A wide field of view video image change detection method and apparatus

Abstract

The application discloses a wide field of view video image change detection method and device, and the method comprises the following steps: using an improved adaptive fast guided filter for video image change detection; after filtering the multi-temporal video image, a logarithmic ratio operator considering neighborhood information is proposed, the mean ratio difference graph generation mode is improved, and a difference graph is generated; the improved MR image and the improved LR image are subjected to image fusion through discrete wavelet transform, and a fused difference graph is obtained; a soft threshold function is used to perform initial classification on the fused difference graph, and an initial change area and an unchanged area are obtained; a cumulative distribution function is used to compress the pixel value of the classification result from [0, 255] to [0, 1]; a super-fast robust constraint fuzzy C-Means clustering algorithm is proposed, and an improved adaptive median filter is used for denoising processing. The device comprises a processor and a memory.

Description

Technical Field

[0001] This invention relates to the field of image processing, and in particular to a method and apparatus for detecting changes in wide field-of-view video images. Background Technology

[0002] Video image change detection is a technique used to detect changes between multiple temporal images and the location of changed regions. It plays an important role in many fields, such as security and target detection. Based on whether label information is used, change detection algorithms can be divided into three types: supervised, semi-supervised, and unsupervised. Supervised change detection relies on supervised classification to determine changed and unchanged regions. In contrast, unsupervised change detection does not require any training samples, and therefore has been more widely used in change detection. However, in practical applications, the monitored areas are vast and the scenes are complex, with significant background noise and high similarity between moving targets and static backgrounds, making wide-field-of-view video image change detection a significant challenge. Based on the above analysis, researching wide-field-of-view video image change detection is crucial.

[0003] In recent years, significant progress has been made in methods for change detection in remote sensing images. However, these methods still have certain limitations for change detection in video images. Under wide field-of-view conditions, targets are typically only a few pixels in size and lack easily distinguishable features such as shape and texture. Furthermore, due to the vast monitoring area and complex scenes, targets are easily obscured by the background, resulting in extremely low signal-to-noise ratios. Current technologies aim to obtain the change areas in multi-temporal video images captured by surveillance equipment. Directly performing change detection on these images will reduce the accuracy of the results to some extent. In addition, difference map analysis is a crucial step in change detection, but common change detection methods are not suitable for video images under wide field-of-view conditions. This is mainly because complex environments and high-density random noise can affect the change detection results. Summary of the Invention

[0004] To address the problem that wide field-of-view conditions, complex environments, and high-density random noise can lead to poor visual quality and low signal-to-noise ratio in video images acquired by image sensors, making it difficult for monitoring systems to detect subtle changes, this invention proposes a wide field-of-view video image change detection method and device. The processed image improves image clarity. The proposed method not only effectively detects changing areas but also exhibits good robustness to unchanged areas, thus improving accuracy. Details are described below:

[0005] A first aspect: a method for detecting changes in wide-field-of-view video images, the method comprising:

[0006] An improved adaptive fast guided filter is used for change detection in video images;

[0007] After filtering multi-temporal video images, a logarithmic ratio operator that considers neighborhood information is proposed to improve the generation method of mean-ratio difference map and generate the difference map.

[0008] Discrete wavelet transform is used to fuse the improved MR image and the improved LR image to obtain a fusion difference map;

[0009] A soft thresholding function is used to perform initial classification on the fused difference map to obtain the initial changed and unchanged regions; the cumulative distribution function is used to compress the pixel values ​​of the classification results from [0,255] to [0,1].

[0010] We propose an ultrafast robust constrained fuzzy C-Means clustering algorithm and employ an improved adaptive median filter for denoising.

[0011] The logarithmic ratio operator is:

[0012]

[0013] Where α is 0.5, β is an adaptive parameter operator, and using the adaptive parameter operator is beneficial for generating difference maps under different scenarios. Y1(i,j) is the previous time-phase image, Y2(i,j) is the next time-phase image a few seconds later, i and j represent pixel coordinates, min represents taking the minimum gray value, and max represents taking the maximum gray value. The improved mean-ratio difference map generation method is as follows:

[0014]

[0015] The ultrafast robust constrained fuzzy C-Means clustering algorithm is as follows:

[0016] Initialize m = 2, iteration = 50, lb = 0, ub = 1, dim = 2, k = 2;

[0017] Adaptive median filtering is applied to the difference image to obtain a new difference image, and the gray-level histogram of the new difference image is calculated.

[0018] Obtain the initial cluster centers c of the dung beetle optimization algorithm, initialize the membership matrix u, and set the number of iterations t; update the cluster centers and update the membership matrix; until the cluster centers no longer change.

[0019] The fitness function of the dung beetle optimization algorithm is the minimum loss function of the fuzzy C-Means clustering algorithm.

[0020]

[0021]

[0022]

[0023]

[0024]

[0025] Where w and h are the width and height of the filtering window, respectively, k is the number of pixels in the filtering window, M(i,j) represents the average pixel value in the filtering window, T represents the threshold, F represents the noise intensity, N represents the number of noise points in the filtering window, and R represents the size of the filtering window.

[0026] The improved adaptive median filter is as follows:

[0027] Step A:

[0028] Calculate A1 = Z med -Z min Calculate A2 = Z med -Z max ;

[0029] If A1 > 0 and A2 < 0, then proceed to step B;

[0030] Otherwise, calculate M and T according to the formula; if |I(i,j)-M(i,j)|>T, then N=N+1;

[0031] Calculate F and R according to the formula, and repeat step A;

[0032] If R = 7, and A1 ≤ 0 and A2 ≥ 0, output Z. med ;Finish;

[0033] Step B:

[0034] First, calculate B1 = Z xy -Z min Calculate B2 = Z xy -Z max ;

[0035] If B1 > 0 and B2 < 0, then output Z. xy Otherwise, output Z. med .

[0036] In a second aspect, a wide-field-of-view video image change detection device, the device comprising: a processor and a memory, the memory storing program instructions, the processor calling the program instructions stored in the memory to cause the device to perform the method described in any one of the first aspects.

[0037] Third aspect, a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method described in any one of the first aspects.

[0038] The beneficial effects of the technical solution provided by this invention are:

[0039] 1. This invention is the first to use an adaptive fast guided filter for change detection in video images, which can effectively preserve image edges and retain image details, and its denoising effect is better than other methods; this invention can accurately and effectively detect subtle changes in video images under wide field of view. Compared with existing methods, this method has higher accuracy and robustness, and shorter running time.

[0040] 2. Inspired by log-ratio difference plots and mean-ratio difference plots, this invention uses improved log-ratio and improved mean-ratio operators to create different images, which improves the contrast of the difference images while preserving detailed information, and uses wavelet fusion to fuse the two difference images.

[0041] 3. This invention uses an improved adaptive median filter based on the fuzzy C-Means algorithm to denoise different noise densities, which improves the noise resistance of the image, preserves more image details, and solves the problems of the current FCM_S1 and FCM_S2 algorithms that do not denoise cleanly and the image is too smooth.

[0042] 4. This invention employs an improved adaptive median filter to cluster the gray-level histogram of the preprocessed image and statistically analyzes the gray-level histogram of the difference image, thereby accelerating the clustering process and reducing computational time complexity. Since the fuzzy C-Means clustering algorithm uses random initialization of centroids, the dung beetle optimization algorithm is used to generate the initial centroids, reducing the number of iterations of the fuzzy C-Means clustering algorithm and lowering the time complexity. Attached Figure Description

[0043] Figure 1 A flowchart of a wide field-of-view video image change detection method;

[0044] Figure 2 The result image shows the transformation detection of the proposed improved difference map fusion algorithm;

[0045] Figure 3 This is a graph showing the results of change detection in a specific scenario.

[0046] Among them, (a) and (b) are video images of the previous and next time phases, (c) is the ground truth map, and (d) is the change detection map of the proposed method. The two time phase maps are 5 seconds apart.

[0047] Figure 4 This is a graph showing the results of change detection in a specific scenario.

[0048] Among them, (a) and (b) are video images of the previous and next time phases, (c) is the truth map, and (d) is the change detection map of the proposed method. The two time phase maps are 8 seconds apart.

[0049] Figure 5 This is a graph showing the change detection results in a specific scenario.

[0050] Among them, (a) and (b) are video images of the previous and next time phases, (c) is the ground truth map, and (d) is the change detection map of the proposed method. The two time phase maps are 6 seconds apart.

[0051] Figure 6 This is an image showing the results of detecting artificial changes in a given scenario.

[0052] In this diagram, (a) represents the video image from the previous time period, (b) represents the video image from the time period after the target was added, (c) is the ground truth image, and (d) is the change detection image of the proposed method. The two time-phase images are 5 seconds apart.

[0053] Figure 7 The image shows the change detection results of the proposed method in an invariant scenario. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below.

[0055] Example 1

[0056] This invention provides a method for detecting changes in wide-field-of-view video images. (See also...) Figure 1 The method includes:

[0057] Step 1: This embodiment of the invention proposes to use an improved adaptive fast guided filter for change detection in video images. This filter can not only effectively preserve image edges, but also retain the detailed information of the image.

[0058] Step 2: After filtering the multi-temporal video images, a logarithmic ratio operator that considers neighborhood information is proposed to improve the generation method of the mean ratio difference map, so that the changed areas and unchanged areas have high contrast, and a difference map with more detailed information is generated and preserved.

[0059] Step 3: Wavelet transform is widely used in image fusion due to its multi-scale decomposition and excellent reconstruction capabilities. In this embodiment of the invention, discrete wavelet transform is used to fuse the improved MR image and the improved LR image, and the fused difference map is named D. LP ;

[0060] Step 4: Although most of the noise in the image is effectively suppressed by filtering, a small amount of noise may still cause false alarms in the monitoring system. First, a soft thresholding function is used to perform initial classification of the difference map to obtain the initial changed and unchanged areas.

[0061] Step 5: After obtaining the initial classification results, the cumulative distribution function is used to compress the pixel values ​​of the classification results from [0,255] to [0,1] to further overcome the influence of residual noise on the classification results;

[0062] Step 6: Propose an ultra-fast robust constrained fuzzy C-Means clustering algorithm. The fuzzy C-Means clustering algorithm uses an improved adaptive median filter for denoising, which denoises according to different noise densities, improves the noise resistance of the image, preserves more image details, and solves the problems of the current FCM_S1 and FCM_S2 algorithms that do not denoise cleanly and the image is too smooth.

[0063] In this embodiment of the invention, the gray-level histogram of the image after preprocessing with an improved adaptive median filter is used for clustering. Furthermore, the gray-level histogram of the difference image is statistically analyzed instead of the entire image, which speeds up the clustering process and reduces the computational time complexity. Since the fuzzy C-Means clustering algorithm uses random initial centroids, the dung beetle optimization algorithm is used to generate the initial centroids, which reduces the number of iterations of the fuzzy C-Means clustering algorithm and lowers the time complexity.

[0064] Example 2

[0065] The experimental subjects used in this embodiment of the invention are video images with a wide field of view acquired by an eagle-eye device. The technical solution in Embodiment 1 will be further described below with reference to the accompanying drawings. See the following description for details:

[0066] Step 1: Guided filtering;

[0067] Unlike change detection in remote sensing images, in wide-field-of-view video images, targets are typically only a few pixels in size and lack easily identifiable features such as shape and texture. Furthermore, due to the vast and complex monitoring area, targets can easily be obscured by the background, resulting in extremely low signal-to-noise ratios. Directly using these images would reduce the accuracy of change detection. Guided filters are linear smoothing filters with excellent edge-preserving properties. Compared to bilateral filters, the computational power of guided filters is unaffected by the filtering window when filtering images. Based on this characteristic, guided filters extract target features with high accuracy, effectively suppressing noise and smoothing image edges, achieving good image enhancement. Fast guided filters offer even higher algorithmic efficiency than guided filters. Therefore, using adaptive fast guided filtering for detail enhancement can effectively reduce noise while preserving image edge and detail information.

[0068] Regarding the guided filter, a locally linear model is first defined, which includes a guided image I, an input image p, and an output image q. The guided image I and the input image p are provided by the user. In the guided filter, the guided image I can be the same as or different from the input image q. Within this locally linear model, the relationship between the output image and the guided image is defined as linear:

[0069]

[0070] In this linear expression, the coefficient a k and b k In the box filter window ω k If the mean remains constant, and we differentiate both sides of the above linear expression, we have: Therefore, the above linear expression ensures that the edges of the output image q are similar to the edges of the guiding image I.

[0071] The fast guided filtering algorithm reduces the number of pixels by downsampling, based on guided filtering, and then calculates the weight value 'a' of a linear function. k The bias value b of the linear function k The average value is then upsampled to restore the original size. For example, if the scaling ratio is S, then the number of pixels after scaling down is N / S. 2 The time complexity is O(N / S) 2 ).

[0072] To better enhance the detail information of the image, the output image is redefined as shown in formula (2):

[0073] Q=q+γ(Iq) (2)

[0074] Detail enhancement of an image is achieved by defining the gradient of the output image as the gradient of the detail enhancement gradient, as shown in Equation (3):

[0075]

[0076] in, I represents the original image. The final result is:

[0077]

[0078] Step 2: Improved the method for generating the mean-to-difference plot;

[0079] In this invention, fast guided filtering is used to denoise wide-field video images Y1 and Y2 to obtain denoised images Y1 and Y2. Fast guided image filtering can remove some speckles from video images, but the residual speckle noise will still affect the subsequent change detection process. Compared with the difference operator and the ratio operator, the logarithmic ratio operator can convert multiplication and division operations into addition and subtraction operations through logarithmic operations. The change in noise properties makes it widely used in the acquisition of difference maps for change detection. This invention proposes a logarithmic ratio operator that considers neighborhood information, and the formula is shown in equation (5). The image mean ratio method replaces the texture feature value or gray value of the corresponding pixel by taking the average value of the neighboring pixels of the pixel. This method can suppress speckles of the shape of a single pixel. Its disadvantage is that it cannot effectively eliminate blocky speckles. The contrast between the changed area and the unchanged area is not high, which will affect the final change detection effect during the difference map analysis. On this basis, the mean ratio difference map generation method is improved so that the changed area and the unchanged area have a higher contrast, and the formula is shown in equation (6).

[0080]

[0081]

[0082] Here, α is set to 0.5, β is an adaptive operator used to avoid the case where the denominator is 0. Using an adaptive parameter operator is beneficial for generating difference maps under different scenarios. i and j represent pixel coordinates, min represents taking the minimum gray value, max represents taking the maximum gray value, lg is a logarithmic operation with base 10, Y1(i,j) is the previous time phase image, and Y2(i,j) is the next time phase image a few seconds later.

[0083] Step 3: Discrete Wavelet Transform (DWT);

[0084] Among them, wavelet transform is widely used in the field of image fusion due to its multi-scale decomposition and excellent reconstruction capabilities. DWT is one of the most commonly used wavelet transforms because of its time-domain and frequency-domain analysis capabilities, the independence of decomposed signals at different scales, and the directionality of decomposition.

[0085] In this embodiment of the invention, Discrete Wavelet Transform (DWT) is used to fuse the improved MR image and the improved LR image, and the fused difference map is named DLP.

[0086] Step 4: Use a soft thresholding function to perform initial classification on the difference map to obtain the initial changed and unchanged regions;

[0087] Most of the noise in the image was effectively suppressed through filtering, but a small amount of noise still caused false alarms in the monitoring system. First, a soft thresholding function was used to initially classify the difference map, obtaining the initial changed and unchanged regions. The formula is shown in equation (7):

[0088] S(x,θ)≡{sign(x)max|x|-θ,0}+θ (7)

[0089] Where x represents the pixel value of the difference map, and θ is the segmentation threshold of the soft thresholding function.

[0090] Step 5: Cumulative distribution function;

[0091] After obtaining the pre-classification results, the cumulative distribution function is used to compress the pixel values ​​of the classification results from [0,255] to [0,1], in order to further overcome the influence of residual noise on the classification results. The formula is as follows:

[0092]

[0093] Where erf is the error function used to solve the cumulative distribution function, and α is the function that adjusts the slope of the cumulative distribution function.

[0094] Step 6: A new fuzzy C-Means clustering algorithm;

[0095] In this study, this invention proposes a novel fuzzy C-Means clustering algorithm. This algorithm can remove noise of varying sizes and densities. When noise is low, a small filtering window can be used; when noise is high, a large filtering window can be used. This ensures effective clustering segmentation while also shortening the clustering time. Furthermore, to further reduce clustering complexity, gray-level histograms are used to cluster the filtered image, and the membership matrix is ​​obtained through iterative calculations. Fuzzy C-Means clustering randomly initializes the cluster centers. To reduce the number of iterations and shorten the clustering time, a dung beetle optimization algorithm is used to initialize the cluster centers.

[0096]

[0097]

[0098]

[0099]

[0100] In the filtering process, w and h are the width and height of the filtering window, respectively, k is the number of pixels in the filtering window, M(i,j) represents the average pixel value within the filtering window, and T represents the threshold. Here, F represents the noise intensity, N represents the number of noise points within the filtering window, and R represents the size of the filtering window.

[0101] This invention proposes an improved adaptive median filter, which introduces a noise intensity parameter F. This filter adaptively selects the window size based on the noise intensity, thereby effectively reducing the number of iterations required by the classic median filter. The filtering algorithm process is shown in steps A and B:

[0102] Step A:

[0103] Step 1: First, calculate A1 = Z med -Z min ;

[0104] Step 2: Next, calculate A2 = Z med -Z max ;

[0105] Step 3: If A1 > 0 and A2 < 0, then proceed to step B;

[0106] Step 4: Otherwise, calculate M and T according to formulas (9) and (10);

[0107] Step 5: If |I(i,j)-M(i,j)|>T, then N=N+1;

[0108] Step 6: Calculate F and R according to formulas (11) and (12), and repeat step A;

[0109] Step 7: If R = 7, and A1 ≤ 0 and A2 ≥ 0, output Z. med ;

[0110] Step 8: End.

[0111] Step B:

[0112] Step 1: First, calculate B1 = Z xy -Z min ;

[0113] Step 2: Next, calculate B2 = Z xy -Z max ;

[0114] Step 3: If B1 > 0 and B2 < 0, then output Z xy ;;

[0115] Step 4: Otherwise, output Z med ;

[0116] In the above steps, in the embodiments of the present invention, the symbol S xy represents the area covered by the filtering window, Z min represents the minimum gray value of the filtering window, Z max represents the maximum gray value of the filtering window, Z med represents the median of the filtering window, Z xy represents the gray value of the pixel at the x-th row and y-th column.

[0117] The embodiments of the present invention propose a new objective function, aiming to improve the robustness to noise and reduce the running time of the algorithm, which is defined as follows:

[0118]

[0119] where, r i is the number of pixel points with the gray value of i, [[ID=3​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​The dung beetle optimization algorithm is a recently proposed swarm intelligence optimization algorithm. Since the fuzzy C-Means clustering algorithm randomly initializes cluster centers, it can lead to the formation of local optima during the clustering process. Therefore, this embodiment of the invention improves the initialization process of its cluster centers by utilizing the global search capability of the dung beetle optimization algorithm to generate the initial cluster centers for the fuzzy C-Means clustering algorithm, and repeats this process until k cluster centers are selected. As shown in equation (18), the fitness function of the dung beetle optimization algorithm is the minimum loss function of the fuzzy C-Means clustering algorithm:

[0129]

[0130] Based on the above analysis, the present invention proposes an ultrafast robust constrained fuzzy C-means clustering algorithm.

[0131] The steps are summarized as follows:

[0132] Step 1: Initialize m=2, iteration=50, lb=0, ub=1, dim=2, k=2;

[0133] Step 2: Perform adaptive median filtering on the difference image to obtain a new difference image, and calculate the gray-level histogram of the new difference image;

[0134] Step 3: Obtain the initial cluster centers c of the dung beetle optimization algorithm, initialize the membership matrix u, and set the number of iterations t;

[0135] Step 4: Update cluster centers and update membership matrix;

[0136] Step 5: Repeat the steps in Step 4 until the cluster centers no longer change.

[0137] Example 3

[0138] A wide-field-of-view video image change detection device includes a processor and a memory. The memory stores program instructions, and the processor calls the program instructions stored in the memory to cause the device to perform the following method steps in Embodiment 1:

[0139] An improved adaptive fast guided filter is used for change detection in video images;

[0140] After filtering multi-temporal video images, a logarithmic ratio operator that considers neighborhood information is proposed to improve the generation method of mean-ratio difference map and generate the difference map.

[0141] Discrete wavelet transform is used to fuse the improved MR image and the improved LR image to obtain a fusion difference map;

[0142] A soft thresholding function is used to perform initial classification on the fused difference map to obtain the initial changed and unchanged regions;

[0143] The cumulative distribution function is used to compress the pixel values ​​of the classification results from [0,255] to [0,1].

[0144] We propose an ultrafast robust constrained fuzzy C-Means clustering algorithm and employ an improved adaptive median filter for denoising.

[0145] The logarithmic ratio operator is:

[0146]

[0147] Where α is 0.5, β is an adaptive parameter operator, and using the adaptive parameter operator is beneficial for generating difference maps in different scenarios. Y1(i,j) is the previous time phase image, Y2(i,j) is the next time phase image a few seconds later, i and j represent pixel coordinates, min represents taking the minimum gray value, and max represents taking the maximum gray value.

[0148] The improved method for generating the mean-to-difference plot is as follows:

[0149]

[0150] The ultrafast robust constrained fuzzy C-Means clustering algorithm is as follows:

[0151] Initialize m = 2, iteration = 50, lb = 0, ub = 1, dim = 2, k = 2;

[0152] An adaptive median filter is applied to the difference image to obtain a new difference image, and the gray-level histogram of the new difference image is calculated.

[0153] Obtain the initial cluster centers c of the dung beetle optimization algorithm, initialize the membership matrix u, and set the number of iterations t;

[0154] Update the cluster centers and the membership matrix until the cluster centers no longer change.

[0155] The fitness function of the dung beetle optimization algorithm is the minimum loss function of the fuzzy C-Means clustering algorithm.

[0156]

[0157]

[0158]

[0159]

[0160]

[0161] Where w and h are the width and height of the filtering window, respectively, k is the number of pixels in the filtering window, M(i,j) represents the average pixel value in the filtering window, T represents the threshold, F represents the noise intensity, N represents the number of noise points in the filtering window, and R represents the size of the filtering window.

[0162] The improved adaptive median filter is as follows:

[0163] Step A:

[0164] Calculate A1 = Z med -Z min Calculate A2 = Z med -Z max ;

[0165] If A1 > 0 and A2 < 0, then proceed to step B;

[0166] Otherwise, calculate M and T according to the formula; if |I(i,j)-M(i,j)|>T, then N=N+1;

[0167] Calculate F and R according to the formula, and repeat step A;

[0168] If R = 7, and A1 ≤ 0 and A2 ≥ 0, output Z. med ;Finish;

[0169] Step B:

[0170] First, calculate B1 = Z xy -Z min Calculate B2 = Z xy -Z max ;

[0171] If B1 > 0 and B2 < 0, then output Z. xy Otherwise, output Z. med .

[0172] It should be noted that the device descriptions in the above embodiments correspond to the method descriptions in the embodiments, and the embodiments of the present invention will not be repeated here.

[0173] The execution entities of the aforementioned processor and memory can be devices with computing functions such as computers, microcontrollers, and single-chip microcomputers. In specific implementations, the embodiments of the present invention do not limit the execution entities and can select them according to the needs of actual applications.

[0174] Data signals are transmitted between the memory and the processor via a bus, which will not be elaborated upon in this embodiment of the invention.

[0175] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium, the storage medium including a stored program, which, when the program is running, controls the device where the storage medium is located to execute the method steps in the above embodiments.

[0176] The computer-readable storage medium includes, but is not limited to, flash memory, hard disk, solid-state drive, etc.

[0177] It should be noted that the description of the readable storage medium in the above embodiments corresponds to the description of the method in the embodiments, and the embodiments of the present invention will not be repeated here.

[0178] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated.

[0179] A computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in or transmitted through a computer-readable storage medium. A computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic or semiconductor, etc.

[0180] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0181] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting changes in wide-field-of-view video images, characterized in that, The method includes: An improved adaptive fast guided filter is used for change detection in video images; After filtering multi-temporal video images, a logarithmic ratio operator that considers neighborhood information is proposed to improve the generation method of mean-ratio difference map and generate the difference map. Discrete wavelet transform is used to fuse the improved MR image and the improved LR image to obtain the fusion difference map; A soft thresholding function is used to perform initial classification on the fused difference map to obtain the initial changed and unchanged regions; The cumulative distribution function is used to compress the pixel values ​​of the classification results from [0,255] to [0,1]. An ultrafast robust constrained fuzzy C-Means clustering algorithm is proposed, and an improved adaptive median filter is used for denoising. The logarithmic ratio operator is: ； in, It is 0.

5. For adaptive parameter operators, using adaptive parameter operators is beneficial for generating difference maps in different scenarios. This is an image from the previous time phase. This is an image of the next time phase, several seconds later. and This represents the pixel coordinates; min represents taking the minimum grayscale value, and max represents taking the maximum grayscale value. The ultrafast robust constrained fuzzy C-Means clustering algorithm is as follows: initialization ; An adaptive median filter is applied to the difference image to obtain a new difference image, and the gray-level histogram of the new difference image is calculated. Obtain the initial cluster centers of the dung beetle optimization algorithm. Initialize the membership matrix Set the number of iterations t; Update the cluster centers and the membership matrix; continue until the cluster centers no longer change. The fitness function of the dung beetle optimization algorithm is the minimum loss function of the fuzzy C-Means clustering algorithm: ； ； ； ； ； in, and These are the width and height of the filtering window, respectively. The number of pixels in the filter window. This represents the average pixel value within the filter window. Indicates the threshold. Indicates noise intensity. This indicates the number of noise points within the filtering window. Indicates the size of the filtering window. Indicates a guide image. Indicates the output image. Is the grayscale value The number of pixels, It is grayscale value Relative to the number of clusters Fuzzy membership degree, It is a weighting index for controlling ambiguity. It is the median gray value of the neighboring pixels after adaptive filtering. This represents the first gray-level histogram after statistical analysis. 1 pixel Indicates the first Cluster centers.

2. The wide field-of-view video image change detection method according to claim 1, characterized in that, The improved mean-to-difference map generation method is as follows: 。 3. The wide field-of-view video image change detection method according to claim 1, characterized in that, The improved adaptive median filtering is as follows: Step A: calculate ;calculate ; if and Then proceed to step B; Otherwise, calculate according to the formula. and ;if , ; Calculate according to the formula and Repeat step A; if ,and and Output ;Finish; Step B: First, calculate ;calculate ; if and Then output Otherwise output ; This represents the minimum grayscale value within the filter window. This represents the maximum grayscale value in the filter window. This represents the median of the filter window. This represents the grayscale value of the pixel in the x-th row and y-th column.

4. A wide field-of-view video image change detection device, characterized in that, The device includes a processor and a memory, the memory storing program instructions, the processor invoking the program instructions stored in the memory to cause the device to perform the method according to any one of claims 1-3.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method described in any one of claims 1-3.

Images