A method and system for locating a target distortion

Abstract

The application provides a target distortion positioning method and system. The method can acquire a to-be-positioned image, and construct an image pyramid according to the to-be-positioned image, wherein the to-be-positioned image comprises a to-be-positioned target, and the image pyramid comprises pyramid sub-images of multiple layers with different resolutions. A target detection point in the pyramid sub-image is searched, a target deformation amount of the target detection point is calculated, and the target deformation amount is diffused to obtain a deformation amount image of the pyramid sub-image. The deformation amount image of each layer of pyramid sub-image is sequentially mapped to a next layer of pyramid sub-image in a direction from a top layer to a bottom layer of the image pyramid. A target detection point in the bottom layer of pyramid sub-image is searched, and a target position of the to-be-positioned target is positioned according to the target detection point. The method can position different degrees of deformation accurately based on an image local template, diffuse the deformation amount, and transfer the deformation amount based on the image pyramid, thereby reducing the positioning time and realizing accurate positioning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image positioning, in particular to a target distortion positioning method and system. BACKGROUND

[0002] In the field of image positioning, a template matching method can be used to realize target positioning. By setting a template image, the matching degree between the template image and the measured object in the measured image is analyzed, and then the measured object is positioned. However, all image-based target positioning algorithms assume that the measured object is parallel to a known plane in space, and the pose of the object on the plane can usually be described as a similarity transformation. However, when the measured object is distorted, for example, there is a certain degree of tilt, or there is local deformation due to the softness of the object material. Shape matching based on templates cannot achieve accurate positioning, reducing positioning efficiency. SUMMARY

[0003] The present application provides a target distortion positioning method and system to solve the problem of reduced positioning accuracy due to target distortion.

[0004] In a first aspect, the present application provides a target distortion positioning method, comprising:

[0005] Obtaining a to-be-positioned image, the to-be-positioned image comprising a to-be-positioned target;

[0006] Constructing an image pyramid according to the to-be-positioned image, the image pyramid comprising a plurality of pyramid sub-images with different resolutions;

[0007] Searching for a target detection point in the pyramid sub-image, the target detection point being a pixel point obtained based on image local template positioning;

[0008] Calculating a target deformation amount of the target detection point, and diffusing the target deformation amount to obtain a deformation amount image of the pyramid sub-image;

[0009] Mapping the deformation amount image of each pyramid sub-image in a direction from the top layer to the bottom layer of the image pyramid to the next layer of the pyramid sub-image;

[0010] Searching for a target detection point in the bottom layer of the pyramid sub-image, and positioning a target position of the to-be-positioned target according to the target detection point.

[0011] In an optional embodiment, the step of constructing an image pyramid according to the input image further comprises:

[0012] Image down-sampling the to-be-positioned image according to a preset number of images to generate a preset number of pyramid sub-images;

[0013] combining pyramid sub-images of a preset image quantity according to a pyramid shape to construct the image pyramid.

[0014] In an optional embodiment, the step of searching the target detection point in the pyramid sub-image further comprises:

[0015] obtaining an image local template;

[0016] traversing a pixel point in the top layer pyramid sub-image according to the image local template;

[0017] searching a pixel point with a similarity greater than a similarity threshold to the image local template as a target detection point in the top layer pyramid sub-image.

[0018] In an optional embodiment, the step of searching the target detection point in the pyramid sub-image further comprises:

[0019] obtaining a deformation quantity image of the previous layer pyramid sub-image, the deformation quantity image comprising a row deformation image and a column deformation image, the row deformation image being used to represent a deformation quantity in a row direction of a pixel point, and the column deformation image being used to represent a deformation quantity in a column direction of a pixel point;

[0020] locating an initial detection point in the current layer pyramid sub-image according to the deformation quantity image of the previous layer pyramid sub-image;

[0021] searching a pixel point with a similarity greater than a similarity threshold to the image local template in a preset size neighborhood of the initial detection point as a target detection point in the current layer pyramid sub-image.

[0022] In an optional embodiment, the step of diffusing the target deformation quantity further comprises:

[0023] obtaining a deformation quantity image of the previous layer pyramid sub-image;

[0024] initializing the deformation quantity image according to a target deformation quantity of a target detection point in the current layer pyramid sub-image;

[0025] dividing the deformation quantity image into a boundary region and a to-be-repaired region, the boundary region being a region of the target detection point on the deformation quantity image, and the to-be-repaired region being a region outside the boundary region on the deformation quantity image;

[0026] diffusing a deformation quantity of the to-be-repaired region according to the boundary region to obtain a deformation quantity image of the current layer pyramid sub-image.

[0027] In an optional embodiment, the step of initializing the deformation quantity image according to a target deformation quantity of a target detection point in the current layer pyramid sub-image further comprises:

[0028] calculating a row offset and a column offset of the target detection point according to the target deformation amount;

[0029] converting the row offset and the column offset into image brightness information respectively, and filling the image brightness information into the deformation amount image.

[0030] In an optional implementation, the step of diffusing the deformation amount of the region to be repaired according to the boundary region further comprises:

[0031] constructing a deformation amount diffusion equation according to the boundary region and the region to be repaired, the deformation amount diffusion equation being used to diffuse the deformation amount of the region to be repaired;

[0032] solving the deformation amount diffusion equation to obtain a deformation amount image of the pyramid sub-image of the current layer.

[0033] In an optional implementation, the deformation amount diffusion equation is constructed according to a preset extremum geometry algorithm.

[0034] iteratively updating the deformation amount diffusion equation so that a value of the deformation amount diffusion equation tends to zero;

[0035] stopping the iterative updating when the value of the deformation amount diffusion equation is less than a preset threshold value, to obtain the deformation amount image of the pyramid sub-image of the current layer.

[0036] In an optional implementation, the step of obtaining the deformation amount image of the pyramid sub-image further comprises:

[0037] filtering out outliers in the deformation amount image with a frequency greater than a first frequency threshold value based on a median filter;

[0038] filtering out noise in the deformation amount image with a frequency greater than a second frequency threshold value based on a smoothing filter.

[0039] In a second aspect, the present application provides a target distortion positioning system, comprising an image acquisition module, an image preprocessing module and an image positioning module, wherein:

[0040] the image acquisition module is configured to acquire a to-be-positioned image, the to-be-positioned image comprising a to-be-positioned target;

[0041] the image preprocessing module is configured to construct an image pyramid according to the to-be-positioned image, the image pyramid comprising multiple layers of pyramid sub-images with different resolutions;

[0042] the image positioning module is configured to search for a target detection point in the pyramid sub-image, the target detection point being a pixel point obtained based on image local template positioning;

[0043] calculate a target deformation of the target detection point, and diffuse the target deformation to obtain a deformation image of the pyramid sub-image;

[0044] map the deformation image of each pyramid sub-image in a direction from a top layer to a bottom layer of the image pyramid to a next layer of the pyramid sub-image;

[0045] search for a target detection point in the bottom layer of the pyramid sub-image, and locate a target position of the target to be located according to the target detection point.

[0046] According to the technical solutions, the method can acquire an image to be located, and construct an image pyramid according to the image to be located, wherein the image to be located includes a target to be located, and the image pyramid includes pyramid sub-images with different resolutions. A target detection point in the pyramid sub-image is searched, a target deformation of the target detection point is calculated, and the target deformation is diffused to obtain a deformation image of the pyramid sub-image, wherein the target detection point is a pixel point located based on an image local template. The deformation image of each pyramid sub-image is mapped to a next layer of the pyramid sub-image in a direction from a top layer to a bottom layer of the image pyramid. A target detection point in the bottom layer of the pyramid sub-image is searched, and a target position of the target to be located is located according to the target detection point. The method can calculate the deformation of the detection point in each layer of the image pyramid based on the image local template, diffuse the deformation of the detection point to the whole image based on an interpolation method, map the deformation calculated in the last layer of the image pyramid to the current layer of the image pyramid for tracking, and perform local search based on the current layer of the image pyramid. The transmission is performed from top to bottom along the image pyramid until the bottom layer of the image pyramid, the time consumption of location is reduced, and accurate location is realized. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0048] Figure 1 A flowchart for constructing an image pyramid is provided for the embodiments of the present application.

[0049] Figure 2 A schematic diagram of an image pyramid is provided for the embodiments of the present application.

[0050] Figure 3 A flowchart for locating a target to be located is provided for the embodiments of the present application.

[0051] Figure 4 A schematic diagram of a diffusion target deformation amount provided for an embodiment of the present application is shown in FIG. 8.

[0052] Figure 5 A schematic diagram of a diffusion target deformation amount based on an equation provided for an embodiment of the present application is shown in FIG. 9. DETAILED DESCRIPTION

[0053] The embodiments will be described in detail below with reference to examples thereof as illustrated in the accompanying drawings. In the following description, same numbers in different drawings represent same or similar elements unless otherwise represented. The implementations described in the following embodiments do not represent all implementations consistent with the present application. They are merely examples of systems and methods consistent with some aspects of the present application as detailed in the claims.

[0054] In the field of image positioning, image-based target positioning algorithms all assume that the measured object is parallel to a certain known plane in space, and the pose of the object on the plane can usually be described as a similarity transformation. However, when the measured object is distorted, for example, there is a certain degree of tilt, or due to the softness of the object material, there is local deformation, which leads to inaccurate positioning and reduces positioning efficiency.

[0055] To solve the above problems, a target distortion positioning method is provided in some embodiments of the present application, which can accurately position different degrees of deformation based on image local template positioning, and diffuse the deformation amount, while transmitting the deformation amount based on image pyramid, thereby reducing the positioning time and achieving accurate positioning. The target distortion positioning method provided by the present application comprises the following steps:

[0056] An image to be positioned is obtained, wherein the image to be positioned includes a target to be positioned. The specific implementation of obtaining the image to be positioned is not specifically limited in the embodiments of the present application, and those skilled in the art can adjust it according to the actual situation. For example, the surface of the target to be positioned can be photographed by a camera or other device to obtain the image to be positioned, or the image to be positioned sent by other devices can be received, or the pre-stored image to be positioned can be read from the local or cloud.

[0057] After obtaining the image to be positioned, an image pyramid can be constructed according to the image to be positioned. The image pyramid includes a plurality of pyramid sub-images with different resolutions. It can be understood that the image pyramid is a multi-scale expression of the image, and the image pyramid of the image to be positioned is a series of images arranged in a pyramid shape (from bottom to top) with gradually reduced resolutions. That is, the image pyramid constructed according to the image to be positioned is composed of a plurality of images to be positioned with different resolutions, and the higher the level, the smaller the image and the lower the resolution.

[0058] As Figure 1As shown, Figure 1 A flowchart for constructing an image pyramid is provided in the embodiments of the present application. When constructing an image pyramid based on a to-be-positioned image, the to-be-positioned image can be image down-sampled according to a preset image quantity to generate pyramid sub-images of the preset image quantity. Then, the pyramid sub-images of the preset image quantity can be combined according to a pyramid shape to construct the image pyramid.

[0059] The image quantity can be set according to user demand, for example, the image quantity is set to 5, that is, a 5-layer image pyramid is constructed. After the to-be-positioned image is acquired, the to-be-positioned image is image down-sampled according to echelon to generate five pyramid sub-images with different resolutions. Then, the five pyramid sub-images with different resolutions are arranged according to a pyramid shape (from bottom to top) to construct the image pyramid. As shown in Figure 2 As shown, Figure 2 A schematic diagram of an image pyramid is provided in the embodiments of the present application, which includes five pyramid sub-images, and the resolution gradually decreases along the direction from the bottom layer to the top layer of the image pyramid.

[0060] As shown, Figure 3 As shown, Figure 3 A flowchart for positioning a to-be-positioned target is provided in the embodiments of the present application. After the image pyramid is constructed, the positioning of the to-be-positioned target can be performed based on the image pyramid. Specifically, a target detection point in the pyramid sub-image can be searched, and a target deformation amount of the target detection point can be calculated, wherein the target detection point is a pixel point positioned based on an image local template. Then, the target deformation amount is diffused to obtain a deformation amount image of the pyramid sub-image. The deformation amount image is used to represent the deformation amount of the pixel point in the pyramid sub-image, including but not limited to coordinates, scaling scale, angle, etc.

[0061] After the deformation amount image of the current layer pyramid sub-image is acquired, the deformation amount image is mapped to the next layer pyramid sub-image, and the deformation amount image of the next layer pyramid sub-image is calculated and mapped to the next layer pyramid sub-image. That is, the deformation amount image of each layer pyramid sub-image is sequentially mapped to the next layer pyramid sub-image along the direction from the top layer to the bottom layer of the image pyramid. In this way, the deformation amount is transmitted based on the image pyramid, and the transmission is performed until the bottom layer pyramid sub-image of the image pyramid.

[0062] After the transmission to the bottom layer pyramid sub-image, a target detection point in the bottom layer pyramid sub-image can be searched, and a target position of the to-be-positioned target can be positioned according to the target detection point.

[0063] In the embodiments of the present application, the to-be-positioned target can be positioned based on the image pyramid. The to-be-positioned image is image down-sampled according to a preset image quantity to generate pyramid sub-images of the preset image quantity. Then, the pyramid sub-images of the preset image quantity are combined according to a pyramid shape to construct the image pyramid. After the image pyramid is constructed, the positioning of the to-be-positioned target can be performed based on the image pyramid. Specifically, a target detection point in the pyramid sub-image can be searched, and a target deformation amount of the target detection point can be calculated. Then, the target deformation amount is diffused to obtain a deformation amount image of the pyramid sub-image. After the deformation amount image of the current layer pyramid sub-image is acquired, the deformation amount image is mapped to the next layer pyramid sub-image, and the deformation amount image of the next layer pyramid sub-image is calculated and mapped to the next layer pyramid sub-image. That is, the deformation amount image of each layer pyramid sub-image is sequentially mapped to the next layer pyramid sub-image along the direction from the top layer to the bottom layer of the image pyramid. In this way, the deformation amount is transmitted based on the image pyramid, and the transmission is performed until the bottom layer pyramid sub-image of the image pyramid. Figure 2As shown in the image pyramid, the first layer pyramid sub-image, the second layer pyramid sub-image, the third layer pyramid sub-image, the fourth layer pyramid sub-image and the fifth layer pyramid sub-image are in the direction from the top layer to the bottom layer of the image pyramid. The deformation of the target detection point in the first layer pyramid sub-image can be calculated based on the image local template positioning, and the deformation of the target detection point is diffused to the full image based on the interpolation method to obtain the deformation image of the first layer pyramid sub-image, which is mapped to the second layer pyramid sub-image. The deformation of the target detection point in the second layer pyramid sub-image is calculated based on the image local template positioning, and is diffused to the full image to obtain the deformation image of the second layer pyramid sub-image, which is mapped to the third layer pyramid sub-image. In this way, the deformation is transmitted based on the image pyramid until it is transmitted to the fifth layer pyramid sub-image, the target detection point in the fifth layer pyramid sub-image is positioned based on the image local template, and then the target to be positioned is positioned.

[0064] It can be understood that, since the image pyramid is constructed by image down-sampling, the resolution of the upper layer pyramid sub-image of the image pyramid is lower, and there may be a detection point in a certain region of the current layer pyramid sub-image, but there is no detection point in the corresponding region of the upper layer pyramid sub-image, which further causes that the region cannot obtain effective deformation information. Therefore, the deformation is diffused to the full image based on the information of the adjacent region to obtain the deformation of each pixel, and the positioning accuracy of the target distortion is further improved.

[0065] In some embodiments, when the deformation of the target detection point is calculated based on the image local template positioning, the preset image local template can be matched with the pyramid sub-image, and the pixel point with a similarity greater than a similarity threshold with the image local template in the pyramid sub-image can be searched as the target detection point in the pyramid sub-image, and the deformation thereof is calculated.

[0066] For the top layer pyramid sub-image, after the image local template is obtained, the pixel points in the top layer pyramid sub-image can be traversed according to the image local template, and the pixel points with a similarity greater than a similarity threshold with the image local template can be searched as the target detection point in the top layer pyramid sub-image. For the pyramid sub-image below the top layer, the deformation image of the upper layer pyramid sub-image can be obtained, and the initial detection point in the current layer pyramid sub-image is positioned according to the deformation image of the upper layer pyramid sub-image. The initial detection point is the position corresponding to the target detection point in the current layer pyramid sub-image in the upper layer pyramid sub-image. And the pixel points with a similarity greater than a similarity threshold with the image local template in the preset size neighborhood of the initial detection point can be searched as the target detection point in the current layer pyramid sub-image.

[0067] That is, a local search is performed according to the position of the target detection point in the upper layer pyramid sub-image to search for a target detection point in the current layer pyramid sub-image that has a similarity greater than a similarity threshold, the target deformation of the target detection point is calculated, and the target deformation is diffused based on an interpolation method to obtain a deformation image of the current layer pyramid sub-image.

[0068] In addition, in order to improve efficiency, part of the pixel points can be extracted according to the edge information of the target to be positioned in the pyramid sub-image, and then the pixel points with a similarity to the image local template greater than a similarity threshold are screened to serve as the target detection point. For example, for a rectangular pattern in the image, the four vertices thereof can be taken as the detection points.

[0069] It should be noted that the size of the neighborhood and the interpolation method are not specifically limited in the embodiments of the present application, and a person skilled in the art can adjust them according to actual conditions. For example, the deformation can be diffused based on Hermite interpolation.

[0070] The positioning method for target distortion provided in the embodiments of the present application includes local template positioning, deformation diffusion, and pyramid deformation transmission. The deformation of the detection point in each layer of the pyramid sub-image in the image pyramid can be calculated based on the image local template, the deformation of the detection point is diffused to the full image based on the interpolation method, the deformation calculated from the upper layer pyramid sub-image in the image pyramid is mapped to the current layer pyramid sub-image for tracking positioning, and the local template positioning is performed based on the current layer pyramid sub-image. In this way, the transmission is performed along the image pyramid from the top layer to the bottom layer until the bottom layer of the image pyramid, and then the target position of the target to be positioned is positioned, the positioning time is reduced, and accurate positioning is achieved.

[0071] In some embodiments, due to local mismatch, there may be outliers in the deformation of the template point, which may cause discontinuous discontinuities to be introduced in the deformation, or even cause the deformation interpolation to cross. Therefore, while preventing the discontinuity and cross problems through the interpolation method, the influence of outliers can be further reduced through median filtering on the deformation image, and the noise and high-frequency components in the deformation image can be removed through smoothing filtering. That is, in some embodiments, after the deformation image of the pyramid sub-image is obtained, outliers with a frequency greater than a first frequency threshold in the deformation image can be filtered based on the median filtering, and noise with a frequency greater than a second frequency threshold in the deformation image can be filtered based on the smoothing filtering.

[0072] It can be understood that the positioning accuracy of the target distortion depends on the accuracy of the interpolation method, and the time consumption of the entire positioning algorithm is mainly concentrated in the interpolation module. Through the interpolation method, the discontinuity and cross problems can be prevented, and the interpolation accuracy can be improved by adding a filtering operation to remove outliers, so that the optimized interpolation can converge faster and the positioning efficiency can be improved.

[0073] In some embodiments, as shown in FIG. 2, the target deformation amount of the target detection point in the current layer pyramid sub-image is obtained by interpolation. Figure 4 Figure 4 As shown in FIG. 3, the schematic diagram of the target deformation amount provided by the embodiments of the present application can obtain the deformation amount image of the last layer pyramid sub-image based on the interpolation of the target deformation amount, and initialize the deformation amount image according to the target deformation amount of the target detection point in the current layer pyramid sub-image. That is, according to the target deformation amount of the target detection point, the offset of the target detection point on the deformation amount image is initialized.

[0074] The image brightness information can also be used to represent the local deformation of the pixel point in the row and column directions, that is, in some embodiments, the row offset and the column offset of the target detection point can be calculated according to the target deformation amount of the target detection point in the current layer pyramid sub-image, and the row offset and the column offset can be converted into image brightness information respectively, and the image brightness information can be filled into the deformation amount image.

[0075] Since the offset of the target detection point includes the offset in the row and column directions, the deformation amount image can be set to two, which respectively represent the deformation in the row direction and the deformation in the column direction, that is, the deformation amount image includes a row deformation image and a column deformation image, the row deformation image is used to represent the deformation amount of the pixel point in the row direction, and the column deformation image is used to represent the deformation amount of the pixel point in the column direction. For example, the row offset of the pixel point in the 10th row and the 10th column of the pyramid sub-image is 2, and the position of the 10th row and the 10th column in the corresponding row deformation image is filled with 2. Similarly, the column offset of the pixel point in the 10th row and the 10th column of the pyramid sub-image is 4, and the position of the 10th row and the 10th column in the corresponding column deformation image is filled with 4.

[0076] After initializing the deformation amount image, the deformation amount image can be divided into a boundary region and a to-be-repaired region, wherein the boundary region is the region of the target detection point on the deformation amount image, and the to-be-repaired region is the region outside the boundary region on the deformation amount image. The boundary region is the position of the target detection point, and the deformation amount thereof is known. The deformation amount of the to-be-repaired region can be diffused according to the boundary region to obtain the deformation amount image of the current layer pyramid sub-image.

[0077] In some embodiments, as shown in FIG. 4, the target deformation amount of the target detection point in the current layer pyramid sub-image is obtained by equation. Figure 5 Figure 5 As shown in FIG. 5, the schematic diagram of the target deformation amount based on the equation provided by the embodiments of the present application can convert the problem of calculating the diffusion deformation amount into an optimal estimation problem in mathematics, that is, when the deformation amount of the to-be-repaired region is diffused according to the boundary region, the deformation amount diffusion equation can be constructed according to the boundary region and the to-be-repaired region, wherein the deformation amount diffusion equation is used to diffuse the deformation amount of the to-be-repaired region. And solve the deformation amount diffusion equation to obtain the deformation amount image of the current layer pyramid sub-image.

[0078] ​​In some embodiments, the deformation diffusion equation can be constructed according to a preset extremum geometry algorithm, and when solving the deformation diffusion equation, the deformation diffusion equation can be iteratively updated to make the value of the deformation diffusion equation approach zero, and then the row and column deformation of the to-be-repaired region is diffused. When the value of the deformation diffusion equation is less than a preset threshold, the iterative updating is stopped to obtain a smooth image, i.e., the deformation image of the current layer pyramid sub-image.

[0079] In the present application, the extremum geometry algorithm and the method of iteratively updating the deformation diffusion equation are not specifically limited, and those skilled in the art can adjust them according to actual conditions. For example, the deformation diffusion equation is constructed according to Euler-Lagrange Equation, and the deformation diffusion equation is iteratively updated according to Laplace equation.

[0080] In some embodiments, in order to reduce the time-consuming of iteratively solving Hermite interpolation, the conjugate gradient method can be used to solve the deformation diffusion equation, so as to make it converge faster and improve the positioning efficiency.

[0081] In the present application, first, the local template can be used to accurately position the deformation of different degrees; second, the interpolation method can be used to accurately and quickly transfer the deformation, and at the same time, the filter operation can be used to remove outliers and improve the interpolation accuracy, so that the optimized interpolation can converge faster and improve the positioning efficiency; finally, based on the image pyramid transmission deformation, the positioning time can be reduced to realize accurate positioning.

[0082] Based on the above positioning method of target distortion. In some embodiments of the present application, a positioning system of target distortion is also provided, which includes an image acquisition module, an image preprocessing module and an image positioning module.

[0083] The image acquisition module is configured to acquire a to-be-positioned image, and the to-be-positioned image includes a to-be-positioned target.

[0084] The image preprocessing module is configured to construct an image pyramid according to the to-be-positioned image, and the image pyramid includes a plurality of pyramid sub-images with different resolutions.

[0085] The image positioning module is configured to search for a target detection point in the pyramid sub-image, and the target detection point is a pixel point obtained based on image local template positioning.

[0086] The target deformation of the target detection point is calculated, and the target deformation is diffused to obtain a deformation image of the pyramid sub-image.

[0087] mapping the deformation map of each layer of the pyramid sub-image to the next layer of the pyramid sub-image in the direction from the top layer to the bottom layer of the image pyramid.

[0088] searching for a target detection point in the bottom layer of the pyramid sub-image, and locating a target position of the target to be located according to the target detection point.

[0089] According to the above technical solutions, the application provides a target distortion positioning method and system. The method can obtain an image to be located, and construct an image pyramid according to the image to be located, wherein the image to be located includes a target to be located, and the image pyramid includes multiple layers of pyramid sub-images with different resolutions. A target detection point is searched for in the pyramid sub-image, a target deformation of the target detection point is calculated, and the target deformation is diffused to obtain a deformation map of the pyramid sub-image, wherein the target detection point is a pixel point located based on an image local template. The deformation map of each layer of the pyramid sub-image is mapped to the next layer of the pyramid sub-image in the direction from the top layer to the bottom layer of the image pyramid. A target detection point is searched for in the bottom layer of the pyramid sub-image, and a target position of the target to be located is located according to the target detection point. The method can calculate the deformation of the detection point in each layer of the image pyramid based on the image local template, diffuse the deformation of the detection point to the whole image based on an interpolation method, map the deformation calculated in the previous layer of the image pyramid to the current layer of the image pyramid for tracking, and perform local search based on the current layer of the image pyramid. This is transmitted from top to bottom along the image pyramid until the bottom layer of the image pyramid, reduces the positioning time, and realizes accurate positioning.

[0090] The similar parts between the embodiments provided by the application can be referred to each other, the specific embodiments provided above are only several examples under the general concept of the application, and do not limit the protection scope of the application. Any other embodiments extended according to the application scheme without creative labor belong to the protection scope of the application for those skilled in the art.

Claims

1. A method of locating a target distortion, characterized by, The method comprises the following steps: obtaining a to-be-positioned image, the to-be-positioned image comprising a to-be-positioned target; constructing an image pyramid according to the to-be-positioned image, the image pyramid comprising a plurality of pyramid sub-images of different resolutions; searching for a target detection point in the pyramid sub-image, the target detection point being a pixel point obtained based on image local template positioning; calculating a target deformation amount of the target detection point, and diffusing the target deformation amount to obtain a deformation amount image of the pyramid sub-image; mapping the deformation amount image of each pyramid sub-image in a direction from a top layer to a bottom layer of the image pyramid to a next layer of the pyramid sub-image; searching for a target detection point in the bottom layer of the pyramid sub-image, and positioning a target position of the to-be-positioned target according to the target detection point; the step of searching for a target detection point in the pyramid sub-image further comprises the following steps: obtaining an image local template; traversing a pixel point in the top layer of the pyramid sub-image according to the image local template; searching for a pixel point with a similarity greater than a similarity threshold to the image local template as a target detection point in the top layer of the pyramid sub-image; the step of searching for a target detection point in the pyramid sub-image further comprises the following steps: obtaining a deformation amount image of the pyramid sub-image of the previous layer, the deformation amount image comprising a row deformation image and a column deformation image, the row deformation image being used to represent a deformation amount in a row direction of a pixel point, and the column deformation image being used to represent a deformation amount in a column direction of a pixel point; positioning an initial detection point in the pyramid sub-image of the current layer according to the deformation amount image of the pyramid sub-image of the previous layer; searching for a pixel point with a similarity greater than a similarity threshold to the image local template in a preset size neighborhood of the initial detection point as a target detection point in the pyramid sub-image of the current layer; the step of diffusing the target deformation amount further comprises the following steps: obtaining a deformation amount image of the pyramid sub-image of the previous layer; initializing the deformation amount image according to a target deformation amount of the target detection point in the pyramid sub-image of the current layer; dividing the deformation amount image into a boundary region and a to-be-repaired region, the boundary region being a region where the target detection point is located on the deformation amount image, and the to-be-repaired region being a region outside the boundary region on the deformation amount image; diffusing the deformation amount of the to-be-repaired region according to the boundary region to obtain a deformation amount image of the pyramid sub-image of the current layer.

2. The positioning method according to claim 1, characterized in that, The step of constructing an image pyramid according to the to-be-positioned image further comprises the following steps: performing image down-sampling on the to-be-positioned image according to a preset image quantity to generate a preset number of pyramid sub-images; combining the preset number of pyramid sub-images in a pyramid shape to construct the image pyramid.

3. The positioning method of claim 1, wherein, The step of initializing the deformation amount image according to a target deformation amount of the target detection point in the pyramid sub-image of the current layer further comprises the following steps: calculating a row offset and a column offset of the target detection point according to the target deformation amount; converting the row offset and the column offset into image brightness information respectively, and filling the image brightness information into the deformation amount image.

4. The positioning method of claim 1, wherein, The step of diffusing the deformation amount of the region to be repaired according to the boundary region further comprises: constructing a deformation amount diffusion equation according to the boundary region and the region to be repaired, the deformation amount diffusion equation being used to diffuse the deformation amount of the region to be repaired; solving the deformation amount diffusion equation to obtain the deformation amount image of the pyramid sub-image of the current layer.

5. The positioning method according to claim 4, characterized in that, The method further comprises: constructing the deformation amount diffusion equation according to a preset extremum geometry algorithm; iteratively updating the deformation amount diffusion equation so that the value of the deformation amount diffusion equation tends to zero; stopping the iterative updating when the value of the deformation amount diffusion equation is less than a preset threshold to obtain the deformation amount image of the pyramid sub-image of the current layer.

6. The positioning method of claim 1, wherein, After the step of obtaining the deformation amount image of the pyramid sub-image, the method further comprises: filtering out outliers in the deformation amount image with a frequency greater than a first frequency threshold based on a median filter; filtering out noise in the deformation amount image with a frequency greater than a second frequency threshold based on a smoothing filter.

7. A targeting distortion positioning system characterized by, The method comprises: an image acquisition module configured to acquire a to-be-positioned image, the to-be-positioned image comprising a to-be-positioned target; an image preprocessing module configured to construct an image pyramid according to the to-be-positioned image, the image pyramid comprising a plurality of pyramid sub-images with different resolutions; an image positioning module configured to search for a target detection point in the pyramid sub-image, the target detection point being a pixel point obtained based on image local template positioning; calculate a target deformation amount of the target detection point, and diffuse the target deformation amount to obtain a deformation amount image of the pyramid sub-image; map the deformation amount image of each pyramid sub-image in turn to the pyramid sub-image of the next layer in the direction from the top layer to the bottom layer of the image pyramid; search for a target detection point in the pyramid sub-image of the bottom layer, and position a target position of the to-be-positioned target according to the target detection point; The step of searching for a target detection point in the pyramid sub-image further comprises: acquiring an image local template; traversing pixel points in the pyramid sub-image of the top layer according to the image local template; searching for pixel points with a similarity greater than a similarity threshold to the image local template as the target detection point in the pyramid sub-image of the top layer; The step of searching for a target detection point in the pyramid sub-image further comprises: acquiring a deformation amount image of the pyramid sub-image of the previous layer, the deformation amount image comprising a row deformation image and a column deformation image, the row deformation image being used to represent a deformation amount in the row direction of a pixel point, and the column deformation image being used to represent a deformation amount in the column direction of a pixel point; positioning an initial detection point in the pyramid sub-image of the current layer according to the deformation amount image of the pyramid sub-image of the previous layer; searching for pixel points with a similarity greater than a similarity threshold to the image local template in a preset size neighborhood of the initial detection point as the target detection point in the pyramid sub-image of the current layer; The step of diffusing the target deformation amount further comprises: acquiring a deformation amount image of the pyramid sub-image of the previous layer; initializing the deformation amount image according to the target deformation amount of the target detection point in the pyramid sub-image of the current layer; The deformation amount image is divided into a boundary region and a region to be repaired, the boundary region is a region where the target detection point is located on the deformation amount image, and the region to be repaired is a region outside the boundary region on the deformation amount image; The deformation amount of the region to be repaired is diffused according to the boundary region to obtain a deformation amount image of the pyramid sub-image of the current layer.

Images