CN107590441A - A kind of pantograph goat's horn on-line measuring device and method based on image procossing - Google Patents

Abstract

The invention discloses an image processing-based online pantograph horn detection device and method. The device includes three parts: an image acquisition module, a data transmission module, and an image processing module. The data transmission module transmits the image data collected by the image acquisition module to the image processing module. The image processing module utilizes the active shape Model learning algorithm to judge the presence or absence of horns. The orientation is: firstly, preprocess the collected images, including filtering, image enhancement, and edge detection; then, mark the feature points of the horn image and build an ASM model; then perform initial area positioning on the horn; finally, according to the constructed horn Model and initial area positioning, precise positioning of the horns, and judging whether the horns are missing. The invention has the advantages of convenient structure layout, stable system and high-precision online non-contact measurement.

Description

A device and method for on-line detection of pantograph horns based on image processing

technical field

The invention relates to the technical field of traffic safety engineering, in particular to an image processing-based on-line pantograph horn detection device and method.

Background technique

The pantograph slide is the only part in contact between the train and the catenary, and is the most important power-taking device for the entire train power supply system. During the operation of the train, the pantograph slide plate is constantly in contact with the catenary, causing wear and tear; if the wear is serious, the catenary wire clip will be in the crack or deep groove on the surface of the slide plate, causing mechanical collision between the bow head and the catenary, resulting in power failure. Bow malfunction. Therefore, accurate detection and identification of pantograph horns in a timely and effective manner is an important measure to ensure pantograph-catenary safety. It can not only prevent various safety accidents, but also provide a basis for the maintenance of urban rail vehicle pantographs.

Domestic urban rail transit is still in the development period, and the pantograph detection method is still dominated by manual detection by climbing to the top. Manual detection requires the train to return to the depot and the catenary is powered off. The detection efficiency is low, the detection accuracy is poor, and it is not conducive to the maintenance and maintenance of the pantograph. The online detection system of the pantograph is still in its infancy, and the basic principles currently used include ultrasonic ranging, CCD imaging, image processing and image recognition.

After analyzing the actual on-site installation and testing environment, Xie Li et al. took multi-angle shots of the pantograph by using three sets of cameras and one set of cameras. Although this method can realize online diagnosis of pantograph slide faults, the system has strict requirements on the installation location of key hardware, and the algorithm adopted by the system cannot effectively eliminate the interference of the surrounding environment on image quality. When the color of the surrounding background environment is close to the color of the skateboard, the upper and lower edges of the pantograph skateboard cannot be correctly extracted, resulting in low detection accuracy, which does not meet the actual application in the field.

Contents of the invention

The object of the present invention is to provide a pantograph horn online detection device and method based on image processing with convenient structure layout and stable system, so as to realize high-precision online non-contact measurement.

The technical solution for realizing the object of the present invention is: a pantograph horn online detection device based on image processing, including an image acquisition module, a data transmission module and an image processing module, wherein:

The image acquisition module includes a first wheel axis position sensor, a first photoelectric sensor, a supplementary light device, a camera module, a second photoelectric sensor and a second wheel axis position sensor arranged sequentially in the direction of travel of the train; the camera There are two groups of modules, and each group contains 2 area array cameras called half-bow cameras, which are installed on the upper side of the roof and set a 30-degree overlooking angle to observe the state of the roof and the pantograph; the two area array cameras are respectively from The images of the pantograph skateboard are collected in left and right directions; two groups of 4 area array cameras respectively photograph the front left half bow of the pantograph skateboard, the front right half bow, the rear left half bow, the rear right left half bow, 4 area array cameras There is a margin to photograph the center area of the pantograph slide;

The data transmission module is used to transmit the image data collected by the image acquisition module to the image processing module;

The image processing module is used to process the received image data, establish an active shape model through learning of horn samples, and use an active shape model learning algorithm for the pantograph horn images captured in real time in combination with the initial positioning of the pantograph horns To judge the presence or absence of horns.

Further, in the image acquisition module, when the first wheel axle position sensor detects the first wheel of the train, it indicates that the train enters the detection area, and the first and second photoelectric sensors are turned on at the same time; When the bow enters the pantograph detection area, turn on the supplementary light equipment to supplement the light in the lighting area, so that the area lighting meets the requirements for taking pictures, and at the same time turn on the camera module to take pictures; when the second photoelectric sensor detects that the pantograph leaves the pantograph detection area , turn off the camera module; when the second wheel shaft position sensor detects the 24th wheel, it indicates that the train has left the detection area, and the image acquisition device and lighting equipment in the image acquisition module are turned off.

A method for on-line detection of pantograph horns based on image processing, comprising the following steps:

Step 1, image acquisition: take pictures by the high-speed camera in the image acquisition module to obtain the original image;

Step 2, image preprocessing: filtering, image enhancement, and edge detection are performed on the image;

Step 3, ASM construction method of croissant: including croissant learning sample calibration and ASM training;

Step 4, preliminary positioning of the goat horn area: perform initial positioning of the goat horn area;

Step 5, detection and recognition of horns: Combine the active shape model learning algorithm to match the horns, and use the single-resolution search algorithm to match the shape of the horns, and judge whether the horns are missing in the initial positioning area.

Further, the croissant ASM construction method described in step 3 includes croissant learning sample calibration, ASM training, specifically as follows:

(3.1) Claw learning sample calibration

After image preprocessing, the boundary points and corner points of the horns were selected as feature points, and the feature points of the horns were marked manually; in the process of marking, the number of feature points of each horn image was required to be consistent and corresponded to each other, and used PDM describes the shape of the horn, that is, the shape of the horn image i is mathematically expressed as:

Wherein, N is the total number of feature points of the horn image;

The image learning sample set of croissants is expressed as:

Wherein, M is the total number of horn images;

(2) ASM training

The first step is to align the feature points. The specific steps are as follows:

a) Translate, rotate, and scale the croissant shape x ⁱ , i=1, 2, 3,..., M one by one to align with the shape x ¹ , so as to obtain the transformed shape set

b) averaging the transformed croissant images to obtain the average shape m:

in,

c) Translate, rotate, and scale the average shape m, and alignment;

d) will Perform translation, rotation, and scaling transformations, and then align and match with the average shape m;

e) If the average shape converges, then stop, otherwise go to step b);

The basis for judging convergence in the above step e is to minimize the sum of the squares of the difference between each horn shape after alignment and the average shape, that is, to find the transformation T _i so that the following formula is minimum:

∑|mT _i ( _xi )| ² (4)

The alignment of horn images is described as: Take two horn shapes as an example, each shape has N coordinate pairs:

First define the transformation matrix T, T is composed of 4 parameters, which are the rotation angle θ, the scale s and the translation vector (t _x , t _y ), to transform the shape x ² :

Assume

Align ^x2 with x1 using the T ^- transform, and the optimal transformation is obtained by minimizing equation (4):

E＝[x ¹ -Rx ² -(t _x ,t _y ) ^T ] ^T (9)

By calculating the partial differential of E to the unknown parameters θ, s, t _x , _ty , and making the differential equation zero, the transformation matrix T is obtained by solving;

The second step, ASM establishment

M training shapes are obtained after alignment processing Each shape is given by N pairs of coordinates: The average shape is set to: Then the covariance matrix is:

Among them, S is a 2N×2N matrix;

The variation of the training shape in certain directions is obtained through the eigenvectors of the covariance matrix S, i.e. solving the linear equation:

Sp _k =λ _k p _k ,k=1,2,3,...,2N (11)

Wherein, the eigenvector of S is P, and P is expressed as: P=(p ₁ ,p ₂ ,...,p _2N );

For any vector X, there exists a shape model parameter b such that:

make:

This gives an estimate of the shape:

The vector b _t defines a set of variable model parameters, and different b _t can fit different shapes;

Since the variance of _{bi on the training set is related to the eigenvalue λ i , bi} must satisfy the following formula:

Further, the preliminary positioning of the sheep horn region described in step 4 is as follows:

(1) The collected pantograph horns are distributed on the left and right sides of the image, with the intersection point as the feature point, and the left and right half-bow images of the left and right half-bow images are respectively searched in the direction of the left and right extension lines of the intersection point;

(2) The area contains at least two straight lines l ₁ and l ₂ , and the angle between the two straight lines is described by the straight line method with between the two if satisfied Then the area is the area to be determined; among them, with are the inclination angles of straight lines l ₁ and l ₂ respectively;

After locating the area to be determined by the horns, since the input image in the horns detection algorithm is an edge image, on this basis, the edge line segment in the search area is directly compressed for curve data, and then the length of the edge line segment is less than the set threshold. line segment; since the directions of the edge line segments of the left and right horns are in the range of 35° to 55° and 125° to 145° respectively, the Hough transform is used to detect the straight line, and the straight line segment whose angle range is in the above two intervals is searched according to the result, and then the The area of the inclined edge line segment is used as the initial positioning area of the horns.

Further, the horn detection and identification described in step 5 are as follows:

(1) According to the average shape in the horn active shape model and the initial position of the horn, initialize the horn shape, expressed as follows:

(2) Search along the boundary normal direction at each marker point in the initial positioning area of the horns, and then obtain the pixel point with the largest gradient, mark this point as the best target point, and move the marker point to the best target point, If no new target point is found, the position of the marked point will not move;

(3) After moving the above mark points, the shape will change, and there is a displacement vector between the changed shape and the initial horn shape According to formula (15), it can be seen that the expression after displacement is:

Deduced from formula (15) and formula (18):

(4) Repeat steps (2) to (3). If p _2N , p _2N-1 , ... is less than the threshold σ and σ tends to zero after the set number of repetitions, it is judged that the horn exists, otherwise it is judged that the horn in the image missing.

Compared with the prior art, the present invention has the following remarkable advantages: (1) convenient structure layout, stable system, and high-precision online non-contact measurement; (2) pantograph horns can be measured during train operation The detection not only ensures the safety of operation, but also improves the detection efficiency.

The present invention will be described in further detail below in conjunction with the accompanying drawings.

Description of drawings

Fig. 1 is a system overall structure diagram of the pantograph horn on-line detection device based on image processing in the present invention.

Fig. 2 is a flow chart of the online pantograph horn detection method based on image processing in the present invention.

Fig. 3 is the croiss part figure that collects in the present invention, and wherein (a)～(l) is 12 croiss part figure that collect respectively.

Fig. 4 is a schematic diagram of marking characteristic points of sheep's horns in the present invention.

Fig. 5 is a schematic diagram of the position of the sheep's horn in the image in the present invention, wherein (a) is a schematic diagram of the left half-arch, and (b) is a schematic diagram of the right half-arch.

Fig. 6 is a display diagram of the detection result of horns in the present invention.

Fig. 7 is a diagram of misdetection results of horns in the present invention, wherein (a) is a diagram of misdetection results of horns with uneven fill light, and (b) is a diagram of misdetection results of horns with excessive angle of inclination.

detailed description

The invention is an image processing-based on-line detection device and method for pantograph horns. Firstly, filtering and contrast enhancement processing are performed on the collected images of the pantograph slide plate; The model (Active Shape Model, ASM) algorithm accurately locates the horn image to be recognized. If the positioning is successful, it is judged that the horn exists, otherwise, it is judged that the horn is missing.

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments.

In conjunction with Fig. 1, the pantograph horn online detection device based on image processing of the present invention includes an image acquisition module, a data transmission module and an image processing module, wherein:

The image acquisition module includes a first wheel axis position sensor, a first photoelectric sensor, a supplementary light device, a camera module, a second photoelectric sensor and a second wheel axis position sensor arranged sequentially in the direction of travel of the train; the camera There are two groups of modules, and each group contains 2 area array cameras called half-bow cameras, which are installed on the upper side of the roof and set a 30-degree overlooking angle to observe the state of the roof and the pantograph; the two area array cameras are respectively from The images of the pantograph skateboard are collected in left and right directions; two groups of 4 area array cameras respectively photograph the front left half bow of the pantograph skateboard, the front right half bow, the rear left half bow, the rear right left half bow, 4 area array cameras There is a margin to photograph the center area of the pantograph slide;

The data transmission module is used to transmit the image data collected by the image acquisition module to the image processing module;

The image processing module is used to process the received image data, establish an active shape model through learning of horn samples, and use an active shape model learning algorithm for the pantograph horn images captured in real time in combination with the initial positioning of the pantograph horns To judge the presence or absence of horns.

In the image acquisition module, when the first wheel axis position sensor detects the first wheel of the train, it indicates that the train enters the detection area, and the first and second photoelectric sensors are turned on at the same time; when the first photoelectric sensor detects that the pantograph enters the In the pantograph detection area, turn on the supplementary light equipment to supplement the light in the lighting area, so that the area lighting meets the requirements for taking pictures, and at the same time turn on the camera module to take pictures; when the second photoelectric sensor detects that the pantograph has left the pantograph detection area, turn off the camera Module; when the second wheel shaft position sensor detects the 24th wheel, it indicates that the train has left the detection area, and the image acquisition device and lighting equipment in the image acquisition module are turned off.

In conjunction with Fig. 2, the pantograph horn online detection method based on image processing of the present invention comprises the following steps:

Step 1, image acquisition: take pictures by the high-speed camera in the image acquisition module to obtain the original image;

Arrange system devices and obtain original images: Figure 1 is the overall design of the system. The pantograph horn detection system uses dynamic non-contact image measurement technology to detect the condition of the pantograph horn. It is mainly composed of on-site detection equipment and control, support and processing equipment located in the equipment room. On-site detection equipment includes image acquisition module, data transmission module and image processing module.

The image acquisition module is used to acquire pantograph images. When the wheel shaft position sensor 1 detects the first wheel of the train, it indicates that the train has entered the detection area, and the photoelectric sensor is turned on simultaneously. When the photoelectric sensor 1 detects that the pantograph enters the pantograph detection area, the supplementary light device is turned on to supplement the light in the lighting area so that the area lighting meets the requirements for taking pictures, and at the same time, the high-speed camera is turned on to take pictures. When the photoelectric sensor 2 detects that the pantograph leaves the pantograph detection area, the high-speed camera is turned off. When the wheel shaft position sensor 2 detects the 24th wheel, it indicates that the train has left the detection area, and the image acquisition equipment and lighting equipment are turned off.

The data transmission module is used to transmit the collected image data to the data processing module. In this solution, GigE network cable is used for transmission.

The image processing module is used to process the images collected by the image collection module. Firstly, image segmentation is performed on the collected images to remove useless information areas and retain useful information areas, and then image denoising, image enhancement and other image preprocessing are performed on the useful information areas after image segmentation, and then the active shape is established through croissant sample learning Combined with the initial positioning of the pantograph horns, the active shape model learning algorithm is used to judge the presence or absence of the pantograph horns images captured in real time. If the system detects a pantograph failure, it will give an audible alarm and submit the cause of the failure, so as to repair the pantograph failure in time to ensure the safety of pantograph-catenary

Step 2, image preprocessing: filtering, image enhancement, and edge detection are performed on the image;

Step 3, ASM construction method of croissant: including croissant learning sample calibration and ASM training;

Claw ASM construction method:

ASM is based on the point distribution model (Point Distribution Model, PDM). The statistical information of the sample feature point distribution is obtained through the training image sample, and the change direction of the undetermined target is allowed to realize the position of the corresponding feature point on the target image. Search.

The specific steps of the Chow Horn ASM construction method are as follows:

(3.1) Claw learning sample calibration

Figure 3 is an image set of horn parts, where (a) to (l) are 12 pictures of horn parts collected respectively. After image preprocessing, the boundary points and corner points of the horns were selected as feature points, and the feature points of horns were marked manually. In the marking process, the number of feature points of each horn image is required to be consistent and correspond to each other, and the shape of the horn is described by PDM, that is, the shape of the horn image i can be mathematically expressed as:

Among them, N is the total number of feature points of the horn image. The learning sample set of croissant images can be expressed as:

Among them, M is the total number of horn images. The schematic diagram of the sheep horn feature point marking is shown in Figure 4.

(2) ASM training

ASM training is divided into two steps, the first step is to align the feature points, and the second step is to establish ASM. Feature point alignment is shape normalization, the purpose of which is to eliminate the non-shape interference of horns caused by external factors such as angle, distance, and pose transformation in the image, so as to make the model more effective. Generally, the generalized procrustes analysis (GPA) method was used for normalization. This method is to align a series of point distribution models to the same PDM without changing the PDM through appropriate translation, rotation, and scaling transformations, so that the acquired original data is no longer in a messy state, so as to reduce non-shape factors. interference. The specific steps of alignment are as follows:

a. Translate, rotate, and scale the croissant shape x ⁱ , i=1, 2, 3,..., M one by one to align with the shape x ¹ , so as to obtain the transformed shape set

b. Calculate the transformed horn image for averaging:

in,

c. Translate, rotate, and scale the average shape m, and alignment;

d. Will Perform translation, rotation, and scaling transformations, and then align and match with the average shape m;

e. If the average shape converges, stop, otherwise go to step b.

The basis for judging convergence in the above step e is to minimize the sum of the squares of the difference between each horn shape after alignment and the average shape, that is, to find the transformation T _i so that the following formula is minimum:

∑|mT _i ( _xi )| ² (4)

Alignment of horn images can be described as: Take two horn shapes as an example, each shape has N coordinate pairs:

First define the transformation matrix T, T is composed of 4 parameters, which are the rotation angle θ, the scale s and the translation vector (t _x , t _y ), to transform the shape x ² :

Assume

Using the T transformation to align x ² with x ¹ , the optimal transformation can be obtained by minimizing equation (4.37):

E＝[x ¹ -Rx ² -(t _x ,t _y ) ^T ] ^T (9)

By calculating the partial differential of E to the unknown parameters θ, s, t _x , _ty , and making the differential equation zero, the transformation matrix T is obtained.

After normalizing the shape of the horn image, an ASM can be built. After alignment processing, M training shapes can be obtained Each shape is given by N pairs of coordinates: The average shape can be set to: Then the covariance matrix is:

Among them, S is a 2N×2N matrix.

The variation of the training shape in certain directions can describe important properties of the horn shape, and these properties can be obtained through the eigenvectors of the covariance matrix S, that is, solving the linear equation:

Sp _k =λ _k p _k ,k=1,2,3,...,2N (11)

Wherein, the feature vector of S is P, and P can be expressed as: P=(p ₁ ,p ₂ ,...,p _2N ).

For any vector X, there exists a shape model parameter b such that:

Vectors with large eigenvalues are used to describe the direction in which the training shape changes greatly. When describing the deviation between a reasonable shape and the average shape, p _2N , p _2N-1 ,... can be ignored, so it can be ordered:

This gives an estimate of the shape:

If X is of reasonable shape relative to the training set, then for sufficiently large t the estimate fits the true shape well.

The vector b _t defines a set of variable model parameters, and different b _t can fit different shapes. Since the variance of b _i on the training set is related to the eigenvalue λ _i , for a better shape, b _i should satisfy the following formula:

Step 4, preliminary positioning of the goat horn area: perform initial positioning of the goat horn area;

Since the pantograph horns collected by the four CCD cameras are distributed on the left and right sides of the image, the schematic diagrams of their positions are shown in Figure 5(a)-(b). The search steps for the horn area are as follows:

(1) Taking the intersection point as the feature point, the left half-bow and right half-bow images in Fig. 5 take the direction of the left and right extension lines of the intersection point as the search direction respectively;

(2) The area contains at least two straight lines l ₁ and l ₂ , and the angle between the two straight lines is described by the straight line method with between the two if satisfied Then this area is the area to be determined. in, with are the inclination angles of straight lines _l1 and _l2 , respectively.

After locating the possible area of the horns, since the input image in the horns detection algorithm is an edge image, it is possible to directly compress the edge segments in the search area on the basis of the curve data, and then remove the shorter segments of the edge segments. Since the directions of the edge segments of the left and right horns are in the range of 35°-55° and 125°-145° respectively, the Hough transform is used to detect the straight line and the straight line segment whose angle range is within the above two intervals is searched according to the result, and then the inclined edge line segment The region is used as the initial positioning region of the horns.

Step 5, detection and recognition of horns: Combine the active shape model learning algorithm to match the horns, and use the single-resolution search algorithm to match the shape of the horns, and judge whether the horns are missing in the initial positioning area.

Combined with the active shape model learning algorithm, the horns can be accurately matched, and the single-resolution search algorithm is used to accurately match the shape of the horns, and then judge whether the horns actually exist in the initial positioning area. The specific steps of the algorithm are as follows:

(1) According to the average shape in the horn active shape model and the initial position of the horn, initialize the horn shape, expressed as follows:

(2) Search along the boundary normal direction at each marker point initially positioned in the horn shape, and then obtain the pixel point with the largest gradient, mark this point as the best target position, and move the marker point to the best target point, If no new target point is found, the position of the marked point will not move;

(3) After moving the above mark points, the shape will change, and there is a displacement vector between the changed shape and the initial horn shape According to formula (15), it can be seen that the expression after displacement is:

It can be deduced from formula (15) and formula (18):

(4) Repeat steps (2) and (3). If the attitude parameters p _2N , p _2N-1 ,... are negligible after several repetitions, it is judged that the horn exists, otherwise it is judged that the horn is missing in the image.

Using the above algorithm to identify and locate the horn image, the correct detection result of the horn is shown in Figure 6.

Example 1

In this experiment, 396 images of horns are selected as the training set. The horns in these images have different poses and exposure levels. The change of horn posture is about 25°, and the image brightness includes insufficient fill light, excessive fill light, uneven fill light, etc. The specific proportion of the horn image quality used for detection is shown in Table 1. Before training the samples, 396 images are manually calibrated, and each image is calibrated with 28 feature points, so as to establish the horn ASM model.

Table 1 Statistical Table of Image Types

In this experiment, 264 pictures were selected as the test set to detect the existence of horns, and the images in the training set and the test set were not repeated. In addition to normal horn images, the test set images also include partially or completely missing horns, and images with uneven illumination. The specific proportions of the test set images are shown in Table 2.

Table 2 Statistical table of test set image types

Using the horn detection algorithm proposed in this paper to detect the above 264 images of horns, the recognition time of each image is about 340ms, and there are actually 8 images in the test images that lack horns, and the test detects that there are no horns in 25 images. The statistical table of test results is shown in Table 3. Among them, 8 images with missing horns were correctly identified, while the images with horns in the remaining 17 images were judged as missing horns, and the false detection rate of horns was 6.4%. Second, the recognition accuracy of the horns for non-ideal images is 79.45%. Among them, the degree of inclination of the horns has a great influence on the detection accuracy of the horns. When the degree of inclination of the horns is large, the recognition accuracy of the horns is only 60%. In addition, the effect of supplementary light also has a great impact on the accuracy of horn recognition. It can be seen from Table 3 that the accuracy of horn recognition is low when the effect of image supplementary light is not ideal.

Table 3 Statistical table of test results

Since this patent uses an algorithm based on the active shape model to accurately locate the horns, theoretically, the light should have little effect on the detection results, but it can be seen from the table that the detection accuracy is relatively low when the light supplement effect is not ideal. This is mainly because the poor lighting effect tends to make the gray distribution of the horn image uneven. Secondly, the shadow cast by the light will strengthen or weaken the original horn characteristics, thereby reducing the accuracy of horn recognition. In order to analyze the specific influence of the illumination and angle of inclination of the horns on the recognition of the horns, two images that were misdetected in the experiment were selected, and the images of the misdetection results of the horns are shown in Figure 7. By analyzing Figure 7(a), it can be seen that when the system collects images, the images in most areas of the pantograph horns are dark due to uneven light supplementation. Although the image is enhanced, it cannot effectively compensate for the impact caused by uneven illumination. When the horn localization method based on the active shape model establishes the local gray-scale structure model, the extracted gray-scale value is the derivative value of the feature point along its contour normal to the gray-scale change, rather than the absolute value, which can overcome the external environment to a certain extent. The effect of light on positioning accuracy. However, due to the poor quality of the test image and the uneven light, some information is missing, which leads to errors in the horn detection. Figure 7(b) is due to the excessive inclination angle of the sheep’s horns, which causes false detection. Although the sheep’s horns positioning method based on the active shape model has certain robustness to the influence of different attitudes, if the angle of the sheep’s horns is beyond a certain range, it cannot be detected. Find the region that matches the feature point in the image, and then the horn detection fails.

Claims (6)

1. a pantograph horn online detection device based on image processing, is characterized in that, comprises image acquisition module, data transmission module and image processing module, wherein: The image acquisition module includes a first wheel axis position sensor, a first photoelectric sensor, a supplementary light device, a camera module, a second photoelectric sensor and a second wheel axis position sensor arranged sequentially in the direction of travel of the train; the camera There are two groups of modules, and each group contains 2 area array cameras called half-bow cameras, which are installed on the upper side of the roof and set a 30-degree overlooking angle to observe the state of the roof and the pantograph; the two area array cameras are respectively from The images of the pantograph skateboard are collected in left and right directions; two groups of 4 area array cameras respectively photograph the front left half bow of the pantograph skateboard, the front right half bow, the rear left half bow, the rear right left half bow, 4 area array cameras There is a margin to photograph the center area of the pantograph slide; The data transmission module is used to transmit the image data collected by the image acquisition module to the image processing module; The image processing module is used to process the received image data, establish an active shape model through learning of horn samples, and use an active shape model learning algorithm for the pantograph horn images captured in real time in combination with the initial positioning of the pantograph horns To judge the presence or absence of horns. 2. The pantograph horn online detection device based on image processing according to claim 1, characterized in that, in the image acquisition module, when the first wheel shaft position sensor detects the first wheel of the train, it indicates that the train Enter the detection area, turn on the first and second photoelectric sensors at the same time; when the first photoelectric sensor detects that the pantograph enters the pantograph detection area, turn on the supplementary light device to supplement the light in the lighting area, so that the area lighting meets the requirements for taking pictures, and turn on the The camera module takes pictures; when the second photoelectric sensor detects that the pantograph has left the pantograph detection area, close the camera module; when the second wheel axle position sensor detects the 24th wheel, it indicates that the train has left the detection area, Turn off the image acquisition equipment and lighting equipment in the image acquisition module. 3. a pantograph horn online detection method based on image processing, is characterized in that, comprises the following steps: Step 1, image acquisition: take pictures by the high-speed camera in the image acquisition module to obtain the original image; Step 2, image preprocessing: filtering, image enhancement, and edge detection are performed on the image; Step 3, ASM construction method of croissant: including croissant learning sample calibration and ASM training; Step 4, preliminary positioning of the goat horn area: perform initial positioning of the goat horn area; Step 5, detection and recognition of horns: Combine the active shape model learning algorithm to match the horns, and use the single-resolution search algorithm to match the shape of the horns, and judge whether the horns are missing in the initial positioning area. 4. the pantograph horn online detection method based on image processing according to claim 3, is characterized in that, the horn ASM construction method described in step 3 comprises horn learning sample calibration, ASM training, specifically as follows: (3.1) Claw learning sample calibration After image preprocessing, the boundary points and corner points of the horns were selected as feature points, and the feature points of the horns were marked manually; in the process of marking, the number of feature points of each horn image was required to be consistent and corresponded to each other, and used PDM describes the shape of the horn, that is, the shape of the horn image i is mathematically expressed as: <mrow><msup><mi>x</mi><mi>i</mi></msup><mo>=</mo><msup><mrow><mo>(</mo><msubsup><mi>x</mi><mn>1</mn><mi>i</mi></msubsup><mo>,</mo><msubsup><mi>y</mi><mn>1</mn><mi>i</mi></msubsup><mo>,</mo><msubsup><mi>x</mi><mn>2</mn><mi>i</mi></msubsup><mo>,</mo><msubsup><mi>y</mi><mn>2</mn><mi>i</mi></msubsup><mo>,</mo><mo>...</mo><mo>,</mo><msubsup><mi>x</mi><mi>N</mi><mi>i</mi></msubsup><mo>,</mo><msubsup><mi>y</mi><mi>N</mi><mi>i</mi></msubsup><mo>)</mo></mrow><mi>T</mi></msup><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>1</mn><mo>)</mo></mrow></mrow> Wherein, N is the total number of feature points of the horn image; The croissant image learning sample set is expressed as: <mrow><msup><mi>x</mi><mi>i</mi></msup><mo>=</mo><msup><mrow><mo>(</mo><msubsup><mi>x</mi><mn>1</mn><mi>i</mi></msubsup><mo>,</mo><msubsup><mi>y</mi><mn>1</mn><mi>i</mi></msubsup><mo>,</mo><msubsup><mi>x</mi><mn>2</mn><mi>i</mi></msubsup><mo>,</mo><msubsup><mi>y</mi><mn>2</mn><mi>i</mi></msubsup><mo>,</mo><mo>...</mo><mo>,</mo><msubsup><mi>x</mi><mi>N</mi><mi>i</mi></msubsup><mo>,</mo><msubsup><mi>y</mi><mi>N</mi><mi>i</mi></msubsup><mo>)</mo></mrow><mi>T</mi></msup><mo>,</mo><mi>i</mi><mo>=</mo><mn>1</mn><mo>,</mo><mn>2</mn><mo>,</mo><mn>3</mn><mo>,</mo><mo>...</mo><mo>,</mo><mi>M</mi><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>2</mn><mo>)</mo></mrow></mrow> Wherein, M is the total number of horn images; (2) ASM training The first step is to align the feature points. The specific steps are as follows: a) Translate, rotate, and scale the croissant shape x ⁱ , i=1, 2, 3,..., M one by one to align with the shape x ¹ , so as to obtain the transformed shape set b) averaging the transformed croissant images to obtain the average shape m: in, c) Translate, rotate, and scale the average shape m, and alignment; d) will Perform translation, rotation, and scaling transformations, and then align and match with the average shape m; e) If the average shape converges, then stop, otherwise go to step b); The basis for judging convergence in the above step e is to minimize the sum of the squares of the difference between each horn shape after alignment and the average shape, that is, to find the transformation T _i so that the following formula is minimum: ∑|mT _i ( _xi )| ² (4) The alignment of horn images is described as: Take two horn shapes as an example, each shape has N coordinate pairs: <mrow><msup><mi>x</mi><mn>1</mn></msup><mo>=</mo><msup><mrow><mo>(</mo><msubsup><mi>x</mi><mn>1</mn><mn>1</mn></msubsup><mo>,</mo><msubsup><mi>y</mi><mn>1</mn><mn>1</mn></msubsup><mo>,</mo><msubsup><mi>x</mi><mn>2</mn><mn>1</mn></msubsup><mo>,</mo><msubsup><mi>y</mi><mn>2</mn><mn>1</mn></msubsup><mo>,</mo><mo>...</mo><mo>,</mo><msubsup><mi>x</mi><mi>N</mi><mn>1</mn></msubsup><mo>,</mo><msubsup><mi>y</mi><mi>N</mi><mn>1</mn></msubsup><mo>)</mo></mrow><mi>T</mi></msup><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>5</mn><mo>)</mo></mrow></mrow> <mrow><msup><mi>x</mi><mn>2</mn></msup><mo>=</mo><msup><mrow><mo>(</mo><msubsup><mi>x</mi><mn>1</mn><mn>2</mn></msubsup><mo>,</mo><msubsup><mi>y</mi><mn>1</mn><mn>2</mn></msubsup><mo>,</mo><msubsup><mi>x</mi><mn>2</mn><mn>2</mn></msubsup><mo>,</mo><msubsup><mi>y</mi><mn>2</mn><mn>2</mn></msubsup><mo>,</mo><mo>...</mo><mo>,</mo><msubsup><mi>x</mi><mi>N</mi><mn>2</mn></msubsup><mo>,</mo><msubsup><mi>y</mi><mi>N</mi><mn>2</mn></msubsup><mo>)</mo></mrow><mi>T</mi></msup><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>6</mn><mo>)</mo></mrow></mrow> First define the transformation matrix T, T is composed of 4 parameters, which are the rotation angle θ, the scale s and the translation vector (t _x , t _y ), to transform the shape x ² : <mrow><mi>T</mi><mrow><mo>(</mo><msup><mi>x</mi><mn>2</mn></msup><mo>)</mo></mrow><mo>=</mo><mfenced open = "(" close = ")"><mtable><mtr><mtd><mrow><mi>s</mi><mi></mi><mi>c</mi><mi>o</mi><mi>s</mi><mi>&amp;theta;</mi></mrow></mtd><mtd><mrow><mo>-</mo><mi>s</mi><mi>i</mi><mi>n</mi><mi>&amp;theta;</mi></mrow></mtd></mtr><mtr><mtd><mrow><mi>s</mi><mi></mi><mi>s</mi><mi>i</mi><mi>n</mi><mi>&amp;theta;</mi></mrow></mtd><mtd><mrow><mi>s</mi><mi></mi><mi>c</mi><mi>o</mi><mi>s</mi><mi>&amp;theta;</mi></mrow></mtd></mtr></mtable></mfenced><mfenced open = "(" close = ")"><mtable><mtr><mtd><msubsup><mi>x</mi><mi>i</mi><mn>2</mn></msubsup></mtd></mtr><mtr><mtd><msubsup><mi>y</mi><mi>i</mi><mn>2</mn></msubsup></mtd></mtr></mtable></mfenced><mo>+</mo><mfenced open = "(" close = ")"><mtable><mtr><mtd><msub><mi>t</mi><mi>x</mi></msub></mtd></mtr><mtr><mtd><msub><mi>t</mi><mi>y</mi></msub></mtd></mtr></mtable></mfenced><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>7</mn><mo>)</mo></mrow></mrow> Assume <mrow><mi>R</mi><mo>=</mo><mfenced open = "(" close = ")"><mtable><mtr><mtd><mrow><mi>s</mi><mi></mi><mi>c</mi><mi>o</mi><mi>s</mi><mi>&amp;theta;</mi></mrow></mtd><mtd><mrow><mo>-</mo><mi>s</mi><mi>i</mi><mi>n</mi><mi>&amp;theta;</mi></mrow></mtd></mtr><mtr><mtd><mrow><mi>s</mi><mi></mi><mi>s</mi><mi>i</mi><mi>n</mi><mi>&amp;theta;</mi></mrow></mtd><mtd><mrow><mi>s</mi><mi></mi><mi>c</mi><mi>o</mi><mi>s</mi><mi>&amp;theta;</mi></mrow></mtd></mtr></mtable></mfenced><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>8</mn><mo>)</mo></mrow></mrow> Align ^x2 with x1 using the T ^- transform, and the optimal transformation is obtained by minimizing equation (4): E＝[x ¹ -Rx ² -(t _x ,t _y ) ^T ] ^T (9) By calculating the partial differential of E to the unknown parameters θ, s, t _x , _ty , and making the differential equation zero, the transformation matrix T is obtained by solving; The second step, ASM establishment M training shapes are obtained after alignment processing Each shape is given by N pairs of coordinates: The average shape is set to: Then the covariance matrix is: Among them, S is a 2N×2N matrix; The variation of the training shape in certain directions is obtained through the eigenvectors of the covariance matrix S, i.e. solving the linear equation: Sp _k =λ _k p _k ,k=1,2,3,...,2N (11) Wherein, the eigenvector of S is P, and P is expressed as: P=(p ₁ ,p ₂ ,...,p _2N ); For any vector X, there exists a shape model parameter b such that: <mrow><mi>x</mi><mo>=</mo><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mo>+</mo><mi>P</mi><mi>b</mi><mo>=</mo><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mo>+</mo><msub><mi>p</mi><mn>1</mn></msub><msub><mi>b</mi><mn>1</mn></msub><mo>+</mo><mo>...</mo><mo>+</mo><msub><mi>p</mi><mrow><mn>2</mn><mi>N</mi></mrow></msub><msub><mi>b</mi><mrow><mn>2</mn><mi>N</mi></mrow></msub><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>12</mo>mn><mo>)</mo></mrow></mrow> make: <mrow><mtable><mtr><mtd><mrow><msub><mi>P</mi><mi>t</mi></msub><mo>=</mo><mrow><mo>(</mo><mrow><msub><mi>p</mi><mn>1</mn></msub><mo>,</mo><msub><mi>p</mi><mn>2</mn></msub><mo>,</mo><msub><mi>p</mi><mn>3</mn></msub><mo>,</mo><mn>...</mn><mo>,</mo><msub><mi>p</mi><mi>t</mi></msub></mrow><mo>)</mo></mrow></mrow></mtd></mtr><mtr><mtd><mrow><msub><mi>b</mi><mi>t</mi></msub><mo>=</mo><msup><mrow><mo>(</mo><mrow><msub><mi>b</mi><mn>1</mn></msub>msub><mo>,</mo><msub><mi>b</mi><mn>2</mn></msub><mo>,</mo><msub><mi>b</mi>mi><mn>3</mn></msub><mo>,</mo><mn>...</mn><mo>,</mo><msub><mi>b</mi><mi>t</mi></msub></mrow><mo>)</mo></mrow><mi>T</mi></msup></mrow></mtd></mtr></mtable><mo>,</mo><mi>t</mi><mo>&amp;le;</mo><mn>2</mn><mi>N</mi><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>13</mn><mo>)</mo></mrow></mrow> This gives an estimate of the shape: <mrow><mi>x</mi><mo>&amp;ap;</mo><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mo>+</mo><msub><mi>P</mi><mi>t</mi></msub><msub><mi>b</mi><mi>t</mi></msub><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>14</mn><mo>)</mo></mrow></mrow> <mrow><msub><mi>b</mi><mi>t</mi></msub><mo>&amp;ap;</mo><msubsup><mi>P</mi><mi>t</mi><mi>T</mi></msubsup><mrow><mo>(</mo><mi>x</mi><mo>-</mo><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mo>)</mo></mrow><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>15</mn><mo>)</mo></mrow></mrow> The vector b _t defines a set of variable model parameters, and different b _t can fit different shapes; Since the variance of _{bi on the training set is related to the eigenvalue λ i , bi} must satisfy the following formula: <mrow><mo>-</mo><mn>3</mn><msqrt><msub><mi>&amp;lambda;</mi><mi>i</mi></msub></msqrt><mo>&amp;le;</mo><msub><mi>b</mi><mi>i</mi></msub><mo>&amp;le;</mo><mn>3</mn><msqrt><msub><mi>&amp;lambda;</mi><mi>i</mi></msub></msqrt><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>16</mn><mo>)</mo></mrow><mo>.</mo></mrow> 5. the pantograph horn online detection method based on image processing according to claim 3, is characterized in that, the preliminary positioning of the horn region described in step 4 is specifically as follows: (1) The collected pantograph horns are distributed on the left and right sides of the image, with the intersection point as the feature point, and the left and right half-bow images of the left and right half-bow images are respectively searched in the direction of the left and right extension lines of the intersection point; (2) The area contains at least two straight lines l ₁ and l ₂ , and the angle between the two straight lines is described by the straight line method with between the two if satisfied Then the area is the area to be determined; among them, with are the inclination angles of straight lines l ₁ and l ₂ respectively; After locating the area to be determined by the horns, since the input image in the horns detection algorithm is an edge image, on this basis, the edge line segment in the search area is directly compressed for curve data, and then the length of the edge line segment is less than the set threshold. line segment; since the directions of the edge line segments of the left and right horns are in the range of 35° to 55° and 125° to 145° respectively, the Hough transform is used to detect the straight line, and the straight line segment whose angle range is in the above two intervals is searched according to the result, and then the The area of the inclined edge line segment is used as the initial positioning area of the horns. 6. the pantograph horn online detection method based on image processing according to claim 3, is characterized in that, the horn detection and identification described in step 5 are specifically as follows: (1) According to the average shape in the horn active shape model and the initial position of the horn, initialize the horn shape, expressed as follows: (2) Search along the boundary normal direction at each marker point in the initial positioning area of the horns, and then obtain the pixel point with the largest gradient, mark this point as the best target point, and move the marker point to the best target point, If no new target point is found, the position of the marked point will not move; (3) After moving the above mark points, the shape will change, and there is a displacement vector between the changed shape and the initial horn shape According to formula (15), it can be seen that the expression after displacement is: <mrow><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mo>+</mo><mi>&amp;zeta;</mi><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mo>&amp;ap;</mo><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mo>+</mo><msub><mi>P</mi><mi>t</mi></msub><mrow><mo>(</mo><mrow><msub><mi>b</mi><mi>t</mi></msub><mo>+</mo><msub><mi>&amp;zeta;b</mi><mi>t</mi></msub></mrow><mo>)</mo></mrow><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>18</mn><mo>)</mo></mrow></mrow> Deduced from formula (15) and formula (18): <mrow><msub><mi>&amp;zeta;b</mi><mi>t</mi></msub><mo>&amp;ap;</mo><msubsup><mi>P</mi><mi>t</mi><mi>T</mi></msubsup><mi>&amp;zeta;</mi><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>19</mo>mn><mo>)</mo></mrow></mrow> (4) Repeat steps (2) to (3). If p _2N , p _2N-1 , ... is less than the threshold σ and σ tends to zero after the set number of repetitions, it is judged that the horn exists, otherwise it is judged that the horn in the image missing.