CN110827354B - A train positioning method based on the counting of the poles of the wayside grid - Google Patents

Abstract

The present invention provides a train positioning method based on the counting of wayside power grid poles. The high-speed camera in the middle of the train head collects the running video of the train; image recognition algorithm is used to identify the track line and the vanishing point of the running video; according to the trained classifier, the power grid pole on the right side of the vanishing point is identified , and use the grid pole counting algorithm to record the total number of currently identified grid poles; determine the current position of the train according to the current total number of grid poles. The method can improve the situation that the existing train positioning methods are expensive and require regular maintenance, and reduce the economic burden of the railway transportation industry.

Description

A train positioning method based on the counting of the poles of the wayside grid

technical field

The invention relates to the technical field of rail traffic control, in particular to a train positioning method based on trackside grid line pole counting.

Background technique

With the development of the modern railway transportation industry, the pressure of railway operation is increasing day by day, and the high-speed railway is developing rapidly. In order to ensure the safety of railway traffic, the train positioning technology in the train control system is indispensable. The existing train positioning methods are mainly based on speed measurement positioning, transponder positioning, and track circuit positioning as auxiliary. Among them, the transponder positioning method has the highest positioning accuracy and can correct the cumulative error of the train during the speed measurement and positioning process. However, the transponder is expensive and requires regular maintenance, which consumes a lot of manpower and material resources, and brings a serious economic burden to the railway transportation industry. maintenance workload. At present, some scholars are studying several new types of train positioning methods, such as satellite positioning, induction loop positioning, wireless signal strength positioning, multi-positioning method fusion positioning, etc., but they are easily interfered by the external environment and the positioning accuracy is not high. In this case, it is of great practical significance to explore a new generation of low-cost, high-reliability train positioning technology.

SUMMARY OF THE INVENTION

The present invention provides a train positioning method based on the counting of the line poles of the trackside power grid, so as to improve the situation that the existing train positioning method is expensive and requires regular maintenance, and reduces the economic burden of the railway transportation industry.

In order to achieve the above objects, the present invention adopts the following technical solutions.

The invention provides a method for locating a train based on the counting of the poles of a trackside power grid, comprising:

Aiming at the wayside power grid poles, the classifier is trained by using the visual geometry group neural network classifier algorithm based on convolutional neural network;

The running video of the train is collected by the high-speed camera installed on the front of the train;

Using an image recognition algorithm to identify the track line and the vanishing point of the running video;

According to the trained classifier, identify the power grid pole on the right side of the vanishing point, and record the total number of the currently identified power grid poles by using the power grid pole counting algorithm;

Determine the current position of the train based on the total number of current grid poles.

Preferably, the visual geometry group neural network classifier based on convolutional neural network consists of 5 layers of convolution layers, 3 layers of fully connected layers and normalized exponential output layer, wherein the convolution layer uses 3*3 convolution kernels, The layers are separated by a 2*2 max pooling layer, and the activation units of all hidden layers use a linear rectification function.

Preferably, for the trackside power grid poles, a classifier is trained by using the visual geometry group neural network classifier algorithm based on convolutional neural network, including: the positive sample is a picture of the trackside power grid pole in the foreground, and the negative sample refers to only containing For the local information of the trackside power grid poles or the pictures that do not contain the trackside power grid poles, the positive sample pictures are included in the training set and the positive example test set, and the negative sample pictures are included in the negative example test set.

The running video of the train is collected by a high-speed camera installed at the head of the train, including: the high-speed camera is installed in the middle of the head of the train.

Preferably, an image recognition algorithm is used to identify the track line and the vanishing point of the running video, including:

The collected image is filtered by Gaussian filter to remove high frequency components;

Use Canny algorithm for edge detection, add a trapezoidal window to the detection result, and filter out the foreground;

Perform Hough transform on the foreground to find a straight line, where the longer straight line is given a higher weight when calculating the mean slope;

Distinguish the left and right tracks according to the positive and negative slope of the found straight line;

Iteratively calculate the difference between the slope of each of the straight lines and the mean value of the slope, and remove the straight line whose difference is greater than a predetermined value;

Perform linear regression on the straight line sets of the remaining left and right orbital lines to obtain the final left and right orbital lines, and the intersection of the final left and right orbital lines is the vanishing point.

Preferably, according to the trained classifier, identify the power grid poles on the right side of the vanishing point, and record the total number of the currently identified power grid poles, including the following steps:

For the currently collected pictures, from right to left, from top to bottom, use the 112×112×3 window to take out the picture containing the complete line rod in turn, and put the picture containing the complete line rod through 64 3×3 windows. The convolution kernel performs two convolutions and linear rectification, and the size after convolution becomes 112×112×64; the pooling unit with size 2×2 is used for maximum pooling, and the size after pooling becomes 64× 64×64; after 128 convolution kernels of 3×3 do two convolutions and linear rectification, the size becomes 64×64×128; for 2×2 max pooling, the size becomes 32×32×128; After 256 3 × 3 convolution kernels are used for cubic convolution and linear rectification, the size becomes 32 × 32 × 256; after 2 × 2 maximum pooling, the size becomes 16 × 16 × 256; after 512 3 × The convolution kernel of 3 is used for cubic convolution and linear rectification, and the size becomes 16×16×512; for 2×2 maximum pooling, the size becomes 8×8×512; after 512 3×3 convolution kernels For cubic convolution and linear rectification, the size becomes 8×8×512; for 2×2 maximum pooling, the size becomes 4×4×512; with two layers of 1×1×4096, one layer of 1×1× 1000 is fully connected linear rectification; 1000 prediction results are output through the normalized exponential function; the first one of the prediction results is a power grid pole and the probability is greater than 98%, it is considered that the power grid pole is identified; this area is in the original image Mark with a rectangular frame adjusted by length and width, and record the horizontal and vertical coordinates and width and height of the upper left corner of the rectangular frame; move the position of the window by a certain area in sequence.

Preferably, the total number of currently identified grid poles is recorded, including:

For each frame of image, the classifier will look for power grid poles in the original image in order from right to left and top to bottom;

Each time a power grid pole is found, the classifier returns the horizontal and vertical coordinates and width and height information of the upper left corner of the current power grid pole;

The search ends when the number of grid poles found reaches three or the abscissa of the vanishing point has been scanned;

If the power grid pole on the far right of the current frame is still within the image acquisition range in the next frame, the maximum abscissa scanned will be larger than the previous frame; The abscissa value will jump and become the second largest abscissa value. If the jumping process does not repeat for 5 consecutive frames, it is considered that a power grid pole has left the camera's field of view. At this time, the power grid line The rod count value is incremented by 1.

Preferably, the current position of the train is determined according to the total number of current grid poles, including:

Store the corresponding relationship between the number of grid poles and the operating position in the database, and query the total number of current grid poles to determine the current position of the train according to the corresponding relationship between the number of grid poles and the operating position stored in the database.

It can be seen from the technical solution provided by the above-mentioned method for locating the train based on the counting of the trackside power grid poles of the present invention, the method of the present invention takes into account the fixed position of the train trackside power grid poles, so long as it can be accurately counted, the train can be determined. location, which can improve existing train positioning methods that are expensive and require regular maintenance, reducing the financial burden on the rail transport industry.

Additional aspects and advantages of the present invention will be set forth in part in the following description, which will be apparent from the following description, or may be learned by practice of the present invention.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

FIG. 1 provides a schematic flowchart of a method for locating a train based on the counting of poles in a wayside power grid in the present embodiment;

2 is a schematic diagram of a recognition process of the image recognition algorithm of the present embodiment;

Fig. 3 is the schematic flow chart of adopting the vanishing point detection algorithm to identify the power grid pole;

FIG. 4 is a schematic diagram of a process of recording the total number of currently identified power grid poles by using a grid pole counting algorithm.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but not to be construed as a limitation of the present invention.

It will be understood by those skilled in the art that the singular forms "a", "an", "the" and "the" as used herein can include the plural forms as well, unless expressly stated otherwise. It should be further understood that the word "comprising" used in the description of the present invention refers to the presence of stated features, integers, steps, operations, elements and/or components, but does not exclude the presence or addition of one or more other features, Integers, steps, operations, elements, components and/or groups thereof. It will be understood that the term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as herein, are not to be taken in an idealized or overly formal sense. explain.

In order to facilitate the understanding of the embodiments of the present invention, the following will take specific embodiments as examples for further explanation and description in conjunction with the accompanying drawings.

Example

FIG. 1 provides a schematic flowchart of a method for locating a train based on the counting of the poles of a trackside power grid for this embodiment, including:

S1 uses the Visual Geometry Group (VGG16) neural network classifier algorithm based on Convolutional Neural Networks (CNN) to train the classifier for the trackside power grid poles.

Convolutional Neural Networks are a class of feedforward neural networks that contain convolutional computations and deep structures, which are very efficient in the field of image recognition. VGG is a deep convolutional network jointly developed by Oxford University's Computer Vision Group and DeepMind, which has achieved extremely high achievements in the field of image classification. VGG16 is a widely used network structure in VGG.

The visual geometry group neural network classifier based on convolutional neural network consists of 5 layers of convolution layers, 3 layers of fully connected layers and normalized index output layer, in which the convolution layer uses 3*3 convolution kernels, layers and layers A 2*2 max pooling layer is used to separate them, and the activation units of all hidden layers use a linear rectification function.

Training a classifier requires preparing samples in advance. In this embodiment, the positive samples are pictures with the trackside power grid poles in the foreground, and the negative samples refer to pictures that only contain the local information of the trackside power grid poles or do not include the trackside power grid lines and poles. The positive sample pictures are classified into the training set and the positive samples Example test set, negative sample pictures are included in the negative example test set. This embodiment specifically collects 600 positive samples and 1600 negative samples. Adjust the size of the above samples to 112px*112px without grayscale processing. Among the positive sample images, 500 positive sample images are included in the training set, 100 positive sample images are included in the positive sample test set, and negative samples are included in the training set. Counterexample test set.

After initializing the neural network and related parameters, input the positive and negative samples prepared in the steps into the neural network, and after debugging the related parameters, after 15 rounds of training, the generated CNN-VGG16 classifier for power grid poles can be obtained.

S2 collects the running video of the train through a high-speed camera installed at the head of the train.

The train running image used in this embodiment is collected by a high-speed camera installed in the front and middle of the head of the running rail locomotive. From the image, the train track line and the power grid poles on the left and right sides can be clearly distinguished.

S3 uses an image recognition algorithm to identify the track line and vanishing point of the running video.

FIG. 2 is a schematic diagram of the recognition process of the image recognition algorithm of the present embodiment, and with reference to FIG. 2, it specifically includes:

S31 filters out the high-frequency components of the collected image through a Gaussian filter.

S32 uses the Canny algorithm for edge detection, and adds a trapezoidal window to the detection result to filter out the foreground.

S33 performs Hough transform on the foreground, and finds a straight line, wherein a longer straight line is given a higher weight when calculating the mean value of the slope.

S34 discriminates the left and right tracks according to the positive and negative slope of the straight line found. Divide a line to belong to the left track or the right track.

S35 iteratively calculates the difference between the slope of each of the straight lines and the mean value of the slope, and removes the straight line whose difference is greater than a predetermined value;

S36 performs linear regression on the set of straight lines of the remaining left and right orbital lines to obtain the final left and right orbital lines, and the intersection of the final left and right orbital lines is the vanishing point.

S4 identifies the power grid poles on the right side of the vanishing point according to the trained classifier, and records the total number of the currently identified power grid lines.

Figure 3 is a schematic diagram of the process of identifying power grid poles using the vanishing point detection algorithm. Referring to Figure 3, it specifically includes: from right to left, from top to bottom, sequentially using a 112x112x3 window to extract the complete line pole from the currently collected picture. Picture, the picture containing the complete line is subjected to 64 3x3 convolution kernels for twice convolution and linear rectification, and the size after convolution becomes 112 × 112 × 64; using a pooling unit with a size of 2 × 2 For maximum pooling, the size after pooling becomes 64×64×64; after 128 3×3 convolution kernels do twice convolution and linear rectification, the size becomes 64×64×128; make 2× 2 maximum pooling, the size becomes 32×32×128; after 256 3×3 convolution kernels are used for cubic convolution and linear rectification, the size becomes 32×32×256; 2×2 maximum pooling , the size becomes 16×16×256; after 512 3×3 convolution kernels are used for cubic convolution and linear rectification, the size becomes 16×16×512; for 2×2 maximum pooling, the size becomes 8 ×8×512; 512 3×3 convolution kernels are used for cubic convolution and linear rectification, and the size becomes 8×8×512; after 2×2 maximum pooling, the size becomes 4×4×512; Fully connected linear rectification with two layers of 1×1×4096 and one layer of 1×1×1000 (three layers in total); output 1000 prediction results through the normalized exponential function; the first of the prediction results is the power grid line If the pole and the probability is greater than 98%, it is considered that the power grid pole is recognized; mark the area in the original image with a rectangular frame adjusted by length and width, and record the horizontal and vertical coordinates and width and height of the upper left corner of the rectangular frame; place the positions of the windows in order Move a certain area to avoid duplicate acquisitions.

The grid pole counting algorithm is used to record the total number of currently recognized grid poles, so that each grid pole is only recorded once after being identified, and the total number of currently recognized grid poles is recorded. Figure 4 shows the use of grid poles. A schematic diagram of the process of recording the total number of power grid poles currently identified by the counting algorithm. Referring to Figure 4, the specific process includes:

For each frame of image, the classifier will look for power grid poles in the original image in order from right to left and top to bottom;

Each time a power grid pole is found, the classifier returns the horizontal and vertical coordinates and width and height information of the upper left corner of the current power grid pole;

The search ends when the number of grid poles found reaches three or the abscissa of the vanishing point has been scanned;

If the power grid pole on the far right of the current frame is still within the image acquisition range in the next frame, the maximum abscissa scanned will be larger than the previous frame; The abscissa value will jump and become the second largest abscissa value. If the jumping process does not repeat for 5 consecutive frames, it is considered that a power grid pole has left the camera's field of view. At this time, the power grid line The rod count value is incremented by 1.

S5 determines the current position of the train according to the total number of current grid poles.

Store the corresponding relationship between the number of grid poles and the operating position in the database, and query the total number of current grid poles to determine the current position of the train according to the corresponding relationship between the number of grid poles and the operating position stored in the database.

This embodiment uses ten test videos recorded on the Beijing-Tianjin intercity line, including traction conditions, braking conditions, coasting conditions, cruising conditions, straight tracks, and turning tracks. For processing by the embodiment method, the train running image is collected by a high-speed camera installed at the head of the rail train running on the Beijing-Tianjin intercity line. rod to count. Using a computer with a main frequency of 2.0GHz CPU i7-8550U, 16G memory, and a graphics card GTX1060, install the pycharm integrated development environment, build the python3.7 development environment and the opencv+tensorflow framework. Based on the trained CNN-VGG16 classifier for power grid poles, a vanishing point detection program and a power grid pole counting program are added to process the collected video. After the processing is completed, the error value is recorded, and the final result shows that the positioning error is within 9m, which meets the practical requirements.

Those skilled in the art should understand that the above-mentioned application types of input boxes are only examples, and other existing or possible future application types of input boxes, if applicable to the embodiments of the present invention, should also be included within the protection scope of the present invention, and It is hereby incorporated by reference.

Those of ordinary skill in the art can understand that the accompanying drawing is only a schematic diagram of an embodiment, and the modules or processes in the accompanying drawing are not necessarily necessary to implement the present invention.

From the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solutions of the present invention can be embodied in the form of software products in essence or the parts that make contributions to the prior art. The computer software products can be stored in storage media, such as ROM/RAM, magnetic disks, etc. , CD, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited to this. Substitutions should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (7)

1. A train positioning method based on trackside power grid line pole counting is characterized by comprising the following steps:

aiming at a trackside power grid line pole, training a classifier by adopting a visual geometry group neural network classifier algorithm based on a convolutional neural network;

collecting a running video of the train through a high-speed camera arranged at the head of the train;

identifying the track line and the vanishing point of the running video by adopting an image identification algorithm, wherein the method comprises the following steps of:

filtering high-frequency components of the acquired image through a Gaussian filter;

performing edge detection by using a Canny algorithm, adding a trapezoidal window to a detection result, and filtering out a foreground;

carrying out Hough transform on the foreground to find straight lines, wherein longer straight lines are endowed with higher weight when calculating the slope average value;

distinguishing left and right tracks of the found straight line according to the positive and negative slopes;

respectively and iteratively calculating the difference between the slope of each straight line and the mean value of the slopes, and removing the straight lines with the difference value larger than a preset value;

respectively performing linear regression on the straight line set of the left and right track lines to obtain a final left and right track line, wherein the intersection point of the final left and right track lines is a vanishing point;

according to the trained classifier, identifying the power grid poles on the right side of the vanishing point, and recording the total number of the currently identified power grid poles by adopting a power grid pole counting algorithm;

and determining the current position of the train according to the total number of the current power grid poles.

2. The method of claim 1, wherein the visual geometry group neural network classifier based on convolutional neural network is composed of 5 convolutional layers, 3 fully-connected layers and normalized exponential output layers, wherein the convolutional layers use 3x3 convolutional kernels, the layers are separated by 2x 2 maximal pooling layers, and the activation units of all hidden layers use linear rectification function.

3. The method according to claim 1, wherein for the trackside power grid mast, training a classifier by using a visual geometry group neural network classifier algorithm based on a convolutional neural network comprises: the positive sample is a picture with the foreground of the trackside power grid pole, the negative sample is a picture only containing trackside power grid pole local information or not containing the trackside power grid pole, the positive sample picture is classified into a training set and a positive example test set, and the negative sample picture is classified into a negative example test set.

4. The method as claimed in claim 1, wherein the step of acquiring the running video of the train by the high-speed camera installed at the head of the train comprises the following steps: the high-speed camera is arranged in the middle of the train head.

5. The method of claim 1, wherein the step of identifying the grid poles to the right of the vanishing point and recording the total number of currently identified grid poles according to the trained classifier comprises the steps of:

sequentially taking out pictures containing complete lines and poles by using 112 multiplied by 3 windows from right to left and from top to bottom for the currently collected pictures, carrying out convolution twice and linear rectification on the pictures containing the complete lines and poles by 64 3 multiplied by 3 convolution kernels, wherein the size after the convolution is changed into 112 multiplied by 64; the pooling unit with the size of 2 multiplied by 2 is adopted for maximizing pooling, and the size after pooling is changed into 64 multiplied by 64; performing two times of convolution and linear rectification by 128 convolution kernels of 3 × 3, wherein the size of the obtained product is 64 × 64 × 128; making 2X 2 maximal pooling, and changing the size to 32X 128; carrying out three times of convolution and linear rectification by 256 convolution kernels of 3 × 3, wherein the size of the convolution kernels is changed into 32 × 32 × 256; making 2 × 2 maximal pooling, and changing the size to 16 × 16 × 256; carrying out three times of convolution and linear rectification by 512 convolution kernels of 3 × 3, wherein the size of the obtained product is changed into 16 × 16 × 512; making 2 × 2 maximal pooling, and changing the size to 8 × 8 × 512; carrying out three times of convolution and linear rectification by 512 convolution kernels of 3 × 3, wherein the size of the convolution kernels is 8 × 8 × 512; making 2X 2 maximal pooling, and changing the size to 4X 512; carrying out full-connection linear rectification with two layers of 1 multiplied by 4096 and one layer of 1 multiplied by 1000; outputting 1000 prediction results through a normalized exponential function; the first one of the prediction results is a power grid pole, and the probability is greater than 98%, the power grid pole is considered to be identified; marking the identified area of the power grid pole on an original picture by using a rectangular frame with the length and width adjusted, and recording the horizontal and vertical coordinates and the width and height of the upper left corner of the rectangular frame; the position of the window is sequentially moved by a certain area.

6. The method of claim 1, wherein said recording the total number of currently identified grid poles comprises:

for each frame of image, the classifier searches for the power grid poles in the original image from right to left and from top to bottom;

when a power grid pole is found, the classifier returns the horizontal and vertical coordinates and width and height information of the upper left corner of the current power grid pole;

finishing searching when the number of the found power grid poles reaches three or the abscissa of the vanishing point is scanned;

if the power grid pole at the rightmost side of the current frame is still in the image acquisition range of the next frame, the scanned maximum abscissa is larger than that of the previous frame; when the rightmost power grid pole disappears in the camera view, the maximum abscissa value jumps to be the original second largest abscissa value, if the jumping process of 5 continuous frames does not repeat, a power grid pole is considered to leave the camera view, and the power grid pole count value is increased by 1.

7. The method of claim 1, wherein determining the current location of the train based on the total number of current grid poles comprises:

and storing the corresponding relation of the number of the electric network poles and the operation position through a database, and inquiring the total number of the current electric network poles according to the corresponding relation of the number of the electric network poles and the operation position stored in the database to determine the current position of the train.

Images