Positioning method and system based on trackside components

Abstract

The present invention relates to a positioning method and system based on trackside components. The method includes marking the relative positions of the components beside the track to be detected; selecting a component in a picture under normal conditions as a reference target, marking the absolute coordinates of the reference target and the target to be located in the picture according to the relative position, and using this as a training set; fitting the mathematical relationship between the target to be located and the reference target based on a neural network according to a deep learning regression algorithm based on the training set; training and establishing a positioning prediction model based on deep learning based on the mathematical relationship; obtaining the absolute position of the target to be located using the positioning prediction model according to the absolute position of the reference target. By marking the positions of the trackside components of various models, fitting the mathematical relationship between the reference target and the target to be located using a regression network, and establishing a positioning prediction model for each specific model, the data demand is greatly solved, the algorithm complexity is reduced, and the calculation speed is improved.

Description

Technical Field

[0001] The present application relates to the technical field of rail data processing, and in particular to a positioning method and system based on railside components. Background Art

[0002] The maintenance and inspection of railway tracks is a very important task in the daily operation of railway tracks. In the past, the inspection relied on the naked eyes of the maintenance workers to observe the key components of the train one by one on site after the train entered the depot to check whether there were any faults or signs of faults.

[0003] With the development of digital technology, information technology and computer technology, the maintenance method has gradually changed from on-site naked eye observation to using linear array cameras to take pictures of both sides of the train when the train enters the depot, uploading them to the computer, and then the maintenance workers view them on the computer.

[0004] Of course, it is still troublesome to drag and drop the entire map to view it. Therefore, with the development of artificial intelligence technology, some semi-automatic or fully automatic fault identification algorithms have emerged one after another. Before the maintenance workers check, they can first run through the algorithm to know which positions are healthy and which positions may have faults. This can effectively assist and speed up the maintenance work of the maintenance workers.

[0005] Existing component localization methods include target detection algorithms based on traditional machine learning (such as HOG+SVM) and target detection algorithms based on deep learning (such as faster rcnn, yolov3, etc.), but all have shortcomings in data, accuracy or speed.

[0006] At present, there is no effective solution to the problems of large data volume and slow computing speed in the process of railway data processing. Summary of the invention

[0007] The embodiments of the present application provide a positioning method and system based on trackside components to at least solve the problems of large data volume and slow computing speed in related technologies.

[0008] In a first aspect, an embodiment of the present application provides a positioning method based on a trackside component, characterized in that it includes the following steps:

[0009] A relative position marking step, marking the relative positions of the components beside the rail to be inspected;

[0010] A training set acquisition step, selecting a component in a picture under normal conditions as a reference target, marking the absolute coordinates of the reference target and the target to be located in the picture according to the relative positions, and using this as a training set;

[0011] A relationship fitting step, using a neural network to fit the mathematical relationship between the target to be located and the reference target according to a deep learning regression algorithm;

[0012] A model training step, based on the mathematical relationship, establishing a positioning prediction model according to deep learning training;

[0013] The positioning prediction step is to obtain the absolute position of the target to be positioned by using the positioning prediction model according to the absolute position of the reference target.

[0014] In some embodiments, the relationship fitting step specifically includes:

[0015] According to the deep learning regression algorithm, the processed absolute coordinates of the reference target are used as input, and the processed absolute coordinates of the target to be located are used as output, and a nonlinear neural network is obtained by using a fully connected network and a nonlinear activation function.

[0016] In some embodiments, the model training step specifically includes the following steps:

[0017] A training set data processing step, performing data enhancement processing on the training set;

[0018] A data input step, inputting the data-enhanced training set into the nonlinear neural network;

[0019] The positioning prediction model training step obtains a gradient vector through forward propagation and back propagation, and adjusts the weight accordingly to adjust the error score of the positioning prediction model to zero or convergence trend.

[0020] In some embodiments, the positioning prediction method specifically includes the following steps:

[0021] A reference target data input step, using the positioning prediction model to input the absolute coordinates of the reference target, and obtain the relative coordinates of all the targets to be positioned relative to the reference target;

[0022] The absolute coordinate calculation step is to obtain the absolute coordinates of all the targets to be located according to the absolute target of the reference target and the relative coordinates.

[0023] In some embodiments, when the position of the reference target is not fixed, the training set acquisition step specifically includes the following steps:

[0024] The target detector training step is to select pictures that meet the preset features to train a target detector;

[0025] The training is followed by a labeling step, in which the absolute coordinates of the reference target and the target to be located in the image are labeled using the target detector.

[0026] In a second aspect, an embodiment of the present application provides a positioning system based on a trackside component, applying the positioning method based on a trackside component described in the first aspect, including:

[0027] The relative position marking module marks the relative positions of the components beside the rails to be inspected;

[0028] A training set acquisition module selects a component in a picture under normal conditions as a reference target, marks the absolute coordinates of the reference target and the target to be located in the picture according to the relative positions, and uses this as a training set;

[0029] A relationship fitting module, using a neural network to fit the mathematical relationship between the target to be located and the reference target according to a deep learning regression algorithm;

[0030] A model training module, based on the mathematical relationship and according to deep learning training, establishes a positioning prediction model;

[0031] The positioning prediction module obtains the absolute position of the target to be positioned by using the positioning prediction model according to the absolute position of the reference target.

[0032] In some of the embodiments, the relationship fitting module uses a deep learning regression algorithm, takes the processed absolute coordinates of the reference target as input, and takes the processed absolute coordinates of the target to be located as output, uses a fully connected network, and adopts a nonlinear activation function to obtain a nonlinear neural network.

[0033] In some embodiments, the model training module specifically includes:

[0034] A training set data processing unit, performing data enhancement processing on the training set;

[0035] A data input unit, inputting the data-enhanced training set into the nonlinear neural network;

[0036] The positioning prediction model training unit obtains a gradient vector through forward propagation and back propagation, and adjusts the weight accordingly to adjust the error score of the positioning prediction model to zero or converge.

[0037] In some embodiments, the positioning prediction system specifically includes:

[0038] A reference target data input unit, using the positioning prediction model, inputs the absolute coordinates of the reference target to obtain the relative coordinates of all targets to be positioned relative to the reference target;

[0039] The absolute coordinate calculation unit processes the absolute coordinates of all the targets to be located according to the absolute target of the reference target and the relative coordinates.

[0040] In some embodiments, when the position of the reference target is not fixed, the training set acquisition module specifically includes:

[0041] The object detector training unit selects pictures that meet the preset features to train an object detector;

[0042] The training and labeling unit utilizes the target detector to label the absolute coordinates of the reference target and the target to be located in the picture.

[0043] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the positioning method based on trackside components as described in the first aspect above is implemented.

[0044] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the positioning method based on trackside components as described in the first aspect above.

[0045] Compared with the related art, the positioning method and system based on trackside components provided in the embodiments of the present application can be applied to the field of deep learning technology, and can also be specifically applied to the field of computer vision technology. By marking the relative positions of various types of trackside components, using a deep regression network to fit the mathematical relationship between the reference target and the target to be located, and establishing a positioning prediction model for each specific model, it greatly solves the data demand and the number of annotations, reduces the algorithm complexity, and improves the calculation speed. At the same time, it is completely unaffected by imaging quality problems caused by external environmental interference.

[0046] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0048] Figure 1 is a flow chart of a positioning method based on trackside components according to an embodiment of the present application;

[0049] Figure 2is a flowchart of the model training steps according to an embodiment of the present application;

[0050] Figure 3 is a flowchart of a positioning prediction method according to an embodiment of the present application;

[0051] Figure 4 is a flowchart of the training set acquisition steps according to an embodiment of the present application;

[0052] Figure 5 is a flow chart of another positioning method based on trackside components according to an embodiment of the present application,

[0053] Figure 6 It is a flow chart of a positioning method based on trackside components according to a preferred embodiment of the present application.

[0054] Figure 7 This is a picture example 1 of a positioning method based on trackside components according to a preferred embodiment of the present application.

[0055] Figure 8 This is a second picture example of a positioning method based on trackside components according to a preferred embodiment of the present application;

[0056] Fig. 9 Design graphs for deep learning regression networks;

[0057] Fig.10 A schematic diagram of an original picture in an embodiment of the present application;

[0058] Fig.11 This is a schematic diagram of the actual results in the examples of this application;

[0059] Fig.12 A schematic diagram of the prediction results in the embodiment of this application.

[0060] Fig.13 is a structural block diagram of a positioning system based on trackside components according to an embodiment of the present application;

[0061] Fig.14 Schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application.

[0062] Fig.15 Schematic diagram of the Yolov3 algorithm structure. Description of the drawings:

[0064] Relative position labeling module 1; training set acquisition module 2; relationship fitting module 3;

[0065] Model training module 4; Positioning prediction module 5; Training set data processing unit 41;

[0066] Data input unit 42; Positioning prediction model training unit 43; Reference target data input unit 51;

[0067] Absolute coordinate calculation unit 52; target detector training unit 21; training and labeling unit 22;

[0068] Processor 81; memory 82; communication interface 83; bus 80. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated 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 application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0070] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0071] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0072] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0073] The existing component positioning methods mainly include target detection algorithms based on traditional machine learning and target detection and recognition algorithms based on regression, among which:

[0074] The traditional target detection and recognition methods in the target detection algorithm based on traditional machine learning can be mainly expressed as: target feature extraction -> target recognition -> target positioning.

[0075] The features used here are all artificially designed, such as SIFT (Scale Invariant Feature Transform), HOG (Histogram of Oriented Gradient), SURF (Speeded Up Robust Features), etc. These features are used to identify the target, and then the target is located in combination with the corresponding strategy.

[0076] Taking HOG features as an example: HOG+SVM detection algorithm

[0077] The main steps are:

[0078] a Extract HOG features;

[0079] b. Train the SVM classifier;

[0080] c. Use the sliding window to extract the target area and classify it;

[0081] d. Use the NMS algorithm to screen and merge candidate boxes;

[0082] e outputs the final detection result;

[0083] The HOG features are:

[0084] aFirst, grayscale and gamma transform the image;

[0085] bCalculate the gradient map

[0086] c. Divide the image into multiple small cells and count the gradient histogram of each cell.

[0087] d Multiple cells form a block for feature normalization

[0088] e. Splice multiple blocks and normalize them

[0089] Currently, target detection and recognition based on deep learning has become the mainstream method, which can be mainly expressed as: deep feature extraction of images -> target recognition and positioning based on deep neural networks, among which the main deep neural network model used is the convolutional neural network CNN.

[0090] At present, the main target detection algorithms based on deep learning are: target detection and recognition algorithms based on region proposals, such as the R-CNN series (R-CNN, Fast-R-CNN, Faster-R-CNN, etc.); target detection and recognition algorithms based on regression, such as the YOLO series (yolov2, yolov3, yolov4, etc.); this type of algorithm only needs to label the data, train offline to obtain the model file, input the picture when using it, call the model file for inference, and you can get the detection result directly in one step (N targets, each target contains coordinate box and confidence score information).

[0091] Taking the Yolov3 algorithm as an example, Fig.15 It is a schematic diagram of the Yolov3 algorithm structure, such as Fig.15 shown.

[0092] Currently, the shortcomings of the existing solutions are as follows:

[0093] Disadvantages of traditional target detection algorithms: complex implementation; low accuracy; slow speed; extremely susceptible to abnormal interference such as deformation, brightness changes, and noise.

[0094] Disadvantages of deep learning-based target detection algorithms: They work well when there is enough data, but they are ineffective or even fail when the amount of data is small; they cannot completely avoid abnormal interference such as deformation, brightness changes, and noise; the higher the accuracy, the slower the target detection algorithm.

[0095] Accordingly, the present invention is mainly aimed at trackside image data, especially subway trackside image data, because subway models are limited, that is, the appearance and relative positions of components on the subway trackside remain unchanged all the time or for a long period of time.

[0096] However, existing target detection algorithms are all aimed at natural scenes (i.e., the target appearance, quantity, and position are uncertain), and are easily affected by external interference, which affects the image quality and leads to false detections and missed detections. At the same time, the speed is not particularly fast.

[0097] Therefore, this embodiment provides a positioning method based on trackside components. Figure 1 is a flow chart of a positioning method based on trackside components according to an embodiment of the present application, such as Figure 1 As shown, the process includes the following steps:

[0098] Relative position marking step S1, marking the relative positions of the components beside the rail to be inspected;

[0099] In the training set acquisition step S2, a component in a picture under normal conditions is selected as a reference target, and the absolute coordinates of the reference target and the target to be located in the picture are marked according to the relative positions, and this is used as the training set;

[0100] Relationship fitting step S3, using a neural network to fit the mathematical relationship between the target to be located and the reference target according to a deep learning regression algorithm;

[0101] Model training step S4, based on mathematical relationships and deep learning training, establishes a positioning prediction model;

[0102] Positioning prediction step S5, according to the absolute position of the reference target, using the positioning prediction model, to obtain the absolute position of the target to be positioned.

[0103] It should be noted that the above-mentioned normal conditions refer to the situation where the trackside components meet the precise position specified by the rail industry standards or are in normal operation and meet the allowable error range of the rail industry standards.

[0104] Through the above steps, the rule-based and deep learning regression methods are used to mark the relative positions of various components on different models, and the deep learning target detection algorithm is used to select the reference target. Then, the deep learning regression algorithm is used to fit the mathematical relationship of the relative positions of various components relative to the reference target. Finally, the absolute coordinates of other targets are obtained based on the absolute coordinates of the reference target and the relative coordinate relationship between other targets and the reference target.

[0105] This process greatly reduces the amount of data required and the number of annotations. There is no need to design complex mathematical formulas and reasoning processes, which reduces the complexity of the algorithm. The mathematical relationship between the reference target and the target to be located is obtained by fitting. The regression model obtained by fitting is very small, the reasoning speed is fast, and the calculation speed is improved. Due to the characteristics of the rails themselves, the objective positional relationship between the components beside the rails is directly replaced by the image features, which is completely unaffected by the imaging quality problems caused by external environmental interference, further improving the accuracy of positioning prediction.

[0106] In some embodiments, the relationship fitting step S3 specifically includes:

[0107] According to the deep learning regression algorithm, the absolute coordinates of the processed reference target are used as input, and the absolute coordinates of the processed target to be located are used as output. A nonlinear neural network is obtained by using a fully connected network and a nonlinear activation function.

[0108] Figure 2 is a flowchart of the model training steps according to an embodiment of the present application, such as Figure 2 As shown, in some embodiments, the model training step S4 specifically includes the following steps:

[0109] Training set data processing step S41, performing data enhancement processing on the training set;

[0110] Data input step S42, inputting the data-enhanced training set into the nonlinear neural network;

[0111] The positioning prediction model training step S43 obtains a gradient vector through forward propagation and back propagation, and adjusts the weight accordingly to adjust the error score of the positioning prediction model to zero or converge.

[0112] In practical applications, the steps of deep learning training models include: 1. Preprocessing the training data (i.e., the training set described in the embodiments of the present application); 2. Inputting the data into the neural network for forward propagation, specifically, each neuron first inputs the value weighted accumulation and then inputs the activation function as the output value of the neuron to obtain an error score; 3. Inputting the "error score" into the error function, comparing it with the expected value to obtain the error, and multiple values ​​are summed, and the recognition degree is judged by the error; 4. Through back propagation, specifically, reverse differentiation, the error function and each activation function in the neural network are required to minimize the error to determine the gradient vector; 5. Finally, each weight is adjusted by the gradient vector to adjust the "error score" so that the error tends to 0 or converges; 6. Repeat the above process until the set number of times or the average value of the loss no longer decreases (the lowest point).

[0113] Figure 3 is a flowchart of a positioning prediction method according to an embodiment of the present application. Figure 3 As shown, in some embodiments, the positioning prediction method S5 specifically includes the following steps:

[0114] In step S51, the reference target data is inputted, using the positioning prediction model, the absolute coordinates of the reference target are inputted to obtain the relative coordinates of all the targets to be positioned relative to the reference target;

[0115] The absolute coordinate calculation step S52 is to process and obtain the absolute coordinates of all the targets to be located according to the absolute target and relative coordinates of the reference target.

[0116] Figure 4 is a flowchart of the training set acquisition steps according to an embodiment of the present application, such as Figure 4 As shown, in some embodiments, when the position of the reference target is not fixed, the training set acquisition step S2 specifically includes the following steps:

[0117] The target detector training step S21 is to select a picture that meets the preset features to train a target detector;

[0118] The training is followed by a labeling step S22, in which the absolute coordinates of the reference target and the target to be located in the image are labeled using a target detector.

[0119] In actual applications, select a reference target. If the absolute position of the reference target is fixed, specify it manually. If the reference target position is not fixed, select the target that is easiest to detect and least likely to be affected by external interference and affect the image quality, and train a simple target detector.

[0120] The specific selection steps are mainly based on the analysis of the data of the actual scene, and the selection is based on comprehensive considerations. The factors that need to be considered include: every picture exists, is not easy to be deformed, rotated, blurred, too bright, too dark, is not too large or too small in size in the original picture, and has obvious contours, textures or colors. In summary: must be present, the image quality is high and stable, the size is not too large or too small, the features are obvious, and it is not easy to cause false detection or missed detection.

[0121] This embodiment also provides a positioning method based on trackside components. Figure 5 is a flow chart of another positioning method based on trackside components according to an embodiment of the present application, such as Figure 5 As shown, the process includes the following steps:

[0122] S501, select a reference target. If the absolute position of the reference target is fixed, specify it manually. If the position of the reference target is not fixed, select the target that is easiest to detect and least likely to be affected by external interference and affect the image quality, and train a simple target detector;

[0123] S502, labeling the training set. Use the target detector to label the absolute coordinates of the reference target and the target to be located in the image;

[0124] S503, design a deep learning regression network. The absolute coordinates of the reference target are transformed as input, and the absolute coordinates of the target to be located are transformed as output. The main body of the network structure uses a fully connected network, and the activation function uses a nonlinear activation function (such as relu).

[0125] The principle of S503 is that any objective mapping relationship can be fitted using more than two layers of nonlinear neural networks.

[0126] S504, training the model. Use the labeled training data and the designed deep learning regression network to perform deep learning training. In this process, the image needs to be resized for data enhancement so that the trained model is robust to the zooming in and out of the image.

[0127] S505, predictive positioning. Use the trained model. When using, input the absolute coordinates of the reference target, use the positioning prediction model (specifically, a model trained with the relative position of the same component model as the target to be predicted) to obtain the absolute coordinates of all targets to be positioned.

[0128] Through the above steps, rule reasoning is used to output the coordinates of other targets relative to the reference target according to the positioning prediction model. The absolute coordinates of other targets can be obtained by post-processing and back-calculation based on the absolute coordinates of the reference target and the relative coordinates output by the algorithm.

[0129] The embodiments of the present application are described and illustrated below through preferred embodiments.

[0130] Figure 6 It is a flow chart of a positioning method based on trackside components according to a preferred embodiment of the present application.

[0131] Figure 7 This is a picture example 1 of a positioning method based on trackside components according to a preferred embodiment of the present application.

[0132] Figure 8 This is a second picture example of a positioning method based on trackside components according to a preferred embodiment of the present application.

[0133] S601, select a reference target and mark the results.

[0134] like Figure 7 As shown, the center of the wheelset in the trackside image is used as the reference target, and its coordinates are as follows: 856,385,1488,1050

[0135] Labeled dataset, labeling result example:

[0136] 916,418,985,507,0

[0137] 1221,434,1289,528,0

[0138] 1327,706,1399,797,0

[0139] 1189,997,1254,1075,0

[0140] 920,989,983,1064,0

[0141] 795,690,866,776,0

[0142] 1293,332,1690,708,4

[0143] 786,416,1406,1075,8

[0144] S602, design deep learning regression network.

[0145] Train deep learning regression networks using labeled datasets. Fig. 9 Design diagram for deep learning regression network, and example results are shown in Fig. 9 shown.

[0146] S603: training a positioning prediction model, and using the positioning prediction model to perform prediction.

[0147] The prediction results are as follows Fig.10 ,11 As shown, Fig.10 A schematic diagram of an original picture in an embodiment of the present application; Fig.11 This is a schematic diagram of the actual results in the examples of this application; Fig.12 Schematic diagram of the prediction results in the examples of this application.

[0148] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0149] This embodiment also provides a positioning system based on trackside components, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the terms "module", "unit", "subunit", etc. can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0150] Fig.13 is a structural block diagram of a positioning system based on trackside components according to an embodiment of the present application, such as Fig.13 As shown, the device comprises:

[0151] Relative position marking module 1, marking the relative positions of various components beside the rail to be inspected;

[0152] Training set acquisition module 2, selects a component in a picture under normal conditions as a reference target, marks the absolute coordinates of the reference target and the target to be located in the picture according to their relative positions, and uses this as a training set;

[0153] The relationship fitting module 3 uses a neural network to fit the mathematical relationship between the target to be located and the reference target according to the deep learning regression algorithm;

[0154] Model training module 4, based on mathematical relationships and deep learning training, establishes a positioning prediction model;

[0155] The positioning prediction module 5 obtains the absolute position of the target to be positioned using the positioning prediction model according to the absolute position of the reference target.

[0156] In some of the embodiments, the relationship fitting module 3 takes the processed absolute coordinates of the reference target as input and the processed absolute coordinates of the target to be located as output according to a deep learning regression algorithm, uses a fully connected network, and adopts a nonlinear activation function to obtain a nonlinear neural network.

[0157] In some embodiments, the model training module 4 specifically includes:

[0158] A training set data processing unit 41 performs data enhancement processing on the training set;

[0159] A data input unit 42 inputs the data-enhanced training set into the nonlinear neural network;

[0160] The positioning prediction model training unit 43 obtains a gradient vector through forward propagation and backward propagation, and adjusts the weight accordingly to adjust the error score of the positioning prediction model to a trend of zero or convergence.

[0161] In some embodiments, the positioning prediction system 5 specifically includes:

[0162] The reference target data input unit 51 uses the positioning prediction model to input the absolute coordinates of the reference target and obtain the relative coordinates of all the targets to be positioned relative to the reference target;

[0163] The absolute coordinate calculation unit 52 processes the absolute coordinates of all the targets to be located according to the absolute target and relative coordinates of the reference target.

[0164] In some embodiments, when the position of the reference target is not fixed, the training set acquisition module 2 specifically includes:

[0165] The object detector training unit 21 selects pictures that meet the preset features to train an object detector;

[0166] The training and labeling unit 22 uses a target detector to label the absolute coordinates of the reference target and the target to be located in the picture.

[0167] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0168] In addition, combined Figure 1 The positioning method of the trackside components described in the embodiment of the present application can be implemented by an electronic device. Fig.14 Schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application.

[0169] The electronic device may include a processor 81 and a memory 82 storing computer program instructions.

[0170] Specifically, the processor 81 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0171] Among them, the memory 82 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 82 may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 82 may include removable or non-removable (or fixed) media.

[0172] In appropriate cases, the memory 82 may be inside or outside the data processing device. In a specific embodiment, the memory 82 is a non-volatile memory. In a specific embodiment, the memory 82 includes a read-only memory (ROM) and a random access memory (RAM). In appropriate cases, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM), or a flash memory (FLASH) or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0173] The memory 82 may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 81 .

[0174] The processor 81 reads and executes the computer program instructions stored in the memory 82 to implement any one of the positioning methods based on trackside components in the above embodiments.

[0175] In some of the embodiments, the electronic device may further include a communication interface 83 and a bus 80. Fig.14As shown, the processor 81, the memory 82, and the communication interface 83 are connected via a bus 80 and communicate with each other.

[0176] The communication interface 83 is used to implement communication between the modules, devices, units and / or equipment in the embodiment of the present application. The communication port 83 can also implement data communication with other components such as: external devices, image / data acquisition equipment, databases, external storage, and image / data processing workstations.

[0177] The bus 80 includes hardware, software or both, and couples the components of the electronic device to each other. The bus 80 includes but is not limited to at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 80 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, bus 80 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0178] The electronic device can perform the model training step in the embodiment of the present application based on the acquired absolute position of the reference target and the relative position of the target to be positioned, thereby realizing the combination of Figure 1 A positioning method based on trackside components is described.

[0179] In addition, in combination with the positioning method based on the trackside component in the above embodiment, the embodiment of the present application can provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the positioning methods based on the trackside component in the above embodiment is implemented.

[0180] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0181] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A positioning method based on trackside components, characterized in that: The following steps are involved: A relative position marking step, marking the relative positions of the components beside the rail to be inspected; A training set acquisition step, selecting a component in a picture under normal conditions as a reference target, marking the absolute coordinates of the reference target and the target to be located in the picture according to the relative positions, and using this as a training set; When the position of the reference target is not fixed, the training set acquisition step specifically includes the following steps: The target detector training step is to select pictures that meet the preset features to train a target detector; A training and labeling step, using the target detector to label the absolute coordinates of the reference target and the target to be located in the image; A relationship fitting step, using a neural network to fit the mathematical relationship between the target to be located and the reference target based on the training set according to a deep learning regression algorithm; A model training step, based on the mathematical relationship, performs training according to deep learning and establishes a positioning prediction model; the model training step specifically includes the following steps: a training set data processing step, performing data enhancement processing on the training set; a data input step, inputting the training set after data enhancement into a nonlinear neural network; a positioning prediction model training step, obtaining a gradient vector through forward propagation and back propagation, and adjusting the weight accordingly to adjust the error score of the positioning prediction model to zero or convergence trend; A positioning prediction step, according to the absolute position of the reference target, using the positioning prediction model, to obtain the absolute position of the target to be located; the positioning prediction step specifically includes the following steps: a reference target data input step, using the positioning prediction model, inputting the absolute coordinates of the reference target, and obtaining the relative coordinates of all the targets to be located relative to the reference target; The absolute coordinate calculation step is to obtain the absolute coordinates of all the targets to be located according to the absolute target of the reference target and the relative coordinates.

2. The positioning method based on trackside components according to claim 1, characterized in that: The relationship fitting step specifically includes: According to the deep learning regression algorithm, the processed absolute coordinates of the reference target are used as input, and the processed absolute coordinates of the target to be located are used as output, and a nonlinear neural network is obtained by using a fully connected network and a nonlinear activation function.

3. A positioning system based on a trackside component, using the positioning method based on a trackside component according to any one of claims 1 to 2, characterized in that: include: The relative position marking module marks the relative positions of the components beside the rails to be inspected; A training set acquisition module selects a component in a picture under normal conditions as a reference target, marks the absolute coordinates of the reference target and the target to be located in the picture according to the relative positions, and uses this as a training set; When the position of the reference target is not fixed, the training set acquisition module specifically includes: The object detector training unit selects pictures that meet the preset features to train an object detector; A training and labeling unit is used to label the absolute coordinates of the reference target and the target to be located in the picture by using the target detector; A relationship fitting module, using a neural network to fit the mathematical relationship between the target to be located and the reference target according to a deep learning regression algorithm; A model training module, based on the mathematical relationship, establishes a positioning prediction model according to deep learning training; the model training module specifically includes: a training set data processing unit, which performs data enhancement processing on the training set; a data input unit, which inputs the training set after data enhancement into the nonlinear neural network; a positioning prediction model training unit, which obtains a gradient vector through forward propagation and back propagation, and adjusts the weight accordingly to adjust the error score of the positioning prediction model to zero or convergence trend; A positioning prediction module, which obtains the absolute position of the target to be positioned by using the positioning prediction model according to the absolute position of the reference target; the positioning prediction module specifically includes: A reference target data input unit, using the positioning prediction model, inputs the absolute coordinates of the reference target to obtain the relative coordinates of all targets to be positioned relative to the reference target; The absolute coordinate calculation unit processes the absolute coordinates of all the targets to be located according to the absolute target of the reference target and the relative coordinates.

4. The positioning system based on trackside components according to claim 3, characterized in that: The relationship fitting module uses the processed absolute coordinates of the reference target as input and the processed absolute coordinates of the target to be located as output according to the deep learning regression algorithm, and uses a fully connected network and a nonlinear activation function to obtain a nonlinear neural network.

Images