Dynamic fluid image registration method and system based on multi-frame sequence image, and medium

Abstract

The invention discloses a dynamic fluid image registration method and system based on multi-frame sequence images and a medium. According to the scheme, the method is improved on the basis of a traditional image registration technology, an infrared flow field image and a visible light flow field image are constructed into a co-occurrence filtering image pyramid, all scale layers of the co-occurrence filtering image pyramid are globally searched, and initial matching points of all the scale layers are found out; the coordinates of the initial matching points are recovered to the coordinates under the initial layer to form a fine matching point set; according to the scheme, nonlinear intensity distortion among different modal flow field images is overcome through a dynamic feature extraction process, global search matching is performed in combination with a three-dimensional feature descriptor of time sequence change, the influence of scale change is resisted, and the registration precision is improved; and mismatching caused by dynamic characteristics of the flow field image is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of image registration, and in particular to a dynamic fluid image registration method, system and medium based on multi-frame sequence images. Background Art

[0002] Image registration is the process of establishing a geometric transformation relationship between the pixel coordinates of two or more images, and then aligning the images with a common spatial coordinate system using the transformation parameters. It is widely used in many fields such as medical image analysis, intelligent transportation, visual navigation and positioning, change monitoring, computer vision, etc. Although image registration technology has been studied for a long time, due to the serious nonlinear intensity and geometric distortion caused by different imaging mechanisms, viewpoints, shooting times, etc., accurate image matching of multimodal images is far from being solved; in practice, infrared and visible light image registration has been widely studied compared with other multimodal images due to its good information complementarity.

[0003] At present, infrared and visible light image registration methods are mainly divided into region-based and feature-based image matching methods. Region-based image matching methods are highly dependent on the appropriate selection of similarity metrics, geometric transformation models and optimization methods, and are greatly affected by overlapping areas and image content; in the case of different modalities, due to severe nonlinear intensity distortion, it is difficult for region-based image matching methods to achieve accurate matching. Feature-based image matching methods usually start with feature extraction, then match features with / without feature descriptions, and then perform transformation model estimation, image resampling and distortion to achieve image registration. Compared with region-based image matching methods, feature-based image matching methods are more robust to geometric deformation and have a wider range of applications; SIFT, SURF and ORB are classic feature-based image matching methods, which were originally designed to process homologous images and therefore cannot handle the nonlinear intensity distortion problem of infrared and visible light images.

[0004] In addition, the current registration methods for infrared and visible light images are mainly designed for rigid objects. When dealing with objects without fixed shapes (such as various fluid objects), the limitations of this method become obvious. Fluid image registration is of great significance for remote sensing analysis and climate observation; however, image registration technology based on fixed objects is no longer applicable to fluid objects without fixed spatial coordinates, and in different modes, nonlinear radiation distortion will cause greater differences between the two, and current infrared and visible light image registration methods will find it difficult to complete accurate registration.

[0005] In summary, the current registration methods still have the following shortcomings:

[0006] (1) When there is large nonlinear intensity distortion in infrared images and visible light images, the existing algorithms have poor robustness;

[0007] (2) When it comes to fluid images without a fixed shape, the existing algorithms have a high mismatch rate and limitations. Summary of the invention

[0008] The technical problem to be solved by the present invention is that the current registration method has poor robustness when there is large nonlinear intensity distortion in infrared images and visible light images; and for fluid images without fixed shapes, the mismatch rate is high and has limitations; the purpose of the present invention is to provide a dynamic fluid image registration method, system and medium based on multi-frame sequence images, improve the method on the basis of traditional image registration technology, and realize dynamic feature extraction by constructing infrared flow field images and visible light flow field images into a co-occurrence filter image pyramid, thereby overcoming the nonlinear intensity distortion between flow field images of different modes, and combining with time-series changing three-dimensional feature descriptors for global search and matching, thereby resisting the influence of scale changes and improving the registration accuracy; and avoiding mismatching caused by dynamic features of flow field images.

[0009] The present invention is achieved through the following technical solutions:

[0010] This solution provides a dynamic fluid image registration method based on multi-frame sequence images, including:

[0011] Acquire infrared flow field images and visible light flow field images of flow field targets, and preprocess the infrared flow field images and visible light flow field images;

[0012] Based on the preprocessed infrared flow field image and visible light flow field image, a co-occurrence filter image pyramid is constructed, and the contour information of the flow field target is determined;

[0013] Dynamic feature extraction is performed based on the co-occurrence filter image pyramid to obtain a common feature point set of each scale layer; the co-occurrence filter image pyramid is composed of an initial layer and a multi-scale downsampling layer;

[0014] Construct a three-dimensional feature descriptor based on the common feature point set, perform a global search on each scale layer of the co-occurrence filter image pyramid, find the initial matching points of each scale layer, and restore the coordinates of the initial matching points to the coordinates under the initial layer to form a precise matching point set;

[0015] Output precise matching results based on FSC algorithm.

[0016] A further optimization scheme is that the co-occurrence filter image pyramid is constructed based on the preprocessed infrared flow field image and the visible light flow field image, including the following method:

[0017] Build an image pyramid;

[0018] Each layer of the image pyramid is filtered based on a co-occurrence filter; the co-occurrence filter satisfies the configuration:

[0019]

[0020] Among them, J p Represents the pixel value of the co-occurrence filter input image, with the pixel index being p; I q Represents the pixel value of the co-occurrence filter output image, and the pixel index is q; Represents the contribution weight of pixel index q to the output value corresponding to pixel index p; represents Gaussian filter; M(I p ,I q ) represents the calculation result of the co-occurrence matrix.

[0021] A further optimization scheme is that the image set SCoFPYD obtained by filtering satisfies:

[0022]

[0023] Among them, σ n Indicates the scale of the nth layer image; i indicates the number of times the initial layer is downsampled; N 0 represents the initial window size of the co-occurrence filter; Represents the nth layer image after filtering; SZ n Indicates the size of the co-occurrence statistics window at the nth layer.

[0024] A further optimization scheme is that the dynamic feature extraction based on the co-occurrence filter image pyramid is performed to obtain a common feature point set of each scale layer, including the following method:

[0025] Extracting consistent local phase information of a co-occurrence filter image pyramid based on a PC algorithm and generating a PC pyramid; the PC pyramid is composed of multi-scale PC image layers;

[0026] Extract feature points from multiple frames of continuous PC images at the same scale layer in the PC pyramid to form a first feature point set;

[0027] Perform differential processing on each PC image layer of the PC pyramid to obtain a sequential differential PC pyramid;

[0028] Extract feature points from the sequence difference PC pyramid to form a second feature point set;

[0029] The first feature point set and the second feature point set are combined to obtain a common feature point set.

[0030] A further optimization scheme is that the PC algorithm is used to extract consistent local phase information of the co-occurrence filter image pyramid; including the following method:

[0031] The consistent local phase information is extracted according to the following formula:

[0032]

[0033] Where I(x,y) represents a two-dimensional image signal; E so (x,y) represents the two-dimensional image signal and the even-symmetric Log-Gabor wavelet L even The first response component obtained by convolution; O so (x,y) represents the two-dimensional image signal and the odd-symmetric Log-Gabor wavelet L odd The second response component obtained by convolution; A so (x, y) represents the amplitude component of the two-dimensional image signal; Φ so (x, y) represents the phase component of the two-dimensional image signal; x represents the horizontal coordinate of the pixel point, and y represents the vertical coordinate of the pixel point; s and o represent the scale and direction of the Log-Gabor wavelet respectively;

[0034] The phase consistency value PC(x,y) obtained by integrating all directions and scales is:

[0035]

[0036] Among them, w o represents the weighting factor, ξ represents a constant; ΔΦ so (x,y) represents the phase deviation function, T represents the noise threshold, The operator prevents the enclosed quantity from taking a negative value. When * is positive, the enclosed quantity is the value of *; otherwise, the enclosed quantity is 0.

[0037] A further optimization scheme is that the three-dimensional feature descriptor is constructed based on the common feature point set, including the method:

[0038] The amplitude component A so (x,y) is summed up at all scales to construct the feature vector A o (x,y):

[0039]

[0040] Perform three-dimensional point sampling on the feature points of the current image, where the sampling points are distributed around the feature points, on the previous frame image of the current image, and on the subsequent frame image of the current image;

[0041] Based on Gaussian weighting, the feature vector A o (x, y) correspondence is integrated near the feature point and the sampling point to form the descriptor GFA;

[0042] The descriptors GFA of all feature points and sampling points are connected in order from top to bottom of the image frame to obtain the final vector, which is used as the feature descriptor.

[0043] A further optimization scheme is that the descriptor GFA is obtained according to the following formula:

[0044]

[0045] σ k =a*R k +b

[0046]

[0047] GFA k,j =[V k,j,1 ; V k,j,2 ;...;V k,j,o ]

[0048] Among them, C k The radius of the concentric circle where the sampling point is located, Gauss(x,y,σ) represents the Gaussian kernel with variance σ, σ k is the Gaussian kernel variance of the sampling point located at the kth concentric circle, a and b are constants, (x p ,y p ) and R k They represent the coordinates of the sampling point and the Gaussian weighted radius of the sampling point respectively, (k, j) represents the index and direction of the concentric circle where the sampling point is located, and o is the direction of the Log-Gabor wavelet.

[0049] A further optimization scheme is that the preprocessing includes the process of: performing frame rate synchronization processing on the infrared flow field image and the visible light flow field image.

[0050] The present solution also provides a dynamic fluid image registration system based on multiple frame sequence images, which is used to implement the above-mentioned dynamic fluid image registration method based on multiple frame sequence images; the system comprises:

[0051] A preprocessing module, used for acquiring an infrared flow field image and a visible light flow field image of a flow field target, and preprocessing the infrared flow field image and the visible light flow field image;

[0052] A construction module is used to construct a co-occurrence filter image pyramid based on the preprocessed infrared flow field image and the visible light flow field image, and determine the contour information of the flow field target;

[0053] A feature extraction module, used for performing dynamic feature extraction based on the co-occurrence filter image pyramid to obtain a common feature point set of each scale layer; the co-occurrence filter image pyramid is composed of an initial layer and a multi-scale downsampling layer;

[0054] A matching module is used to construct a three-dimensional feature descriptor based on a common feature point set, perform a global search on each scale layer of the co-occurrence filter image pyramid, find the initial matching points of each scale layer, and restore the coordinates of the initial matching points to the coordinates under the initial layer to form a precise matching point set;

[0055] The output module is used to output the precise matching results based on the FSC algorithm.

[0056] The present solution also provides a computer-readable medium on which a computer program is stored. The computer program is executed by a processor to implement the dynamic fluid image registration method based on multiple frame sequence images as described above.

[0057] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0058] 1. The present invention provides a dynamic fluid image registration method, system and medium based on multi-frame sequence images; based on the traditional image registration technology, the method is improved, and the infrared flow field image and the visible light flow field image are constructed into a co-occurrence filter image pyramid to realize dynamic feature extraction, resist the scale transformation between images, and combine the time-series changing three-dimensional feature descriptors for global search and matching, thereby improving the registration accuracy; avoiding the mismatch caused by the dynamic features of the flow field image.

[0059] 2. The present invention provides a dynamic fluid image registration method, system and medium based on multi-frame sequence images; by establishing a multi-frame co-occurrence filter image pyramid to extract edge features to resist scale transformation, the influence of nonlinear intensity distortion between visible light and infrared images is overcome based on the phase consistency algorithm, and at the same time, feature points are extracted on the multi-frame phase consistency image to construct a common feature point set, and a new feature descriptor is constructed in the form of three-dimensional point sampling to capture flow field change information, thereby realizing the registration of dynamic objects with no fixed morphology in the flow field. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0061] Figure 1 It is a framework diagram of a dynamic fluid image registration method based on multi-frame sequence images;

[0062] Figure 2 It is a flowchart of a dynamic fluid image registration method based on multi-frame sequence images;

[0063] Figure 3 It is a schematic diagram of the process of extracting the common feature point set;

[0064] Figure 4 Schematic diagram of the process of constructing a three-dimensional feature descriptor;

[0065] Figure 5 Schematic diagram of the multi-scale matching process;

[0066] Figure 6 This figure shows the registration results of the dynamic fluid image registration method based on multi-frame sequence images. DETAILED DESCRIPTION

[0067] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0068] The current registration method has poor algorithm robustness when there is large nonlinear intensity distortion in infrared images and visible light images; and has limitations due to high mismatch rate for fluid images without fixed shapes. In particular, there is still much room for improvement in the registration processing of non-steady-state targets such as floating clouds, water flow and flames; in view of this, this solution provides the following embodiments to solve the above technical problems:

[0069] Example 1

[0070] This embodiment provides a dynamic fluid image registration method based on multiple frame sequence images, such as Figure 1 and Figure 2 As shown, including:

[0071] Step 1: Acquire infrared flow field images and visible light flow field images of the flow field target, and preprocess the infrared flow field images and visible light flow field images; the preprocessing includes the process of: performing frame rate synchronization processing on the infrared flow field image and the visible light flow field image to reduce potential fluctuations.

[0072] Step 2: constructing a co-occurrence filter image pyramid based on the preprocessed infrared flow field image and the visible light flow field image, and determining the contour information of the flow field target; the method of constructing a co-occurrence filter image pyramid based on the preprocessed infrared flow field image and the visible light flow field image includes:

[0073] S21, construct an image pyramid; each image pyramid consists of an initial layer (a 0 ,b 0 ,c 0 ) and downsampling layer (a i ,b i ,c i ) is composed of 0Layer is the original image processed, b 0 is a 0 The layer is downsampled by a factor of 4 / 3, c 0 is a 0 The layer is downsampled by a factor of 3 / 2. i ,b i ,c i They are respectively a 0 ,b 0 ,c 0 The i-th layer is downsampled by a factor of 2, where i is the number of downsampling times of the initial layer;

[0074] S22, filtering each layer of the image pyramid based on a co-occurrence filter; the co-occurrence filter satisfies the configuration:

[0075]

[0076] Among them, J p Represents the pixel value of the co-occurrence filter input image, with the pixel index being p; I q G represents the pixel value of the co-occurrence filter output image, and the pixel index is q; σs (p,q)·M(I p ,I q ) represents the contribution weight of pixel index q to the corresponding output value of pixel index p; G σs (p,q) represents Gaussian filter; M(I p ,I q ) represents the calculation result of the co-occurrence matrix.

[0077] The image set SCoFPYD obtained by filtering satisfies:

[0078]

[0079] Among them, σ n Indicates the scale of the nth layer image; i indicates the number of times the initial layer is downsampled; N 0 represents the initial window size of the co-occurrence filter; Represents the nth layer image after filtering; SZ n Indicates the size of the co-occurrence statistics window at the nth layer.

[0080] Step 3: dynamic feature extraction is performed based on the co-occurrence filter image pyramid to obtain a common feature point set of each scale layer; the co-occurrence filter image pyramid is composed of an initial layer and a multi-scale downsampling layer; the dynamic feature extraction is performed based on the co-occurrence filter image pyramid to obtain a common feature point set of each scale layer, including the following method:

[0081] S31, extracting consistent local phase information of a co-occurrence filter image pyramid based on a PC algorithm, and generating a PC pyramid; the PC pyramid is composed of a multi-scale PC image layer; the method of extracting consistent local phase information of a co-occurrence filter image pyramid based on a PC algorithm includes:

[0082] The consistent local phase information is extracted according to the following formula:

[0083]

[0084] Where I(x,y) represents a two-dimensional image signal; E so (x,y) represents the two-dimensional image signal and the even-symmetric Log-Gabor wavelet L even The first response component obtained by convolution; O so (x,y) represents the two-dimensional image signal and the odd-symmetric Log-Gabor wavelet L odd The second response component obtained by convolution; A so (x, y) represents the amplitude component of the two-dimensional image signal; Φ so (x, y) represents the phase component of the two-dimensional image signal; x represents the horizontal coordinate of the pixel point, and y represents the vertical coordinate of the pixel point; s and o represent the scale and direction of the Log-Gabor wavelet respectively;

[0085] The phase consistency value PC(x,y) obtained by integrating all directions and scales is:

[0086]

[0087] Among them, w o represents the weighting factor, ξ represents a constant; ΔΦ so (x,y) represents the phase deviation function, T represents the noise threshold, The operator prevents the enclosed quantity from taking a negative value. When * is positive, the enclosed quantity is the value of *; otherwise, the enclosed quantity is 0.

[0088] S32, extracting feature points from multiple frames of continuous PC images at the same scale layer in the PC pyramid based on the Shi-Tomasi algorithm to form a first feature point set;

[0089] S33, performing differential processing on each PC image layer of the PC pyramid to obtain a sequence differential PC pyramid;

[0090] S34, extracting feature points from the sequence difference PC pyramid based on the Shi-Tomasi algorithm to form a second feature point set;

[0091] S35, combining the first feature point set and the second feature point set to obtain a common feature point set.

[0092] like Figure 3 As shown, the PC algorithm is used to extract highly consistent local phase information from the image pyramid, and the Shi-Tomasi algorithm is used to extract feature points on multiple frames of continuous PC image layers at the same scale, and then merged to form a common feature point set.

[0093] Step 4: construct a three-dimensional feature descriptor based on the common feature point set, perform a global search on each scale layer of the co-occurrence filter image pyramid, find the initial matching points of each scale layer, and restore the coordinates of the initial matching points to the coordinates under the initial layer to form a precise matching point set;

[0094] The principle of constructing a three-dimensional feature descriptor based on a common feature point set is as follows: Figure 4 As shown, including methods:

[0095] S41, construct the feature vector A by summing the amplitude components at all scale levels o (x,y,o):

[0096]

[0097] S42, perform three-dimensional point sampling on the feature points of the current image, and the sampling points are distributed around the feature points, on the previous frame image of the current image, and on the subsequent frame image of the current image; specifically, for the current image, the sampling points are located at n 1 concentric circles of different radii, each circle is divided into n 2 In each direction of each concentric circle, a sampling point is set. Then, this sampling model is extended to three dimensions. Multiple frames of continuous images are selected and the feature point set layer is placed in the middle. The upper two layers are the first two frames of the middle layer, and the lower two layers are the last two frames of the middle layer. The same point sampling model is used. The radius of the concentric circle can be calculated by the following formula:

[0098] C k+1 =C k +C 1 *k

[0099] Among them, C 1 , C k and C k+1 They represent the radii of the 1st, kth, and k+1th concentric circles respectively.

[0100] S43, based on Gaussian weighting, the feature vector A o (x, y) correspondence is integrated near the feature point and the sampling point to form the descriptor GFA;

[0101] S44, the descriptors GFA of all feature points and sampling points are connected in order from top to bottom of the image frame to obtain a final vector, and the final vector is used as a feature descriptor. The final vector is a 6*(n1 *n 2 +1)*frame, where frame indicates the number of multi-frame images used.

[0102] The descriptor GFA is obtained according to the following formula:

[0103]

[0104] σ k =a*R k +b

[0105]

[0106] GFA k,j =[V k,j,1 ; V k,j,2 ;...;V k,j,o ]

[0107] Where Gauss(*) represents the Gaussian kernel, σ k is the Gaussian kernel variance of the sampling point located at the kth concentric circle, a and b are constants, (x p ,y p ) and R k They represent the coordinates of the sampling point and the Gaussian weighted radius of the sampling point respectively, (k, j) represents the index and direction of the concentric circle where the sampling point is located, and o is the direction of the Log-Gabor wavelet.

[0108] Step 5: Output the precise matching result based on the FSC algorithm.

[0109] like Figure 5 As shown in , in order to detect all matching feature points in the scale space, this scheme performs a global search on the scale space. Feature matching is performed by traversing all image layers. The above process can achieve accurate registration of non-stationary images. The specific registration effect is shown in Figure 6 As shown in Figure 1, (a) is the fusion result after registration, (b) is the chessboard result, and (c) is the correct matching point correspondence result. Figure 6 It can be seen that no artifacts or misalignment occur in the fused image, and good alignment without misalignment can be achieved in the checkerboard image. The above results show that the registration of non-steady-state fluid images is achieved efficiently and accurately through the method of the present invention.

[0110] This scheme establishes a multi-frame co-occurrence filter image pyramid to extract edge features to resist scale transformation. Based on the phase consistency algorithm, it overcomes the influence of nonlinear intensity distortion between visible light and infrared images. At the same time, feature points are extracted on the multi-frame phase consistency map to construct a common feature point set. A new feature descriptor is constructed in the form of three-dimensional point sampling to capture flow field change information, thereby realizing the registration of dynamic objects with no fixed morphology in the flow field. This scheme extracts dynamic features and performs global search and matching with time-series changing three-dimensional feature descriptors to avoid mismatching caused by dynamic features of flow field images. This scheme introduces the PC algorithm to overcome the nonlinear intensity distortion between flow field images of different modes. A simple co-occurrence filter image pyramid and multi-scale matching strategy are designed to enable the algorithm to resist scale changes and improve registration accuracy.

[0111] Example 2

[0112] This embodiment provides a dynamic fluid image registration system based on multiple frame sequence images, which is used to implement the dynamic fluid image registration method based on multiple frame sequence images described in Embodiment 1; the system includes:

[0113] A preprocessing module, used for acquiring an infrared flow field image and a visible light flow field image of a flow field target, and preprocessing the infrared flow field image and the visible light flow field image;

[0114] A construction module is used to construct a co-occurrence filter image pyramid based on the preprocessed infrared flow field image and the visible light flow field image, and determine the contour information of the flow field target;

[0115] A feature extraction module, used for performing dynamic feature extraction based on the co-occurrence filter image pyramid to obtain a common feature point set of each scale layer; the co-occurrence filter image pyramid is composed of an initial layer and a multi-scale downsampling layer;

[0116] A matching module is used to construct a three-dimensional feature descriptor based on a common feature point set, perform a global search on each scale layer of the co-occurrence filter image pyramid, find the initial matching points of each scale layer, and restore the coordinates of the initial matching points to the coordinates under the initial layer to form a precise matching point set;

[0117] The output module is used to output the precise matching results based on the FSC algorithm.

[0118] Example 3

[0119] This embodiment provides a computer-readable medium having a computer program stored thereon. The computer program is executed by a processor to implement the dynamic fluid image registration method based on multiple frame sequence images as described in Embodiment 1.

[0120] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dynamic fluid image registration method based on multi-frame sequence images, characterized in that: include: Acquire infrared flow field images and visible light flow field images of flow field targets, and preprocess the infrared flow field images and visible light flow field images; Based on the preprocessed infrared flow field image and visible light flow field image, a co-occurrence filter image pyramid is constructed, and the contour information of the flow field target is determined; Dynamic feature extraction is performed based on the co-occurrence filter image pyramid to obtain a common feature point set of each scale layer; the co-occurrence filter image pyramid is composed of an initial layer and a multi-scale downsampling layer; Construct a three-dimensional feature descriptor based on the common feature point set, perform a global search on each scale layer of the co-occurrence filter image pyramid, find the initial matching points of each scale layer, and restore the coordinates of the initial matching points to the coordinates under the initial layer to form a precise matching point set; Output precise matching results based on FSC algorithm.

2. The dynamic fluid image registration method based on multi-frame sequence images according to claim 1, characterized in that: The method of constructing a co-occurrence filter image pyramid based on the preprocessed infrared flow field image and the visible light flow field image includes: Build an image pyramid; Each layer of the image pyramid is filtered based on a co-occurrence filter; the co-occurrence filter satisfies the configuration: Among them, J p Represents the pixel value of the co-occurrence filter input image, with the pixel index being p; I q Represents the pixel value of the co-occurrence filter output image, and the pixel index is q; Represents the contribution weight of pixel index q to the output value corresponding to pixel index p; represents Gaussian filter; M(I p ,I q ) represents the calculation result of the co-occurrence matrix.

3. The dynamic fluid image registration method based on multi-frame sequence images according to claim 2 is characterized in that: The image set ScoFSpace obtained by filtering satisfies: Among them, σ n represents the scale of the nth layer image; i represents the number of downsampling of the initial layer; N0 represents the initial window size of the co-occurrence filter; Represents the nth layer image after filtering; SZ n Indicates the size of the co-occurrence statistics window at the nth layer.

4. The dynamic fluid image registration method based on multi-frame sequence images according to claim 1, characterized in that: The method of extracting dynamic features based on the co-occurrence filter image pyramid to obtain a common feature point set of each scale layer includes: Extracting consistent local phase information of a co-occurrence filter image pyramid based on a PC algorithm and generating a PC pyramid; the PC pyramid is composed of multi-scale PC image layers; Extract feature points from multiple frames of continuous PC images at the same scale layer in the PC pyramid to form a first feature point set; Perform differential processing on each PC image layer of the PC pyramid to obtain a sequential differential PC pyramid; Extract feature points from the sequence difference PC pyramid to form a second feature point set; The first feature point set and the second feature point set are combined to obtain a common feature point set.

5. The dynamic fluid image registration method based on multi-frame sequence images according to claim 4 is characterized in that: The PC algorithm is used to extract consistent local phase information of the co-occurrence filter image pyramid; including the following method: The consistent local phase information is extracted according to the following formula: Where I(x,y) represents a two-dimensional image signal; E so (x,y) represents the two-dimensional image signal and the even-symmetric Log-Gabor wavelet L even The first response component obtained by convolution; O so (x,y) represents the two-dimensional image signal and the odd-symmetric Log-Gabor wavelet L odd The second response component obtained by convolution; A so (x, y) represents the amplitude component of the two-dimensional image signal; Φ so (x, y) represents the phase component of the two-dimensional image signal; x represents the horizontal coordinate of the pixel point, and y represents the vertical coordinate of the pixel point; s and o represent the scale and direction of the Log-Gabor wavelet respectively; The phase consistency value PC(x,y) obtained by integrating all directions and scales is: Among them, w o represents the weighting factor, ξ represents a constant; ΔΦ so (x,y) represents the phase deviation function, T represents the noise threshold, The operator prevents the enclosed quantity from taking a negative value. When * is positive, the enclosed quantity is the value of *; otherwise, the enclosed quantity is 0.

6. The dynamic fluid image registration method based on multi-frame sequence images according to claim 5, characterized in that: The method of constructing a three-dimensional feature descriptor based on a common feature point set includes: The amplitude component A so (x,y) is summed up at all scales to construct the feature vector A o (x,y): Perform three-dimensional point sampling on the feature points of the current image, where the sampling points are distributed around the feature points, on the previous frame image of the current image, and on the subsequent frame image of the current image; Based on Gaussian weighting, the feature vector A o (x, y) correspondence is integrated near the feature point and the sampling point to form the descriptor GFA; The descriptors GFA of all feature points and sampling points are connected in order from top to bottom of the image frame to obtain the final vector, which is used as the feature descriptor.

7. The dynamic fluid image registration method based on multi-frame sequence images according to claim 6, characterized in that: The descriptor GFA is obtained according to the following formula: s k =a*R k +b GFA k,j =[V k,j,1 ;V k,j,2 ;...;V k,j,o ] Among them, C k represents the radius of the concentric circle where the sampling point is located, Gauss(x,y,σ) represents the Gaussian kernel with variance σ, σ k is the Gaussian kernel variance of the sampling point located at the kth concentric circle, a and b are constants, (x p ,y p ) and R k They represent the coordinates of the sampling point and the Gaussian weighted radius of the sampling point respectively, (k, j) represents the index and direction of the concentric circle where the sampling point is located, and o is the direction of the Log-Gabor wavelet.

8. The dynamic fluid image registration method based on multi-frame sequence images according to claim 1, characterized in that: The preprocessing includes the following steps: performing frame rate synchronization processing on the infrared flow field image and the visible light flow field image.

9. A dynamic fluid image registration system based on multi-frame sequence images, characterized in that: Used to implement the dynamic fluid image registration method based on multiple frame sequence images as described in any one of claims 1 to 8; the system comprises: A preprocessing module, used for acquiring an infrared flow field image and a visible light flow field image of a flow field target, and preprocessing the infrared flow field image and the visible light flow field image; A construction module is used to construct a co-occurrence filter image pyramid based on the preprocessed infrared flow field image and the visible light flow field image, and determine the contour information of the flow field target; A feature extraction module, used for performing dynamic feature extraction based on the co-occurrence filter image pyramid to obtain a common feature point set of each scale layer; the co-occurrence filter image pyramid is composed of an initial layer and a multi-scale downsampling layer; A matching module is used to construct a three-dimensional feature descriptor based on a common feature point set, perform a global search on each scale layer of the co-occurrence filter image pyramid, find the initial matching points of each scale layer, and restore the coordinates of the initial matching points to the coordinates under the initial layer to form a precise matching point set; The output module is used to output the precise matching results based on the FSC algorithm.

10. A computer readable medium having a computer program stored thereon, characterized in that: The computer program is executed by a processor to implement the dynamic fluid image registration method based on multiple frame sequence images as described in any one of claims 1 to 8.

Images