Dynamic identification and measurement method and system for flexible pantograph-catenary contact point of large iron

Abstract

The invention relates to a dynamic identification and measurement method and system for a flexible pantograph-catenary contact point of a large iron, and the method is realized by adopting an overall architecture of two processes: the first process is located at a vehicle-mounted end, and the pantograph-catenary contact point is identified by adopting an infrared camera to obtain an image, gray matching NCC and Hough straight line detection; and then real-time rapid analysis is carried out to store possible overrun problem data. And in the second process, the main control device located at the ground end carries out secondary analysis on the abnormal video to obtain final abnormal alarm information. Wherein the second process is realized by adopting target detection of deep learning, and an alexnet model framework is used for training to obtain model parameters of the pantograph and the contact point for forward reasoning. The method comprises the following two steps of: identifying a pantograph area, longitudinally expanding part of the area according to a result of the first step, and performing target detection on a cross-like intersection point in pantograph-catenary contact in the area. The method is good in real-time performance and high in anomaly detection accuracy. The practicability is high, and the popularization significance is high.

Description

Technical Field

[0001] The present invention relates to the field of rail transit technology, and in particular to a method and system for dynamically identifying and measuring contact points of a flexible bow-net on a large railway. Background Art

[0002] The pantograph and catenary are the physical components of the high-speed railway catenary, and are the mechanism by which the train obtains electrical energy from the contact wire. The catenary directly affects the safe operation of high-speed electrified railways, and its good service performance is the basic condition for ensuring the reliable and safe operation of high-speed railways; and the identification of the geometric parameters of the pantograph is an important part of judging the safety performance of the catenary.

[0003] At present, the commonly used technical method for pantograph detection is to combine laser ranging and angular displacement sensor devices to obtain measurement data and perform data preprocessing to reduce interference from external factors, and finally obtain the conduction height value and pull-out value through conversion equations. However, laser scanning mainly detects and transmits data through signals. During use, it will be affected by the low electromagnetic waves brought by the train operation, resulting in errors in some data, which will affect the results of laser scanning and the accuracy is poor. Summary of the invention

[0004] Based on this, it is necessary to provide a method and system for dynamically identifying and measuring the contact points of a large iron flexible bow-net in response to the deficiencies in the prior art.

[0005] A method for dynamically identifying and measuring contact points of a large iron flexible bow-net includes the following steps:

[0006] Step S0, calibrate the infrared camera on the pantograph lifting plane to obtain a calibration file; obtain a mapping matrix consisting of three parts: rotation, non-uniform scaling and translation through the mapping relationship between the collected actual points and the pixel points;

[0007]

[0008] Among them, R: rotation matrix, T: translation vector, Sx, Sy: scaling size;

[0009] Step S1: using an infrared camera to collect images of the coordination between the pantograph and the contact network;

[0010] Step S2: using NCC grayscale template matching to obtain the pantograph position;

[0011] Step S3: Reduce the ROI (target area) to the pantograph and the area above it in combination with the pantograph position, and then use Hough transform to extract the contact network;

[0012] Step S4: Calculate the contact point between the pantograph and the contact network, obtain the accurate position area of ​​the pantograph through the grayscale feature, extract the uppermost contour line of the pantograph and the contact line, solve the intersection point by simultaneously solving the two straight line equations, and obtain the contact point position coordinates;

[0013]

[0014] Step S5: filtering the result points according to the results retained by the previous multiple frames to eliminate abnormal points;

[0015] Step S6: Map the result points and transform them into real coordinates by matrix transformation, and obtain the current height guide value and pull-out value, as shown below;

[0016]

[0017] Among them, (Px, Py): image coordinate point; (Qx, Qy): result coordinate point; x: pull out, y: guide height;

[0018] Step S7: judging whether it is abnormal, if the above-obtained conductance value and pull-out value are abnormal values, saving the video information and transmitting it to the main control device;

[0019] Step S8: the main control device identifies abnormal videos;

[0020] Step S9: the main control device performs pantograph-catenary contact point identification and detection to obtain the contact point coordinates;

[0021] Step S10: Determine whether the contact point is abnormal or not according to the threshold setting.

[0022] Furthermore, the calibration file is in the lifting plane where the pantograph is located, with the center of the pantograph as the 0 point, and a ruler is used to collect plane points for infrared camera calibration.

[0023] Furthermore, in step S5, prediction and filtering are performed according to the distribution characteristics of the installation position of the flexible contact network itself, the rules of the pull-out value and the height conduction value.

[0024] Furthermore, the implementation of the filtering algorithm uses the results of the previous multiple frames of data to form a real-time updated dynamic array, uses the least squares linear fitting method to obtain the regression equation, and judges whether the current frame result is within the movement trend range of the contact point to define the outliers. For outliers, the current trend prediction value is used to replace the outliers.

[0025] Furthermore, in steps S8 and S9, a deep learning method is used to identify and detect the pantograph.

[0026] A dynamic identification and measurement system for large railway flexible bow-net contact points comprises an on-board detection system and a ground analysis system, wherein the on-board detection system comprises an image acquisition module, an image processing module, a calculation and analysis module, a judgment module, a data storage module, and a data transmission module; the ground analysis system comprises a deep learning module, a data judgment module, and a database; the image acquisition module acquires an image, and the image processing module performs grayscale matching NCC and Hough line detection on the image to identify the bow-net contact points; the calculation and analysis module calculates the contact points between the pantograph and the contact network, and performs filtering and mapping transformation on the structures retained by multiple frames to obtain the current conduction height value and pull-out value; the judgment module compares and analyzes the analysis value with the standard value, and stores the data information in the data storage module; and the abnormal data information is sent to the ground analysis system through the data transmission module for secondary analysis; the ground analysis system stores the received information in the database, and performs identification and detection through the deep learning module, and the data judgment module compares and analyzes the data information after detection and issues an alarm for confirming the abnormality.

[0027] Furthermore, the image acquisition module is an infrared camera, and the infrared camera is arranged on the roof of the vehicle.

[0028] In summary, the identification and measurement method of the present invention is implemented by an overall architecture of two processes. The first process is located at the vehicle-mounted end, and an infrared camera is used to obtain images, grayscale matching NCC and Hough line detection are used to identify the contact points of the bow and the network, and then real-time rapid analysis is performed to save the possible over-limit problem data. The second process is located in the main control device at the ground end, and the abnormal video is analyzed again to obtain the final abnormal alarm information. Among them, the second process is implemented by deep learning target detection, and the model parameters of the pantograph and the contact point are trained and obtained respectively using the alexnet model framework for forward reasoning. It is carried out in two steps. The first step is to identify the pantograph area, and the second step is to perform target detection on the cross-like intersection of the bow and the network contact in the area after longitudinally expanding part of the area according to the results of the first step. This method combines the rapid response of conventional image algorithms to ensure its real-time performance. The combination of the two processes is used to combine the deep learning strategy in the main control device to perform secondary analysis to ensure the accuracy of abnormal detection. It is highly practical and has a strong promotion significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is one of the calibration file schematic diagrams in step S0 of a method for dynamically identifying and measuring contact points of a large iron flexible bow-net according to the present invention;

[0030] Figure 2 This is one of the schematic diagrams of the NCC matching results in step S2;

[0031] Figure 3 This is one of the schematic diagrams of the results after reducing the ROI in S3;

[0032] Figure 4 This is one of the schematic diagrams of Hough line extraction results in S3;

[0033] Figure 5 This is one of the graphs related to the algorithm processing results in S6;

[0034] Figure 6 This is one of the schematic diagrams of pantograph identification results in S8;

[0035] Figure 7 This is one of the schematic diagrams of the S9 contact point cross recognition results. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0037] like Figure 1 As shown, it is a method for dynamic identification and measurement of large iron flexible bow-net contact points provided by the present invention, which specifically includes the following steps:

[0038] Step S0: Calibrate the infrared camera on the pantograph lifting plane to obtain a calibration file (such as Figure 1 As shown in the figure, it is one of the calibration files. This table expresses the mapping relationship between the points on the image and the actual plane points, and the values ​​are determined according to the actual situation); the calibration file is in the lifting plane where the pantograph is located, with the center of the pantograph as the 0 point, and a ruler is used to collect plane points for infrared camera calibration. The mapping relationship between the collected actual points and pixel points is used to obtain a mapping matrix consisting of three parts: rotation, non-uniform scaling and translation.

[0039]

[0040] Among them, R: rotation matrix, T: translation vector, Sx, Sy: scaling size.

[0041] Step S1: Collect the coordinated image of the pantograph and the contact network; in this embodiment, an infrared camera is set on the roof of the train, and the image of the pantograph and the contact network is obtained by the infrared camera. The infrared camera uses temperature to image, which can avoid complex background interference to a certain extent, and because the current is transmitted between the contact network and the pantograph, the thermal effect of the current will increase the temperature and make the characteristics of the pantograph and the contact network more obvious to a certain extent.

[0042] Step S2: Use NCC grayscale template matching to obtain the pantograph position (such as Figure 2 As shown); NCC grayscale matching can avoid the influence of illumination and ensure the adaptability of the matching algorithm.

[0043] Step S3: Reduce the ROI (target area) to the pantograph and the area above it (such as Figure 3 As shown in the figure), and then use the Hough transform line to extract the contact network (as shown in the figure). Figure 4 As shown). Hough detection is performed after grayscale matching detection, which can ensure that only the area above the pantograph is detected to avoid background interference. For multi-line situations, the contact network position will not change suddenly and has a zigzag feature, so that an effective contact network working branch (the part in contact with the pantograph) can be obtained.

[0044] Step S4: Calculate the contact point between the pantograph and the contact network, obtain the accurate position area of ​​the pantograph through the grayscale feature, extract the top contour line of the pantograph and the contact line, solve the intersection point by solving the two straight line equations, and obtain the contact point position coordinates; where Y1: contact network straight line equation. Y2: pantograph top straight line equation.

[0045]

[0046] Step S5: filter the result points according to the results retained by the previous multiple frames to eliminate abnormal points; the filtering process is based on the flexible contact network itself to ensure the uniform wear of the carbon slide plate installation position distribution characteristics, the pull-out value presents a "Z" shape, and the guide height value presents a small range of wave-like law, to predict and filter; the implementation of the filtering algorithm uses the results of the first 40 frames of data to form a real-time update dynamic array, and uses the least squares linear fitting method to obtain the regression equation. The current frame result is judged whether it is within the range of the movement trend of the contact point to define the abnormal value, and the current trend prediction value is used to replace the abnormal value. It can be understood that in other embodiments, more or less than 40 frames of data can also be used according to actual needs; here, the specific number is not limited.

[0047] Step S6: Mapping transformation is performed on the result point of step S5, and matrix transformation is performed to obtain the actual coordinates, and the current guide height value and pull-out value are obtained, as shown below;

[0048]

[0049] Among them, (Px, Py): image coordinate point; (Qx, Qy): result coordinate point; x: pull out, y: guide height.

[0050] Step S7: Determine whether it is abnormal. If the above-obtained conductance value and pull-out value are abnormal values, save the video information and transmit it to the main control device; the main control device can be a computer, a human-computer interaction machine, etc.;

[0051] Step S8: Identify and detect abnormal videos (such as Figure 5As shown), the main control device is provided with a control system, which identifies the pantograph and contact network in the abnormal video based on deep learning;

[0052] Step S9: Perform bow-net contact point identification and detection to obtain contact point coordinates; in this embodiment, deep learning cross-type contact point identification is adopted, such as Figure 6 As shown, the small rectangle indicates the identification position.

[0053] Step S10: judging whether the contact point exceeds the limit according to the threshold setting; comparing and analyzing the acquired contact point coordinates with the standard coordinate range in the database to judge whether the contact point exceeds the limit abnormality.

[0054] In addition, the present invention also discloses a system for dynamic identification and measurement of contact points of a large iron flexible bow-net, which operates by adopting the above-mentioned method for dynamic identification and measurement of contact points of a large iron flexible bow-net. A dynamic identification and measurement system for large railway flexible bow-net contact points comprises an on-board detection system and a ground analysis system, wherein the on-board detection system comprises an image acquisition module, an image processing module, a calculation and analysis module, a judgment module, a data storage module, and a data transmission module; the ground analysis system comprises a deep learning module, a data judgment module, and a database; the image acquisition module acquires an image, and the image processing module performs grayscale matching NCC and Hough line detection on the image to identify the bow-net contact points; the calculation and analysis module calculates the contact points between the pantograph and the contact network, and performs filtering and mapping transformation on the structures retained by multiple frames to obtain the current conduction height value and pull-out value; the judgment module compares and analyzes the analysis value with the standard value, and stores the data information in the data storage module; and the abnormal data information is sent to the ground analysis system through the data transmission module for secondary analysis; the ground analysis system stores the received information in the database, and performs identification and detection through the deep learning module, and the data judgment module compares and analyzes the data information after detection and issues an alarm for confirming the abnormality.

[0055] Furthermore, the image acquisition module is an infrared camera, and the infrared camera is arranged on the roof of the vehicle.

[0056] In summary, the identification and measurement method of the present invention is implemented by an overall architecture of two processes. The first process is located at the vehicle-mounted end, and an infrared camera is used to obtain images, grayscale matching NCC and Hough line detection are used to identify the contact points of the bow and the network, and then real-time rapid analysis is performed to save the possible over-limit problem data. The second process is located in the main control device at the ground end, and the abnormal video is analyzed again to obtain the final abnormal alarm information. Among them, the second process is implemented by deep learning target detection, and the model parameters of the pantograph and the contact point are trained and obtained respectively using the alexnet model framework for forward reasoning. It is carried out in two steps. The first step is to identify the pantograph area, and the second step is to perform target detection on the cross-like intersection of the bow and the network contact in the area after longitudinally expanding part of the area according to the results of the first step. This method combines the rapid response of conventional image algorithms to ensure its real-time performance. The combination of the two processes is used to combine the deep learning strategy in the main control device to perform secondary analysis to ensure the accuracy of abnormal detection. It is highly practical and has a strong promotion significance.

[0057] The above-mentioned embodiment only expresses one implementation mode of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A method for dynamic identification and measurement of the contact point of a large iron flexible bow-net, characterized in that: The steps include: Step S0, calibrate the infrared camera on the pantograph lifting plane to obtain a calibration file; obtain a mapping matrix consisting of three parts: rotation, non-uniform scaling and translation through the mapping relationship between the collected actual points and the pixel points; Among them, R: rotation matrix, T: translation vector, Sx, Sy: scaling size; Step S1: using an infrared camera to collect images of the coordination between the pantograph and the contact network; Step S2: using NCC grayscale template matching to obtain the pantograph position; Step S3: Reduce the ROI (target area) to the pantograph and the area above it in combination with the pantograph position, and then use the Hough transform straight line to extract the overhead line; Step S4: Calculate the contact point between the pantograph and the contact network, obtain the accurate position area of ​​the pantograph through the grayscale feature, extract the uppermost contour line of the pantograph and the contact line, solve the intersection point by simultaneously solving the two straight line equations, and obtain the contact point position coordinates; ； Step S5: filtering the result points according to the results retained by the previous multiple frames to eliminate abnormal points; Step S6: Map the result points and transform them into real coordinates by matrix transformation, and obtain the current height guide value and pull-out value, as shown below; Among them, (Px, Py): image coordinate point; (Qx, Qy): result coordinate point; x: pull out, y: guide height; Step S7: judging whether it is abnormal, if the above-obtained conductance value and pull-out value are abnormal values, saving the video information and transmitting it to the main control device; Step S8: the main control device identifies abnormal videos; Step S9: the main control device performs pantograph-catenary contact point identification and detection to obtain the contact point coordinates; Step S10: Determine whether the contact point is abnormal or not according to the threshold setting.

2. A method for dynamic identification and measurement of large iron flexible bow-net contact points as claimed in claim 1, characterized in that: The calibration file is in the lifting plane where the pantograph is located, with the center of the pantograph as the 0 point, and a ruler is used to collect plane points for infrared camera calibration.

3. A method for dynamic identification and measurement of large iron flexible bow-net contact points as claimed in claim 1, characterized in that: In step S5, prediction and filtering are performed based on the distribution characteristics of the installation position of the flexible contact network itself, the rules of the pull-out value and the height conduction value.

4. A method for dynamic identification and measurement of large iron flexible bow-net contact points as claimed in claim 3, characterized in that: The implementation of the filtering algorithm uses the results of the previous multiple frames of data to form a real-time updated dynamic array, uses the least squares linear fitting method to obtain the regression equation, and judges whether the current frame result is within the movement trend range of the contact point to define the outliers. For outliers, the current trend prediction value is used to replace the outliers.

5. A method for dynamic identification and measurement of large iron flexible bow-net contact points as claimed in claim 3, characterized in that: In steps S8 and S9, a deep learning method is used to identify and detect the pantograph.

6. A dynamic identification and measurement system for the contact point of a large iron flexible bow-net, characterized in that: It includes an on-board detection system and a ground analysis system. The on-board detection system includes an image acquisition module, an image processing module, a calculation and analysis module, a judgment module, a data storage module, and a data transmission module; the ground analysis system includes a deep learning module, a data judgment module, and a database; the image acquisition module acquires images, and the image processing module performs grayscale matching NCC and Hough line detection on the images to identify the contact points between the pantograph and the contact network; the calculation and analysis module calculates the contact points between the pantograph and the contact network, and performs filtering and mapping transformation on the structures retained in multiple frames to obtain the current conduction height value and pull-out value; the judgment module compares and analyzes the analysis value with the standard value, and stores the data information in the data storage module; The abnormal data information is sent to the ground analysis system through the data transmission module for secondary analysis; the ground analysis system stores the received information in the database, and performs identification and detection through the deep learning module, and the data judgment module compares and analyzes the data information after detection and issues an alarm for confirmed abnormalities.

7. A dynamic identification and measurement system for large iron flexible bow-net contact points as claimed in claim 6, characterized in that: The image acquisition module is an infrared camera, and the infrared camera is arranged on the top of the vehicle.

Images