railway track to the railway track inspection center;

[0045] The railway track inspection vehicle runs on each railway track of the railway network, and is used to collect the railway track image data and the corresponding railway track position data of the railway track in real time, and upload the railway track image data and the corresponding railway track position data of the railway track to the network base station in the current area.

[0046] Further, the railway track inspection center includes a defect detection model construction unit, a railway track digital world 3D model construction unit, a local storage unit, a defect detection application unit, a wireless communication unit and a 3D model visualization unit, the local storage unit is respectively connected to the defect detection model construction unit, the railway track digital world 3D model construction unit, the defect detection application unit, the wireless communication unit and the 3D model visualization unit, and the wireless communication unit is respectively connected to the cloud data center and all network base stations;

[0047] A defect detection model building unit is used to extract a historical railway track image data set in a local storage unit, build a railway track defect detection model based on the historical railway track image data set, and store model metadata of the railway track defect detection model in the local storage unit and the cloud data center;

[0048] A railway track digital world 3D model construction unit is used to extract the video data, ledger information, ledger model and GIS information of each railway track in the railway network from the local storage unit, construct a railway track digital world 3D model of the railway network, and store the railway track digital world 3D model of the railway network in the cloud data center;

[0049] A defect detection application unit is used to extract model metadata of a defect detection model of a railway track and railway track image data of the railway track collected in real time in a local storage unit, build a defect detection model of the railway track, input the railway track image data of the railway track into the defect detection model to perform defect detection, obtain corresponding defect detection results, and store the defect detection results in the local storage unit and the cloud data center;

[0050] The three-dimensional model visualization unit is connected to extract the railway track position data of the railway track collected in real time in the local storage unit, match the indication tag of the corresponding area in the cloud data center according to the railway track position data, obtain the model metadata of the railway track digital world three-dimensional model of the corresponding area, render the model metadata of the railway track digital world three-dimensional model of the corresponding area, obtain the rendered railway track digital world three-dimensional model, extract the defect detection result in the local storage unit, according to the defect detection result, the railway track image data corresponding to the defect, correct the rendered railway track digital world three-dimensional model, and obtain the visual railway track digital world three-dimensional model with highlighted defect targets;

[0051] A wireless communication unit, used to establish a data transmission channel between the railway track detection center and the cloud data center and between the railway track detection center and several network base stations;

[0052] The local storage unit is used to store the model metadata and defect detection results of the defect detection model, and cache the railway track image data of the railway track collected in real time, the corresponding railway track position data and the model metadata of the railway track digital world three-dimensional model of the corresponding area.

[0053] A railway track inspection vehicle is applied to a railway track inspection system. The railway track inspection vehicle comprises a vehicle body, at least two first video acquisition devices, at least one second video acquisition device, a control module, a video processing module, an RDSS communication module, a LoRa communication module, a track geometry measurement module, a vehicle body vibration acceleration measurement module, and a GPS mileage automatic correction module;

[0054] At least two first video acquisition devices are symmetrically arranged on both sides of the bottom of the vehicle body, and each first video acquisition device is arranged at an angle facing the side of the railway track. At least one second video acquisition device is arranged at the top of the vehicle body, and the second video acquisition device is arranged at an angle facing the forward direction of the railway track. The control module, video processing module, RDSS communication module, LoRa communication module, track geometry measurement module, vehicle body vibration acceleration measurement module and GPS mileage automatic correction module are all arranged inside the vehicle body, and the control module is respectively connected to the video processing module, RDSS communication module, LoRa communication module, track geometry measurement module, vehicle body vibration acceleration measurement module and GPS mileage automatic correction module, the RDSS communication module is respectively connected to the external positioning satellite and the network base station in the current area, the LoRa communication module is connected to the network base station in the current area, and the video processing module is respectively connected to the at least two first video acquisition devices and at least one second video acquisition device.

[0055] Furthermore, the railway track detection vehicle further comprises a sensor group and an A / D conversion module, the A / D conversion module is respectively connected to the control module and the sensor group, and the sensor group is arranged inside the vehicle body;

[0056] The sensor group includes a temperature and humidity sensor, a meteorological sensor, and an air pressure and altitude sensor, all of which are connected to the A / D conversion module.

[0057] The beneficial effects of the present invention are:

[0058] The present invention provides a railway track detection method and system based on digital twin and image recognition, which use defect detection models to automatically detect tracks, contact networks and communication equipment along the railway. The method has a high degree of automation, avoids manual inspection, reduces labor cost investment and the workload of inspectors, and improves safety and detection efficiency. At the same time, the digital world three-dimensional model of the railway track that highlights the defect target strengthens the connection between the defect or fault result and the field data and the actual scene of the railway track, so that users can intuitively and conveniently observe the defect or fault location of the railway track, the actual scene conditions and other information, provide guiding opinions on the railway track maintenance work, and improve the practicability of the method and system, as well as the user experience.

[0059] The present invention provides a railway inspection vehicle, which collects railway track image data of the railway track by a first video acquisition device, and uses a second video acquisition device to collect image data of the contact network and communication equipment, thereby improving the efficiency and practicality of railway inspection, and adopts a GPS mileage automatic correction module to automatically correct the railway track position data collected by the RDSS communication module, thereby improving the accuracy of railway inspection, and the RDSS communication module and the LoRa communication module serve as backup for each other, thereby ensuring the data transmission channel with the network base station and improving the stability of data transmission.

[0060] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flowchart of the railway track detection method based on digital twin and image recognition.

[0062] Figure 2 It is a structural block diagram of the railway track detection system based on digital twin and image recognition.

[0063] Figure 3 It is the structural block diagram of the railway track inspection vehicle. DETAILED DESCRIPTION

[0064] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. Example 1

[0065] like Figure 1 As shown, this embodiment provides a railway track detection method based on digital twin and image recognition, comprising the following steps:

[0066] Constructing a railway track defect detection model and a railway track digital world 3D model of a railway network, and storing model metadata of the railway track digital world 3D model of the railway network, includes the following steps:

[0067] Obtaining historical railway track image data sets, including historical rail surface image data, fastener image data, overhead wire suspension device component image data, rail waist nail image data, transponder image data, and leaky cable fixture image data; the defect detection model not only performs defect detection on railway tracks, but also provides defect detection on overhead wire suspension devices and other overhead wire components, thereby improving the practicality and functionality of the method;

[0068] Based on the YOLOv6 algorithm, the initial defect detection model of railway tracks is constructed; the YOLOv6 algorithm supports the full-chain industrial application requirements such as model training, reasoning and multi-platform deployment, and has made many improvements and optimizations in the algorithm level such as network structure and training strategy. YOLOv6 surpasses other algorithms of the same volume in terms of accuracy and speed. Therefore, in the field of railway track defect detection, it has better effects and applications, and can more accurately identify and detect rail surface defects, fastener failures, contact network suspension device component failures, rail waist nail failures, etc. in railway track image data;

[0069] Input the historical railway track image data set into the initial defect detection model for optimization training to obtain the optimal defect detection model for railway tracks;

[0070] Obtain the railway track route drawings of the railway network, video data of each railway track in the railway network, ledger information, ledger model and (Geographic Information System, GIS) information;

[0071] According to the railway track route drawings of the railway network, as well as the ledger information and ledger models of each railway track, the initial railway track digital world 3D model of the railway network is constructed;

[0072] According to the video data of each railway track, the model appearance of the initial railway track digital world 3D model is corrected to obtain a corrected railway track digital world 3D model of the railway network;

[0073] According to the GIS information of each railway track, the track route of the revised railway track digital world 3D model is corrected to obtain the final railway track digital world 3D model of the railway network;

[0074] The railway track digital world 3D model can intuitively and conveniently view the specific conditions of the railway track, and can directly display the railway track inspection results in the railway track digital world 3D model for visual management, which improves practicality and convenience;

[0075] According to the railway track location data in the GIS information of each railway track, the model metadata of different areas of the final railway track digital world 3D model are stored separately, and the railway track location data is used as an indicator label of the model metadata of the railway track digital world 3D model of each area; since the railway track digital world 3D model of the railway network is huge, the railway track digital world 3D model directly rendered as a whole needs to occupy a large amount of memory, so the railway track location data is used as an indicator label, and the railway track digital world 3D model is rendered and displayed in different areas according to the real-time railway track location data, which improves the convenience of observation and reduces the occupation of memory resources;

[0076] Collecting railway track image data and corresponding railway track position data of the railway track in real time, and inputting the railway track image data of the railway track into the defect detection model for defect detection, and obtaining corresponding defect detection results, including the following steps:

[0077] The railway track video data of the railway track collected in real time is input into the defect detection model for frame interception to obtain continuous frame railway track image data;

[0078] Preprocessing the railway track image data of the continuous frames to obtain the preprocessed railway track image data of the continuous frames;

[0079] Performing grid division and prior frame setting on each frame of preprocessed railway track image data to obtain continuous frames of railway track image data with prior frames;

[0080] Obtaining the offsets of all the a priori frames and the corresponding preset prediction frames in the railway track image data with a priori frames of continuous frames, and obtaining the initial prediction frames corresponding to all the grids according to the offsets, to obtain the railway track image data with the initial prediction frames of continuous frames;

[0081] According to a preset intersection-over-union ratio and a preset confidence level, non-maximum suppression screening is performed on all initial prediction frames in the railway track image data with initial prediction frames of the continuous frames to obtain the railway track image data with final prediction frames of the continuous frames;

[0082] According to the position of the final prediction frame, target detection is performed on each frame of railway track image data with the final prediction frame to obtain a number of detection targets of each frame of railway track image data with the final prediction frame;

[0083] Using a target tracking algorithm, target tracking and defect detection are performed on several detection targets of the railway track image data with the final prediction box in consecutive frames to obtain defect detection results corresponding to the same detection target;

[0084] If the defect detection result of the current railway track is that there are defects, then according to the railway track position data of the current railway track, the model metadata of the corresponding railway track digital world 3D model is called for rendering to obtain the rendered railway track digital world 3D model, otherwise, the defect detection is continued;

[0085] According to the defect detection result, the image data of the railway track corresponding to the defect is corrected to obtain the digital world 3D model of the railway track with the defect target highlighted, including the following steps:

[0086] According to the defect detection results, the detection targets corresponding to the defects are matched with the corresponding model components in the rendered three-dimensional model of the railway track in the digital world;

[0087] According to the railway track image data corresponding to the detection target, the corresponding defect target in the rendered railway track digital world three-dimensional model is corrected to obtain the railway track digital world three-dimensional model after the defect target is corrected;

[0088] The defect target is highlighted, and the corresponding defect detection result is added to the digital world 3D model of the railway track after the defect target is corrected, so as to obtain the digital world 3D model of the railway track with the defect target highlighted. Example 2

[0089] like Figure 2 As shown, based on Example 1, this embodiment provides a railway track detection system based on digital twins and image recognition, which is used to implement a railway track detection method. The system includes a railway track detection center, a cloud data center, several network base stations and several railway track detection vehicles. The railway track detection center is connected to the cloud data center and the several network base stations respectively. Each network base station is arranged one-to-one at a key point of data transmission in the railway network, and each network base station is connected to several railway track detection vehicles in a corresponding area respectively. Each railway track detection vehicle is arranged one-to-one at a railway track of the railway network.

[0090] The railway track inspection center is used to construct a defect detection model of the railway track and a railway track digital world three-dimensional model of the railway network, send the model metadata of the railway track defect detection model and the model metadata of the railway track digital world three-dimensional model of the railway network to the cloud data center for storage, receive and store the railway track image data of the railway track collected in real time by the railway track inspection vehicle transmitted by the network base station and the corresponding railway track position data, input the railway track image data of the railway track into the defect detection model for defect detection, obtain the corresponding defect detection result, send the defect detection result to the cloud data center for storage, call the model metadata of the corresponding railway track digital world three-dimensional model stored in the cloud data center for rendering according to the current railway track position data of the railway track, obtain the rendered railway track digital world three-dimensional model, and use the railway track image data to correct the rendered railway track digital world three-dimensional model according to the current railway track defect detection result, so as to obtain the railway track digital world three-dimensional model with the defect target highlighted;

[0091] A cloud data center, used to store model metadata of a defect detection model of a railway track, model metadata of a railway track digital world three-dimensional model of a railway network, railway track image data of the railway track and corresponding railway track position data, and defect detection results obtained by performing defect detection with a defect detection model;

[0092] A network base station, used to receive the railway track image data and corresponding railway track position data of the railway track uploaded by all railway track inspection vehicles in the area, and transmit the railway track image data and corresponding railway track position data of the railway track to the railway track inspection center;

[0093] The railway track inspection vehicle runs on each railway track of the railway network, and is used to collect the railway track image data and the corresponding railway track position data of the railway track in real time, and upload the railway track image data and the corresponding railway track position data of the railway track to the network base station in the current area.

[0094] Preferably, the railway track inspection center includes a defect detection model construction unit, a railway track digital world 3D model construction unit, a local storage unit, a defect detection application unit, a wireless communication unit and a 3D model visualization unit, the local storage unit is respectively connected to the defect detection model construction unit, the railway track digital world 3D model construction unit, the defect detection application unit, the wireless communication unit and the 3D model visualization unit, and the wireless communication unit is respectively connected to the cloud data center and all network base stations;

[0095] A defect detection model building unit is used to extract a historical railway track image data set in a local storage unit, build a railway track defect detection model based on the historical railway track image data set, and store model metadata of the railway track defect detection model in the local storage unit and the cloud data center;

[0096] A railway track digital world 3D model construction unit is used to extract the video data, ledger information, ledger model and GIS information of each railway track in the railway network from the local storage unit, construct a railway track digital world 3D model of the railway network, and store the railway track digital world 3D model of the railway network in the cloud data center;

[0097] A defect detection application unit is used to extract model metadata of a defect detection model of a railway track and railway track image data of the railway track collected in real time in a local storage unit, build a defect detection model of the railway track, input the railway track image data of the railway track into the defect detection model to perform defect detection, obtain corresponding defect detection results, and store the defect detection results in the local storage unit and the cloud data center;

[0098] The three-dimensional model visualization unit is connected to extract the railway track position data of the railway track collected in real time in the local storage unit, match the indication tag of the corresponding area in the cloud data center according to the railway track position data, obtain the model metadata of the railway track digital world three-dimensional model of the corresponding area, render the model metadata of the railway track digital world three-dimensional model of the corresponding area, obtain the rendered railway track digital world three-dimensional model, extract the defect detection result in the local storage unit, according to the defect detection result, the railway track image data corresponding to the defect, correct the rendered railway track digital world three-dimensional model, and obtain the visual railway track digital world three-dimensional model with highlighted defect targets;

[0099] A wireless communication unit, used to establish a data transmission channel between the railway track detection center and the cloud data center and between the railway track detection center and several network base stations;

[0100] The local storage unit is used to store the model metadata and defect detection results of the defect detection model, and cache the railway track image data of the railway track collected in real time, the corresponding railway track position data and the model metadata of the railway track digital world three-dimensional model of the corresponding area.

[0101] The present invention provides a railway track detection method and system based on digital twin and image recognition, which use defect detection models to automatically detect tracks, contact networks and communication equipment along the railway. The method has a high degree of automation, avoids manual inspection, reduces labor cost investment and the workload of inspectors, and improves safety and detection efficiency. At the same time, the digital world three-dimensional model of the railway track that highlights the defect target strengthens the connection between the defect or fault result and the field data and the actual scene of the railway track, so that users can intuitively and conveniently observe the defect or fault location of the railway track, the actual scene conditions and other information, provide guiding opinions on the railway track maintenance work, and improve the practicability of the method and system, as well as the user experience. Example 3

[0102] like Figure 3As shown, based on Example 2, this embodiment provides a railway track detection vehicle, which is applied to a railway track detection system. The railway track detection vehicle includes a vehicle body, at least two first video acquisition devices, at least one second video acquisition device, a control module, a video processing module, a satellite radio positioning system (Radio DeterminationSatel-lite System, RDSS) communication module, a long-range radio (Long Range Radio, LoRa) communication module, a track geometry measurement module, a vehicle body vibration acceleration measurement module and a global positioning system (Global PositioningSystem, GPS) mileage automatic correction module;

[0103] At least two first video acquisition devices are symmetrically arranged on both sides of the bottom of the vehicle body, and each first video acquisition device is arranged at an angle facing the side of the railway track. At least one second video acquisition device is arranged at the top of the vehicle body, and the second video acquisition device is arranged at an angle facing the forward direction of the railway track. The control module, the video processing module, the RDSS communication module, the LoRa communication module, the track geometry measurement module, the vehicle body vibration acceleration measurement module and the GPS mileage automatic correction module are all arranged inside the vehicle body, and the control module is respectively connected to the video processing module, the RDSS communication module, the LoRa communication module, the track geometry measurement module, the vehicle body vibration acceleration measurement module and the GPS mileage automatic correction module, and the RDSS communication module is respectively connected to the external positioning satellite and the network base station in the current area. The LoRa communication module is connected to the network base station in the current area, and the video processing module is respectively connected to at least two first video acquisition devices and at least one second video acquisition device; the RDSS communication module accurately obtains the railway track position data through the Beidou satellite, and also has the Beidou short message communication function, which can transmit images, videos and other data in the form of short messages. The track geometry measurement unit collects the track geometry data of the track gauge, track direction, height, level, triangular pits, superelevation and curvature. The car body vibration acceleration measurement unit collects the vertical acceleration and horizontal acceleration of the car body. The GPS mileage automatic correction unit automatically corrects the track inspection vehicle mileage, eliminates the accumulated error of the measured mileage, and facilitates each section to accurately locate and inspect defective places, find and rectify track diseases, ensure railway driving safety and improve track maintenance quality.

[0104] Preferably, the railway track detection vehicle further comprises a sensor group and an A / D conversion module, the A / D conversion module is connected to the control module and the sensor group respectively, and the sensor group is arranged inside the vehicle body;

[0105] The sensor group includes a temperature and humidity sensor, a meteorological sensor, and an air pressure and altitude sensor, all of which are connected to the A / D conversion module. The collected temperature and humidity information, meteorological information, and air pressure and altitude information can assist in the collection of railway track position data.

[0106] The present invention provides a railway inspection vehicle, which collects railway track image data of the railway track by a first video acquisition device, and uses a second video acquisition device to collect image data of the contact network and communication equipment, thereby improving the efficiency and practicality of railway inspection, and adopts a GPS mileage automatic correction module to automatically correct the railway track position data collected by the RDSS communication module, thereby improving the accuracy of railway inspection, and the RDSS communication module and the LoRa communication module serve as backup for each other, thereby ensuring the data transmission channel with the network base station and improving the stability of data transmission.

[0107] The present invention is not limited to the above optional implementation modes, and anyone can derive other various forms of products under the enlightenment of the present invention.

[0108] The above specific implementation methods should not be understood as limiting the scope of protection of the present invention. The scope of protection of the present invention should be based on the definition in the claims, and the description can be used to interpret the claims.

Claims

1. A railway track detection method based on digital twin and image recognition, characterized in that: The steps include: Constructing a railway track defect detection model and a railway track digital world 3D model of a railway network, and storing model metadata of the railway track digital world 3D model of the railway network, includes the following steps: Obtain railway track route drawings of the railway network, video data of each railway track in the railway network, ledger information, ledger models and GIS information; According to the railway track route drawings of the railway network, as well as the ledger information and ledger models of each railway track, the initial railway track digital world 3D model of the railway network is constructed; According to the video data of each railway track, the model appearance of the initial railway track digital world 3D model is corrected to obtain a corrected railway track digital world 3D model of the railway network; According to the GIS information of each railway track, the track route of the revised railway track digital world 3D model is corrected to obtain the final railway track digital world 3D model of the railway network; According to the railway track location data in the GIS information of each railway track, the model metadata of different areas of the final railway track digital world 3D model are stored separately, and the railway track location data is used as an indication label of the model metadata of the railway track digital world 3D model of each area; Collecting railway track image data and corresponding railway track position data of the railway track in real time, and inputting the railway track image data of the railway track into the defect detection model for defect detection, and obtaining corresponding defect targets and defect detection results, including the following steps: The railway track video data of the railway track collected in real time is input into the defect detection model for frame interception to obtain continuous frame railway track image data; Preprocessing the railway track image data of the continuous frames to obtain the preprocessed railway track image data of the continuous frames; Performing grid division and prior frame setting on each frame of preprocessed railway track image data to obtain continuous frames of railway track image data with prior frames; Obtaining the offsets of all the a priori frames and the corresponding preset prediction frames in the railway track image data with a priori frames of continuous frames, and obtaining the initial prediction frames corresponding to all the grids according to the offsets, to obtain the railway track image data with the initial prediction frames of continuous frames; According to a preset intersection-over-union ratio and a preset confidence level, non-maximum suppression screening is performed on all initial prediction frames in the railway track image data with initial prediction frames of the continuous frames to obtain the railway track image data with final prediction frames of the continuous frames; According to the position of the final prediction frame, target detection is performed on each frame of railway track image data with the final prediction frame, so as to obtain a number of detection targets of each frame of railway track image data with the final prediction frame; Using a target tracking algorithm, target tracking and defect detection are performed on several detection targets of the railway track image data with the final prediction box in consecutive frames to obtain the same defect target and defect detection results; If the defect detection result of the current railway track is that there are defects, then according to the railway track position data of the current railway track, the model metadata of the corresponding railway track digital world 3D model is called for rendering to obtain the rendered railway track digital world 3D model, otherwise, the defect detection is continued; According to the defect detection result, the image data of the railway track corresponding to the defect is corrected to obtain the digital world 3D model of the railway track with the defect target highlighted, including the following steps: According to the defect detection result, the defect target corresponding to the defect is matched with the corresponding model component in the rendered three-dimensional model of the railway track digital world, so as to obtain the defect target of the rendered three-dimensional model of the railway track digital world; According to the railway track image data corresponding to the defect target of the rendered railway track digital world 3D model, the defect target is corrected to obtain the railway track digital world 3D model after the defect target is corrected; The defect target of the digital world 3D model of the railway track after the defect target correction is highlighted, and the corresponding defect detection result is added to the digital world 3D model of the railway track after the defect target correction, so as to obtain the digital world 3D model of the railway track with the defect target highlighted.

2. The railway track detection method according to claim 1, characterized in that: Building a defect detection model for railway tracks includes the following steps: Obtain historical railway track image datasets; Based on the YOLOv6 algorithm, an initial defect detection model for railway tracks was constructed; The historical railway track image dataset is input into the initial defect detection model for optimization training to obtain the optimal defect detection model for railway tracks.

3. The railway track detection method according to claim 2, characterized in that: The historical railway track image data includes historical rail surface image data, fastener image data, overhead wire suspension device component image data, rail waist nail image data, transponder image data and leaky cable fixture image data.

4. A railway track detection system based on digital twin and image recognition, used to implement the railway track detection method according to any one of claims 1 to 3, characterized in that: The system includes a railway track detection center, a cloud data center, a plurality of network base stations and a plurality of railway track detection vehicles. The railway track detection center is connected to the cloud data center and the plurality of network base stations respectively. Each of the network base stations is arranged one by one at a key point of data transmission in the railway network. Each of the network base stations is connected to a plurality of railway track detection vehicles in a corresponding area respectively. Each of the railway track detection vehicles is arranged one by one at a railway track in the railway network. The railway track inspection center is used to construct a defect detection model of the railway track and a railway track digital world three-dimensional model of the railway network, send the model metadata of the railway track defect detection model and the model metadata of the railway track digital world three-dimensional model of the railway network to the cloud data center for storage, receive and store the railway track image data of the railway track collected in real time by the railway track inspection vehicle transmitted by the network base station and the corresponding railway track position data, input the railway track image data of the railway track into the defect detection model for defect detection, obtain the corresponding defect target and defect detection result, send the defect detection result to the cloud data center for storage, call the model metadata of the corresponding railway track digital world three-dimensional model stored in the cloud data center for rendering according to the current railway track position data of the railway track, obtain the rendered railway track digital world three-dimensional model, and use the railway track image data to correct the rendered railway track digital world three-dimensional model according to the current railway track defect detection result, so as to obtain the railway track digital world three-dimensional model with highlighted defect targets; A cloud data center, used to store model metadata of a defect detection model of a railway track, model metadata of a railway track digital world three-dimensional model of a railway network, railway track image data of the railway track and corresponding railway track position data, and defect detection results obtained by performing defect detection with a defect detection model; A network base station, used to receive the railway track image data and corresponding railway track position data of the railway track uploaded by all railway track inspection vehicles in the area, and transmit the railway track image data and corresponding railway track position data of the railway track to the railway track inspection center; The railway track inspection vehicle runs on each railway track of the railway network, and is used to collect the railway track image data and the corresponding railway track position data of the railway track in real time, and upload the railway track image data and the corresponding railway track position data of the railway track to the network base station in the current area.

5. The railway track detection system according to claim 4, characterized in that: The railway track inspection center includes a defect detection model construction unit, a railway track digital world 3D model construction unit, a local storage unit, a defect detection application unit, a wireless communication unit and a 3D model visualization unit, wherein the local storage unit is respectively connected to the defect detection model construction unit, the railway track digital world 3D model construction unit, the defect detection application unit, the wireless communication unit and the 3D model visualization unit, and the wireless communication unit is respectively connected to the cloud data center and all network base stations; A defect detection model building unit is used to extract a historical railway track image data set in a local storage unit, build a railway track defect detection model based on the historical railway track image data set, and store model metadata of the railway track defect detection model in the local storage unit and the cloud data center; A railway track digital world 3D model construction unit is used to extract the video data, ledger information, ledger model and GIS information of each railway track in the railway network from the local storage unit, construct a railway track digital world 3D model of the railway network, and store the railway track digital world 3D model of the railway network in the cloud data center; A defect detection application unit is used to extract model metadata of a defect detection model of a railway track and railway track image data of the railway track collected in real time in a local storage unit, build a defect detection model of the railway track, input the railway track image data of the railway track into the defect detection model for defect detection, obtain corresponding defect targets and defect detection results, and store the defect targets and defect detection results in the local storage unit and the cloud data center; The three-dimensional model visualization unit is connected to extract the railway track position data of the railway track collected in real time in the local storage unit, match the indication tag of the corresponding area in the cloud data center according to the railway track position data, obtain the model metadata of the railway track digital world three-dimensional model of the corresponding area, render the model metadata of the railway track digital world three-dimensional model of the corresponding area, obtain the rendered railway track digital world three-dimensional model, extract the defect detection result in the local storage unit, according to the defect detection result, the railway track image data corresponding to the defect, correct the rendered railway track digital world three-dimensional model, and obtain the visual railway track digital world three-dimensional model with highlighted defect targets; A wireless communication unit, used to establish a data transmission channel between the railway track detection center and the cloud data center and between the railway track detection center and several network base stations; The local storage unit is used to store the model metadata and defect detection results of the defect detection model, and cache the railway track image data of the railway track collected in real time, the corresponding railway track position data and the model metadata of the railway track digital world three-dimensional model of the corresponding area.

6. A railway track inspection vehicle, applied to the railway track inspection system as claimed in claim 5, characterized in that: The railway track inspection vehicle comprises a vehicle body, at least two first video acquisition devices, at least one second video acquisition device, a control module, a video processing module, an RDSS communication module, a LoRa communication module, a track geometry measurement module, a vehicle body vibration acceleration measurement module and a GPS mileage automatic correction module; At least two of the first video acquisition devices are symmetrically arranged on both sides of the bottom of the vehicle body, and each first video acquisition device is arranged at an angle facing the side of the railway track. At least one of the second video acquisition devices is arranged at the top of the vehicle body, and the second video acquisition device is arranged at an angle facing the forward direction of the railway track. The control module, video processing module, RDSS communication module, LoRa communication module, track geometry measurement module, vehicle body vibration acceleration measurement module and GPS mileage automatic correction module are all arranged inside the vehicle body, and the control module is respectively connected to the video processing module, RDSS communication module, LoRa communication module, track geometry measurement module, vehicle body vibration acceleration measurement module and GPS mileage automatic correction module. The RDSS communication module is respectively connected to the external positioning satellite and the network base station in the current area, the LoRa communication module is connected to the network base station in the current area, and the video processing module is respectively connected to the at least two first video acquisition devices and at least one second video acquisition device.

7. The railway track inspection vehicle according to claim 6, characterized in that: The railway track detection vehicle further comprises a sensor group and an A / D conversion module, wherein the A / D conversion module is connected to the control module and the sensor group respectively, and the sensor group is arranged inside the vehicle body; The sensor group includes a temperature and humidity sensor, a meteorological sensor, and an air pressure and altitude sensor, all of which are connected to the A / D conversion module.

Images