CN114067106B - Method, device and storage medium for pantograph deformation detection based on inter-frame comparison - Google Patents

Abstract

The invention discloses a pantograph deformation detection method, equipment and a storage medium based on inter-frame comparison, and relates to the technical field of image processing and image recognition. The method is not interfered by external illumination change, and is only related to the similarity of the gray level projection of the current image, the template matching characteristic and the historical image. For each frame of image, the contrast, the exposure and other external information of the image do not need to be paid much attention. According to the method, external interference information is eliminated, deformation information does not need to be considered too much, a large number of sample supports are not needed, the early-stage workload of detection is small, and the recognition rate is high.

Description

Method, device and storage medium for pantograph deformation detection based on inter-frame comparison

technical field

The invention relates to the technical field of image processing and image recognition, and more particularly to a method, device and storage medium for pantograph deformation detection based on frame comparison.

Background technique

The pantograph is an electrical device for electric traction locomotives to obtain electrical energy from the catenary, which is installed on the roof of the locomotive or motor vehicle. The pantograph can be divided into two types: single-arm bow and double-arm bow, both of which are composed of skateboard, upper frame, lower arm (lower frame for double-arm bow), bottom frame, bow spring, transmission cylinder, supporting insulator and other components. The diamond-shaped pantograph, also known as the diamond pantograph, was very common in the past, but was gradually eliminated due to the high maintenance cost and the tendency to break the catenary in the event of failure. In recent years, single-arm bows are mostly used. The smoothness of the load current passing through the contact surface of the contact wire and the pantograph sliding plate is related to the contact pressure, transition resistance, and contact area between the sliding plate and the contact wire, and depends on the interaction between the pantograph and the catenary.

During the operation of the locomotive or motor train, if the hard points and other defects on the catenary power supply line cause the pantograph to collide when it passes at high speed, the pantograph will shake violently at light level, and the pantograph will be deformed or even fall off in severe cases. The pantograph sliding plate monitoring system is used for video monitoring of special sections and sections of the catenary, such as: station throat area, key tunnel entrance, line fork and phase-splitting links. The high-definition video images of each section are transmitted to the Power supply operation management department. Install video surveillance systems at high-speed railway stations to monitor the pantograph status of operating EMUs or electric locomotives, especially the status of pantograph skateboards.

In the prior art, the Chinese patent with the application number CN202110401700.1 discloses a method for detecting the roof maintenance of an electric locomotive, which utilizes an image recognition technology based on an anthropomorphic multi-layer neural network deep learning algorithm to extract an image of a carbon skateboard. The features are trained by the artificial intelligence recognition platform to become the recognition algorithm library, which can automatically detect and identify the surface cracks and defects on the carbon skateboard. This kind of deep learning method trains the corresponding model and recognizes the deformation of the pantograph. However, due to the unknown state, size, and location of the deformation, it is difficult to obtain the training samples of the neural network model in advance, resulting in the recognition rate and false alarm rate of the neural network model. not ideal.

SUMMARY OF THE INVENTION

In order to overcome the above-mentioned defects in the prior art, the purpose of the present invention is to provide a pantograph deformation detection method, device and storage medium based on inter-frame comparison. Similarity and template matching feature similarity calculation, and according to the total similarity calculation result to determine whether the pantograph is deformed, this method is not disturbed by external light changes, only the grayscale projection of the current image and template matching features and historical images related to the similarity.

In order to achieve the above purpose, the technical scheme adopted in the present invention:

A pantograph deformation detection method based on frame comparison, comprising the following steps:

S1, obtain vehicle number information, and search for the positioning template image and historical frame ROI sample image of the normal pantograph of the model corresponding to the current vehicle number from the sample image set according to the vehicle number information; the positioning template image includes a center positioning template and more than one ROI positioning area; the position of the ROI sample image of the historical frame is consistent with the ROI positioning area.

S2: Acquire an image to be inspected, perform template matching between the image to be inspected and the center positioning template, generate a mapping matrix, and transform the image to be inspected into a shape position consistent with the positioning template image according to the mapping matrix.

S3, according to the positional relationship between the center positioning template and the ROI positioning area, the ROI target image is intercepted from the transformed image to be inspected, and the grayscale projection feature similarity and template matching feature similarity between the historical frame ROI sample image and the ROI target image at the corresponding position are calculated. Spend.

S4, calculate the total similarity of a single ROI according to the similarity of the grayscale projection feature and the similarity of the template matching feature. If the total similarity meets the preset threshold, it is judged to be normal, otherwise it is deformed; traverse all ROI areas, if any ROI If the similarity determination result is abnormal, it is determined that the pantograph deformation occurs in the current image to be inspected.

Further, the first method of obtaining the ROI sample image of the historical frame is: obtaining a historical normal pantograph image, performing template matching between the normal pantograph image and the center positioning template, generating a mapping matrix, and according to the The mapping matrix transforms the normal pantograph image into a shape position consistent with the positioning template image; according to the positional relationship between the center positioning template and the ROI positioning area, the historical frame ROI sample image is intercepted from the transformed normal pantograph image, and the The ROI sample image is saved according to the corresponding ROI positioning area;

And/or Manner 2: Save the ROI target image of the pantograph image detection process previously determined to be normal as the ROI sample image of the corresponding area.

Further, the grayscale projection feature is a normalized grayscale projection gradient feature, and the extraction method includes the following steps:

Calculate the grayscale mean value of each line in the horizontal direction of the image, and calculate the grayscale projection data of the image according to the grayscale mean value;

Normalization processing and gradient transformation are sequentially performed on the grayscale projection data to obtain normalized grayscale projection gradient features.

Further, the template matching in S4 adopts shape-based template matching, and the adopted features are the position of the edge point and the gradient direction of the edge point, and the method for extracting the gradient direction feature of the edge point includes the following steps:

Extract the edges of the two images to be matched, calculate the gradients in the x and y directions of the images, and then calculate the total gradient value and gradient direction of the edge points according to the gradients in the x and y directions.

， Further, the template matching feature similarity calculation process is: using a sliding window method to perform image search, and calculating a matching score in the window, where the matching score is the direction of the template edge point and the corresponding point corresponding to the area to be matched. The cosine of the angle between the vectors:

,

Among them, score is the matching score, p _i is the cosine value of the included angle, and n is the number of points;

，

,

Wherein A _j and B _j are the elements of the direction vector of the corresponding point of the template and the area to be matched;

The above direction vector is expressed as:

，

,

为梯度角。 in

is the gradient angle.

Further, the determination of whether the pantograph is deformed according to the total similarity calculation result is that any of several ROI areas on the pantograph image to be detected is deformed, and it is determined that the pantograph is deformed; if the pantograph is deformed; If there is no deformation in several ROI areas on the pantograph image, it is determined that the pantograph is not deformed.

Further, the sample images are provided with six ROI regions of the pantograph carbon slide plate and the balance rod, and each ROI region saves multiple ROI sample images. When calculating the similarity, each ROI target image is associated with multiple Calculate the total similarity of the ROI sample images, and then take multiple averages of the total similarity as the criterion.

A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the processor executes the steps in the pantograph deformation detection method.

A computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements steps in a pantograph deformation detection method.

Beneficial effects of the present invention:

1. The method for detecting the deformation of the pantograph provided by the present invention calculates the similarity of grayscale projection features and the similarity of template matching features between the image to be detected and the historical frame template, and determines whether the pantograph is not based on the calculation result of the total weighted similarity. deformed. The method is not disturbed by changes in external illumination, and is only related to the grayscale projection of the current image, the similarity between template matching features and historical images. For each frame of image, there is no need to pay too much attention to external information such as image contrast and image exposure. This method eliminates the external interference information, does not need to consider too much deformation information, does not need a large number of samples to support, the early detection workload is small, and the recognition rate is high.

2. The establishment of the sample image First, create the positioning template, then select the ROI area, then traverse the normal pantograph historical images to locate and crop the normal image samples in the corresponding area, and finally save the corresponding template, parameters and sample images for the convenience of subsequent waiting. The detection images are positioned and compared, and the grayscale projection feature similarity and template matching feature similarity are calculated between the to-be-detected image and the historical frame template.

3. The grayscale projection feature is the normalized grayscale projection gradient feature, the extracted grayscale feature is the grayscale projection in the horizontal direction, and the projection data is normalized to weaken the influence of illumination.

4. Template matching adopts shape-based template matching, and the extracted features used are the position of edge points and the gradient direction of edge points, so as to perform template matching feature similarity calculation between the image to be detected and the historical frame template.

Description of drawings

Fig. 1 is the overall block diagram of the present invention;

Fig. 2 is a schematic diagram of 6 ROIs in Fig. 1;

Fig. 3 is the flow chart of positioning template and ROI area creation of the present invention;

4 is the original image of the normal pantograph history image obtained in the data flow chart of the positioning template and the ROI area according to the present invention;

Fig. 5 is the target template region selected in the present invention's positioning template and ROI region creation data flow chart;

Fig. 6 is the target template region cropped in the data flow chart of positioning template and ROI region creation of the present invention;

Fig. 7 is the matching result diagram in the present invention's positioning template and ROI area creation data flow chart;

Fig. 8 is 6 ROI regions selected in the positioning template and ROI region creation data flow chart of the present invention;

Fig. 9 is the flow chart of sample image creation of the present invention;

10 is the original image of the normal pantograph historical image obtained in the data flow chart of the sample image creation of the present invention;

Fig. 11 is the matching result diagram in the data flow chart of the sample image creation of the present invention;

Fig. 12 is the image after affine transformation in the sample image creation data flow chart of the present invention;

Fig. 13 is the ROI area on the affine transformation image selected in the sample image creation data flow chart of the present invention;

Fig. 14 is the ROI region selected by cropping in the sample image creation data flow chart of the present invention;

15 is a flowchart of deformation detection of the present invention;

16 is the original image of the image to be detected obtained from the deformation detection data flow chart of the present invention;

Fig. 17 is the template matching in the deformation detection data flow chart of the present invention;

18 is the process of correcting the image in the deformation detection data flow chart of the present invention;

Fig. 19 is the detection ROI region selected in the deformation detection data flow chart of the present invention;

Fig. 20 is the image cropping in the deformation detection data flow chart of the present invention;

Fig. 21 is the image smoothing in the deformation detection data flow chart of the present invention;

Figure 22 is the extraction of grayscale features in the deformation detection data flow chart of the present invention;

Fig. 23 is the template feature similarity in the deformation detection data flow chart of the present invention;

Fig. 24 is the template feature similarity of another ROI region in the deformation detection data flow chart of the present invention;

25 is a schematic diagram of the grayscale average value of each row in the horizontal direction of grayscale projection calculation according to the present invention;

Figure 26 is the grayscale projection curve of the present invention;

Figure 27 is the normalized grayscale projection curve of the present invention;

FIG. 28 is a grayscale projection curve after gradient transformation of the present invention.

Detailed ways

The concept, specific structure and technical effects of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings, so as to fully understand the purpose, features and effects of the present invention.

Example 1

A pantograph deformation detection method based on inter-frame comparison, as shown in Figures 1 and 2, includes the following steps:

S1, obtain vehicle number information, and search for the positioning template image and historical frame ROI sample image of the normal pantograph of the model corresponding to the current vehicle number from the sample image set according to the vehicle number information; the positioning template image includes a center positioning template and more than one ROI positioning area; the position of the ROI sample image of the historical frame is consistent with the ROI positioning area.

The normal pantograph image is the top view image of the pantograph, including the easily deformable area and the non-deformable area of the pantograph. Among them, the central positioning template is an area that is not easily deformed and can be clearly displayed in the top view of the pantograph, such as insulators, airbags, and parts of the pantograph roof fixing parts, and this area is selected as the positioning template. The ROI is the region of interest, which is the six easily deformable regions on the pantograph carbon sliding plate and the horn of the balance rod.

S2: Acquire an image to be inspected, perform template matching between the image to be inspected and the center positioning template, generate a mapping matrix, and transform the image to be inspected into a shape position consistent with the positioning template image according to the mapping matrix.

S3, according to the positional relationship between the center positioning template and the ROI positioning area, the ROI target image is intercepted from the transformed image to be inspected, and the grayscale projection feature similarity and template matching feature similarity between the historical frame ROI sample image and the ROI target image at the corresponding position are calculated. Spend.

S4, calculate the total similarity of a single ROI according to the similarity of the grayscale projection feature and the similarity of the template matching feature. If the total similarity meets the preset threshold, it is judged to be normal, otherwise it is deformed; traverse all ROI areas, if any ROI If the similarity determination result is abnormal, it is determined that the pantograph deformation occurs in the current image to be inspected.

In this embodiment, the sample images are centrally set with six ROI areas of the pantograph carbon slide and the balance bar, and each ROI area stores multiple ROI sample images. When calculating the similarity, each ROI target image is associated with multiple Calculate the total similarity of each ROI sample image, and then take multiple averages of the total similarity as the criterion. The determination of whether the pantograph is deformed according to the total similarity calculation result is that if any of several ROI regions on the pantograph image to be detected is deformed, the pantograph is deformed, and if all are normal, the pantograph is not deformed.

The pantograph deformation detection method provided in this embodiment belongs to the intelligent identification method of the 5C pantograph deformation. The function of obtaining the car number is that different car numbers correspond to different models, and the corresponding pantograph history frame template is selected for different models. The function of the sample image is to obtain multiple historical normal pantograph historical images of this type of vehicle, and to obtain the sample templates of 6 ROI regions with the same pose and scale.

In the sample images, they are classified according to the car number, because different car numbers correspond to different models, and during the detection, the corresponding normal pantograph historical images are selected for comparison and detection of different models. Specifically, in the sample image creation step, for models with different car numbers, corresponding sample images are created; in the deformation detection step, according to the car number of the detected image, first select the sample image corresponding to the car number in the sample image , and then compare with the sample image.

Through the above method, the pantograph deformation detection is not disturbed by changes in external illumination, and is only related to the grayscale projection of the ROI area of the current image and the similarity between the template matching feature and the historical image. For each frame of image, there is no need to pay too much attention to the transformation of the contrast of the image, nor to external information such as image exposure. The external interference information is eliminated, the deformation information does not need to be considered too much, and there is no need for a large number of samples to support, the early detection workload is small, and the recognition rate is relatively high.

Example 2

In this embodiment, on the basis of Embodiment 1, the sample image in step S1 is further described. The establishment of sample images includes the steps of creating positioning templates and ROI regions and creating steps of ROI sample images of historical frames.

This embodiment adopts the historical frame comparison method. The precondition of using the historical frame comparison method is that the two regions to be compared are consistent in shape and position. Therefore, the precondition is first considered when designing the algorithm. The purpose of establishing the sample image is to satisfy the preconditions of the algorithm. The establishment of the sample image firstly creates the positioning template, then selects 6 ROI regions, then traverses the normal pantograph historical image to locate and crop the normal image samples in the corresponding area, and finally saves the corresponding template, parameters and sample images.

(1) Positioning template and ROI area creation

The function of the positioning template is to transform all the detected images to the same position as the template image to ensure the consistency of shape and position.

As shown in Figure 3, the positioning template and the ROI area creation step are to obtain the original image of the normal pantograph history image, select the target template area on the original image and intercept the target template area, and create an image template with the intercepted target template area As a positioning template; perform template matching between the image template and the original image, select several ROI areas on the original image according to the template results, and save the image template and ROI area.

Figure 4-8 shows the data flow chart of positioning template and ROI area creation. In the original images of several normal pantograph historical images of the same vehicle number, first obtain an original image of the normal pantograph historical image as shown in Figure 4, and select the target template on the original image as shown in Figure 5 area, and create an image template by cropping the target template area as shown in Figure 6. Then, the image template is matched with the original image of the normal pantograph historical image, and the matching result is shown in Figure 7. Next, select 6 ROI regions on the original image as shown in Figure 8. Finally, save the image template information and ROI area information for deformation detection.

(2) Historical frame ROI sample image creation

As shown in FIG. 9 , the historical frame ROI sample image creation step is to obtain a historical normal pantograph image, perform template matching between the normal pantograph image and the center positioning template, and generate a mapping matrix. The mapping matrix transforms the normal pantograph image into a shape position consistent with the positioning template image; according to the positional relationship between the center positioning template and the ROI positioning area, the historical frame ROI sample image is intercepted from the transformed normal pantograph image, and the ROI is The sample image is saved according to the corresponding ROI positioning area.

The sample image creation data flow diagram is shown in Figure 10-15. From the other original images of the other normal pantograph historical images of the same vehicle number, obtain the original images of the normal pantograph historical images as shown in Figure 10, and perform template matching between the original image and the image template, and the matching results are shown in the figure 11; then perform affine transformation on the original image, as shown in Figure 12; secondly, select the ROI area on the affine transformed image, as shown in Figure 13; then crop the selected ROI area, as shown in Figure 14 ; Finally, save the intercepted image sample to the corresponding ROI area position.

The second method may also be used for creating the ROI sample image of the historical frame: saving the ROI target image of the pantograph image detection process previously determined to be normal as the ROI sample image of the corresponding area.

In the sample image establishment step, the created image template and the saved and intercepted ROI area constitute the sample image. For the pantograph of each vehicle number, several original images of normal pantographs with different history can be selected. For example, in this embodiment, four images can be selected, and one of the four original images can be selected to create a positioning template and ROI The other 3 original images are created according to the created positioning template and ROI area. 4 original images of normal pantograph historical images, and create their historical frame templates for comparison of the images to be detected in the subsequent deformation detection steps.

Example 3

This embodiment is further improved on the basis of Embodiment 2, and the deformation detection step includes the calculation step of gray-scale projection feature similarity S1, the calculation step of the average value of matching scores S2, and the step of determining whether the pantograph is deformed.

As shown in FIG. 15 , the gray-scale projection feature similarity S1 calculation step is to obtain the image to be detected, and perform template matching between the image to be detected and the image template; according to the matching result, perform affine transformation on the image to be detected, and intercept the shoot a ROI area image in the transformed image, and calculate the horizontal grayscale projection of the ROI area image of the image to be detected; obtain a sample image, and calculate the horizontal grayscale projection of the corresponding ROI area in the sample image; according to the ROI area of the image to be detected The horizontal grayscale projection of the image and the horizontal grayscale projection of the corresponding ROI area in the sample image are used to calculate the grayscale projection feature similarity S1.

As shown in FIG. 15 , in the calculation step of the average value of the matching scores S2, according to the ROI area image after the affine transformation of the image to be detected intercepted in the step of grayscale projection feature similarity S1, a ROI area template is created, and the sample image is ROI region template matching is performed in , and the average S2 of the matching scores is calculated.

As shown in Figure 15, in the step of determining whether the pantograph is deformed, the total similarity S is calculated according to the gray-scale projection feature similarity S1 and the average value of the matching score S2, and the size of the total similarity S and the threshold T is determined; If S is less than T, the pantograph is deformed; if S is greater than T, it is judged whether to traverse all ROI areas. If so, the pantograph is not deformed. If not, it returns to the calculation step of the gray-scale projection feature similarity S1. The step of intercepting the ROI area image in the affine transformed image, and intercepting another ROI area for deformation detection.

，T1为灰度特征的权 重，T2为模板特征的权重，T1+T2=1。 Among them, the calculation formula of the total similarity S is:

, T1 is the weight of the grayscale feature, T2 is the weight of the template feature, T1+T2=1.

In this embodiment, due to the grayscale feature used in grayscale projection, the grayscale feature will fluctuate with the influence of illumination, so the weight of the grayscale feature will be set lower in the calculation of the final similarity, and the template feature adopts the Shape features are not sensitive to lighting effects, so the weight is higher than that of gray features. The deformation judgment method is:

，

,

Where T is the threshold, generally set to 0.5.

The data flow of deformation detection is shown in Figure 16-26, where the data in the box is the similarity result. First obtain the original image of the image to be detected, as shown in Figure 16; then perform template matching between the original image of the to-be-detected image and the image template, as shown in Figure 17; then correct the image according to the matching result, as shown in Figure 18; then Select the detected ROI area, as shown in Figure 19; then crop the selected and detected ROI area, as shown in Figure 20, and smooth the image, as shown in Figure 21; then extract the grayscale features, such as As shown in Figure 22, and calculate the grayscale feature similarity; then calculate the template feature similarity, as shown in Figure 23; for another ROI area, according to the above method, calculate its grayscale feature similarity and template feature similarity, as As shown in Figure 24, and so on, traverse all the ROI areas, and perform deformation detection on the pantograph.

In this embodiment, the template matching before the affine transformation is used to transform all the detected images to the same positions as the template images, so as to ensure the shape and position consistency. The template matching in the average S2 of the matching scores is calculated, which is used to calculate the template shape matching similarity feature.

Example 4

In this embodiment, on the basis of Embodiment 3, the method for extracting grayscale features is further elaborated. The grayscale projection feature is a normalized grayscale projection gradient feature, and the extraction method is as follows: first calculate the grayscale mean value of each line in the horizontal direction of the image, and calculate the grayscale projection data of the image according to the grayscale mean value; Normalization processing and gradient transformation are sequentially performed on the grayscale projection data to obtain normalized grayscale projection gradient features. Subsequent projected gradient feature similarity computations employ existing methods, such as Euclidean distance or cosine similarity.

Specifically, grayscale projection basically calculates the grayscale mean value of each row in the horizontal direction, as shown by the box selected in Figure 25 .

The gray mean value is calculated as follows:

，

,

为图像的宽度，

为行，

为第

列，

为灰度均值，

为图像，

为坐标在

处的灰度值。照上述的方法计算得到的图像的灰度投影曲线 如图26所示。 in,

is the width of the image,

for the line,

for the first

List,

is the gray mean value,

for the image,

for the coordinates in

grayscale value at . The grayscale projection curve of the image calculated according to the above method is shown in FIG. 26 .

The grayscale projection normalization method is as follows:

，

,

为灰度投影数据，

为灰度投影数据的最小值，

为灰度投影数据 的最大值。归一化的灰度投影曲线如图27所示。 in,

is the grayscale projection data,

is the minimum value of the grayscale projection data,

is the maximum value of grayscale projection data. The normalized grayscale projection curve is shown in Figure 27.

In order to reflect the fluctuation of grayscale features, this embodiment performs gradient transformation on the grayscale projection. The interval of gradient transformation is step, and step is set according to the actual experimental environment. The default value in this embodiment is 1. The curve after gradient transformation is shown in Figure 28. shown.

According to the normalized grayscale projection data obtained by calculation, the grayscale gradient sum is calculated as the grayscale feature.

Example 5

In this embodiment, on the basis of Embodiment 4, template matching is further described. The template matching adopts shape-based template matching, and the extracted features are the position of the edge point and the gradient direction of the edge point. First, the edge point of the template image is extracted, and then the x-direction and y-direction gradients of the image are calculated with the sobel operator. , and then calculate the total gradient value and gradient direction of the edge point according to the x-direction and y-direction gradients, and finally calculate the position of the edge point. The extraction method of edge point gradient direction features is as follows:

Extract the edge points of the template image.

To calculate the gradient direction and gradient value of the edge point, first calculate the x-direction and y-direction gradients of the template image. The calculation method is as follows: take the sobel operator as an example.

The gradients in the X and Y directions are calculated as follows:

Then calculate the gradient value at that point as follows:

，

,

为x方向梯度，

为y方向梯度； in,

is the x-direction gradient,

is the gradient in the y direction;

The final calculated gradient angle is as follows:

，

,

The gradient direction is the direction in which the grayscale of the image increases. The gradient direction corresponding to finding a certain point in the image is calculated by calculating the gradient angle between the point and its 8 neighboring points. The maximum gradient angle is the gradient direction.

The matching process is as follows:

， The image search is performed in a sliding window mode, and the matching score in the window is calculated, and the matching score is the cosine of the angle between the template edge point and the direction vector of the corresponding point corresponding to the area to be matched:

,

Among them, score is the matching score, p _i is the cosine value of the included angle, and n is the number of points;

，

,

Wherein A _j and B _j are the elements of the direction vector of the corresponding point of the template and the area to be matched;

The above direction vector is expressed as:

，

,

为梯度角。 in

is the gradient angle.

Example 6

A computer device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it executes any one of the methods for detecting pantograph deformation in the above-mentioned embodiments 1-5. A step of.

In this embodiment, the processor may be a central processing unit (Central Processing Unit, CPU). The processor may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other Chips such as programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above types of chips.

As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer-executable programs, and units. The processor executes various functional applications of the processor and works data processing by running the non-transitory software programs, instructions and modules stored in the memory, that is, to implement the methods in the above embodiments.

The memory may include a storage program area and a storage data area, wherein the storage program area may store an operating system and an application program required by at least one function; the storage data area may store data created by the processor, and the like. Additionally, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory may optionally include memory located remotely from the processor, which may be connected to the processor through a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

One or more units are stored in the memory, and when executed by the processor, perform the method of any one of the foregoing embodiments 1-5.

Example 7

A computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps in any one of the above-mentioned methods for detecting pantograph deformation in Embodiments 1-5.

The embodiments of the present invention have been specifically described above, but the present invention is not limited to the embodiments. Those skilled in the art can also make various equivalent modifications or substitutions without departing from the spirit of the present invention. These equivalents or substitutions All are included within the scope defined by the claims of the present invention.

Claims (9)

1. The pantograph deformation detection method based on inter-frame contrast is characterized by comprising the following steps of: the method comprises the following steps:

s1, acquiring the car number information, and searching a positioning template image and a historical frame ROI sample image of a normal pantograph of a car type corresponding to the current car number from the sample image set according to the car number information; the positioning template image comprises a central positioning template and more than 1 ROI positioning area; the historical frame ROI sample image is consistent with the position of an ROI positioning area;

s2, acquiring an image to be detected, performing template matching on the image to be detected and the central positioning template to generate a mapping matrix, and transforming the image to be detected to a shape position consistent with the positioning template image according to the mapping matrix;

s3, intercepting an ROI target image from the transformed image to be detected according to the position relation between the central positioning template and the ROI positioning region, and calculating the gray projection feature similarity and template matching feature similarity of the historical frame ROI sample image and the ROI target image at the corresponding positions;

s4, calculating the total similarity of the single ROI according to the gray projection feature similarity and the template matching feature similarity in a weighted mode, if the total similarity meets a preset threshold value, judging that the ROI is normal, and otherwise, judging that the ROI is deformed; and traversing all ROI areas, and if the similarity judgment result of any ROI is abnormal, judging that the pantograph deformation occurs in the current image to be detected.

2. The pantograph deformation detection method according to claim 1, wherein:

the first acquisition mode of the historical frame ROI sample image is as follows: acquiring a historical normal pantograph image, performing template matching on the normal pantograph image and the central positioning template to generate a mapping matrix, and transforming the normal pantograph image to a shape position with the same positioning template image according to the mapping matrix; intercepting a historical frame ROI sample image from the transformed normal pantograph image according to the position relation between the central positioning template and the ROI positioning region, and storing the ROI sample image according to the corresponding ROI positioning region;

and/or mode two: and saving the ROI target image in the pantograph image detection process which is previously determined to be normal as the ROI sample image of the corresponding region.

3. The pantograph deformation detection method according to claim 1, wherein: the gray projection feature is a normalized gray projection gradient feature, and the extraction method comprises the following steps:

calculating the gray level mean value of each line in the horizontal direction of the image, and calculating the gray level projection data of the image according to the gray level mean value;

and sequentially carrying out normalization processing and gradient transformation on the gray projection data to obtain the gradient characteristic of the normalized gray projection.

4. The pantograph deformation detection method according to claim 1, wherein: the template matching in the S4 adopts the template matching based on the shape, the adopted characteristics are the position of the edge point and the gradient direction of the edge point, and the extraction method of the gradient direction characteristics of the edge point comprises the following steps:

and extracting the edges of the two images to be matched, calculating the gradients of the images in the x direction and the y direction, and calculating the total gradient value and the gradient direction of the edge points according to the gradients in the x direction and the y direction.

5. The pantograph deformation detection method according to claim 4, wherein: the template matching feature similarity calculation process comprises the following steps: image search is carried out in a sliding window mode, and a matching score in a window is calculated, wherein the matching score is a cosine value of an included angle between a template edge point and a direction vector of a corresponding point of a corresponding region to be matched:

，

where score is the match score, p_iIs the cosine value of the included angle, and n is the number of points;

，

wherein A is_jAnd B_jElements of direction vectors of corresponding points of the template and the area to be matched;

the above direction vector is expressed as:

，

wherein

Is a gradient angle.

6. The pantograph deformation detection method according to claim 1, wherein: judging whether the pantograph is deformed or not according to the total similarity calculation result, and judging that the pantograph is deformed if any one of a plurality of ROI areas on the pantograph image to be detected is deformed; and if a plurality of ROI areas on the pantograph image to be detected are not deformed, judging that the pantograph is not deformed.

7. The pantograph deformation detection method according to claim 1, wherein: six horn ROI areas of the pantograph carbon slide plate and the balance rod are arranged in the sample image, a plurality of ROI sample images are stored in each ROI area, when the similarity is calculated, the total similarity is calculated between each ROI target image and the ROI sample images, and then the mean value of the total similarities is taken as a criterion.

8. A computer device, characterized by: comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to perform the steps of the method for detecting a powered bow according to any of the preceding claims 1 to 7.

9. A computer-readable storage medium characterized by: stored with a computer program which, when executed by a processor, implements the steps in the power receiving bow variation detection method according to any one of claims 1 to 7.

Images