Method and device for detecting small target change of video image under large field of view at night
Through adaptive superpixel reconstruction and multi-scale singular value decomposition and fusion algorithm, combined with noise removal technology, the noise problem in video image change detection under large field of view at night is solved, and the accuracy and robustness of change detection are significantly improved.
Patent Information
- Application Number
- CN202411030050.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-07-30
AI Technical Summary
Under the large field of view at night, the monitoring video images have problems of complex random noise and low signal-to-noise ratio, which makes it difficult for traditional algorithms to effectively detect small target changes.
Adaptive superpixel reconstruction and multi-scale singular value decomposition and fusion algorithm are used to denoise and fusion the difference graph, combined with Anscombe root transformation and full variation regularization to remove Poisson and Gaussian noise, and use the FCM clustering algorithm to obtain complete detailed structures. Change detection results.
The quality of the difference graph is improved, the edge characteristics of the changing area are significantly enhanced, noise interference is suppressed, the error detection rate is reduced, and the accuracy and robustness of the change detection are improved.
Smart Images

Figure CN119027702A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of nighttime large-field-of-view video image changes, and in particular to a method and device for detecting small target changes in nighttime video images with a large field of view. Background Art
[0002] With the development of technology and the needs of social governance, surveillance equipment has been widely popularized. In conventional scenarios, the use of video image change detection algorithms to detect changes in surveillance video images has become very mature. However, in a nighttime large field of view environment, surveillance video images have complex random noise and low signal-to-noise ratio, making it difficult for people to detect small targets that have changed. Therefore, studying change detection algorithms suitable for nighttime large fields of view has become a very important topic.
[0003] At present, the research work on night-time large-field-of-view video image processing mainly focuses on the acquisition of large-field-of-view images and night-time image enhancement, while the research on change detection of video images under night-time large-field-of-view is relatively insufficient. For the change detection task, the biggest challenge is to suppress the noise in the difference map, which is generated by comparing multi-phase images and represents the degree of change of each pair of corresponding pixels. Due to the different and complex types of noise in the difference map of Hawkeye surveillance video images under night-time large-field-of-view, the difference map generated by traditional subtraction operators and ratio operators often cannot truly reflect the trend of change, which poses a challenge to extracting effective change targets from the difference map submerged by noise.
[0004] In order to solve the above problems, it is necessary to improve the quality of the difference map to better distinguish the changed area from the unchanged area. Summary of the invention
[0005] The present invention provides a method and device for detecting small target changes in video images under a large field of view at night. The present invention enhances the two denoised difference maps by adaptive superpixel reconstruction, effectively enhances the edge features of the difference maps, and fuses the two difference maps by a multi-scale singular value decomposition fusion algorithm to obtain a robust difference map. Finally, a clustering algorithm is used to obtain a change detection result with complete detail structure, as described below:
[0006] In the first aspect, a method for detecting small target changes in a video image under a large field of view at night, the method comprising:
[0007] The comprehensive Anscombe root transform and total variation regularization are used to remove the residual Poisson noise in the difference map, and the denoised image is obtained by inverse transformation;
[0008] The residual Gaussian noise in the denoised image is removed by using total variation regularization based on partial differential equations and minimizing the energy functional; the primal-dual algorithm is used to solve the energy functional;
[0009] The difference map is processed by a multi-scale superpixel segmentation and reconstruction method based on statistical information to enhance the edge features of the changed area and suppress the interference of noise;
[0010] The two difference images are fused based on multi-scale singular value decomposition fusion, and the FCM clustering algorithm is used to segment the fused difference image to obtain the detection result.
[0011] Among them, the Anscomb e The forward transform and inverse transform of the root transform are:
[0012]
[0013]
[0014] Among them, DI k is the original image, X k is the image after the forward transformation of Anscombe transformation, X k - 1 is the image after the inverse Anscombe transform.
[0015] The primal-dual algorithm is used to solve the energy functional:
[0016] Alternately update the original variable and the dual variable p, and update them according to the gradient descent principle of the regularized model. The original variable and the update formula are:
[0017]
[0018] Among them, I k+1 is the updated image, is the gradient of the image, λ is the Lagrange multiplier, I 0 is the original noise image, and the update formula of the Lagrange multiplier λ is:
[0019]
[0020]
[0021] Among them, I gradient is the size of the average gradient of the image, reflecting the overall gradient level of the image, p i,x and p i,y are the horizontal and vertical components of the gradient vector p, corresponding to the gradients in the x and y directions of the image, respectively;
[0022] Then the dual variable p is updated, and the update formula of the dual variable p is:
[0023]
[0024] Where σ is the step size parameter, is the updated image gradient.
[0025] The multi-scale superpixel segmentation and reconstruction method based on statistical information processes the difference map as follows:
[0026] Given a set of pixel values in a superpixel block {x 1 .x 2 , ..., x n}, n is the number of pixel values, and the reconstruction formula is:
[0027] I R =(w std ×O)+(w mean ×M)+(w median ×N)
[0028] Among them, w std Represents the weight of the original pixel, w mean represents the weight of the mean, w median Represents the weight of the median, O is the original pixel value inside the superpixel block, M is the mean value inside the superpixel block, and N is the median value inside the superpixel block.
[0029] The fusion of the two difference images based on multi-scale singular value decomposition fusion is as follows:
[0030] The selection rules for U and V for each layer are as follows:
[0031]
[0032]
[0033] For diagonal matrices, select larger singular values for fusion. The selection rule is:
[0034] ∑ fused =max(∑ 1 ,∑ 2 )
[0035] For Contrast1 and Contrast2, the global standard deviation of different layers of the Laplacian pyramid is calculated to obtain μ 1 k and μ 2 k is the average value of the current layer image, m and n are the number of rows and columns of the image respectively;
[0036]
[0037]
[0038] The final fusion image expression obtained by MSVD is:
[0039] Y k =U fused k ×∑ fused k ×(V fused k ) T
[0040]
[0041] Among them, K is the number of decomposition layers of the Laplacian pyramid, and Y is the final fused difference map.
[0042] In the second aspect, a device for detecting changes in small targets in video images at night with a large field of view is provided, the device comprising: a processor and a memory, the memory storing program instructions, the processor calling the program instructions stored in the memory to enable the device to execute any one of the methods described in the first aspect.
[0043] In a third aspect, a computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes any one of the methods described in the first aspect.
[0044] The beneficial effects of the technical solution provided by the present invention are:
[0045] 1. The present invention adopts a regularized denoising method with dynamically adjusted parameters to remove the noise in the difference map. This method achieves a good balance between image detail preservation and denoising performance, and its denoising effect is better than other methods;
[0046] 2. The present invention designs a new superpixel reconstruction method to enhance the difference map, making the difference between the changed area and the unchanged area more obvious. The difference map obtained by this method can suppress noise and background and highlight the changed pixels, reducing the probability of false detection;
[0047] 3. The present invention proposes a new difference map fusion method based on multi-scale singular value decomposition. This method can effectively overcome the influence of most random noises and more prominently capture the changing parts by selecting different fusion rules. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 A schematic diagram of a method for detecting small target changes in video images at night with a large field of view;
[0049] Figure 2Flowchart of the proposed change detection scheme;
[0050] Figure 3 Schematic diagram of the proposed MSVD fusion algorithm;
[0051] Figure 4 This is a schematic diagram of the change detection results of high-resolution video images;
[0052] Among them, (a) is the phase 1 image; (b) is the phase 2 image; (c) is the change detection result after processing by this algorithm.
[0053] Figure 5 Schematic diagram of change detection results for low-resolution video images.
[0054] Among them, (a) is the phase 1 image; (b) is the phase 2 image; (c) is the change detection result after processing by this algorithm. DETAILED DESCRIPTION
[0055] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in further detail below.
[0056] 1. Regularization-based Difference Map Denoising
[0057] The quality of the difference map directly affects the final change detection result. However, due to the influence of noise, the difference map without further processing often cannot correctly reflect the trend of real changes. For the night video images under the Eagle Eye surveillance camera, the noise in the difference map mainly comes from the surveillance camera itself, including: dark current, shot noise generated by light that follows Poisson distribution, and Gaussian noise with zero mean generated by sensor heat during imaging. In the case of weak light, due to the small number of photons, the randomness of photons will be more significant, resulting in a greater impact of Poisson noise in the image.
[0058] Through the above analysis, the embodiment of the present invention adopts a comprehensive Anscomb e Root transformation and total variation regularization methods remove the residual Poisson noise in the difference map. Anscomb e Transformation is a commonly used data transformation method in statistics. This method converts Poisson noise into approximate Gaussian noise with stable variance, and then uses the denoising method of removing additive Gaussian noise to remove the noise, and finally obtains the denoised image through inverse transformation. e The forward transform and inverse transform formulas are shown in equations (1) and (2):
[0059]
[0060]
[0061] Among them, DI k is the original image, X k is the image after the forward transformation of Anscombe transformation, X k - 1 is the image after the inverse Anscombe transform.
[0062] Subsequently, the embodiment of the present invention uses a total variation regularization denoising method based on partial differential equations to remove the Gaussian noise remaining in the difference map. The total variation model is an anisotropic model that relies on gradient descent flow to smooth the image, which can better protect details such as image edge texture. The principle of denoising is to minimize the energy function, and the energy function is shown in formula (3):
[0063]
[0064] The first term in the above formula is a regularization term that describes the smoothness of the image. The second term (λ∫|II 0 | 2 dΩ) is the data fidelity term, I is the denoised image to be obtained, and I 0 is the original noisy image, the data fidelity term represents the similarity between the denoised image and the original image, and the Lagrange multiplier λ is used to balance the regularization term and the data fidelity term.
[0065] Next, the embodiment of the present invention uses the primal-dual algorithm to solve the above energy functional. In each iteration, the embodiment of the present invention alternately updates the primal variable and the dual variable p, and updates them according to the gradient descent principle of the regularized model. The specific formula for updating the primal variable is shown in (4):
[0066]
[0067] Among them, I k+1 is the updated image, is the gradient of the image, λ is the Lagrange multiplier, I 0 is the original noise image. Here, the embodiment of the present invention adds a process of dynamically updating the Lagrange multiplier λ, where λ is the average gradient value calculated based on the gradient of the image in the current iteration step, and then the initial λ parameter is divided by the average gradient value to adjust the value of λ. The update formula of the Lagrange multiplier λ is:
[0068]
[0069]
[0070] Among them, I gradientis the size of the average gradient of the image, reflecting the overall gradient level of the image, p i,x and p i,y are the horizontal and vertical components of the gradient vector p, corresponding to the gradients in the x and y directions of the image, respectively.
[0071] Then the dual variable p is updated. The update formula of the dual variable p is shown in formula (7):
[0072]
[0073] Where σ is the step size parameter, is the updated image gradient.
[0074] 2. Difference Map Enhancement Based on Adaptive Superpixel Reconstruction
[0075] After denoising the difference map, the embodiment of the present invention uses a multi-scale superpixel segmentation and reconstruction method based on statistical information to further process the difference map, the purpose of which is to enhance the edge features of the change area and further suppress the interference of noise. First, the SLIC (Simple Linear Iterative Clustering) superpixel segmentation method is used to segment the image into multiple homogeneous superpixels on the denoised difference map. In each superpixel block, a weight combination method is designed to use the original pixel value, mean and standard deviation in each superpixel block to perform weighted superpixel block reconstruction, give different weights to each information, and complete the reconstruction of the superpixel block. The weight is determined according to the standard deviation of the pixel value inside the superpixel block. The larger the standard deviation, the more dispersed the distribution of the pixel value, which means that the variability of the pixel value is greater. Therefore, the embodiment of the present invention hopes to retain more original pixel values. On the contrary, if the standard deviation is small, it means that the variability of the pixel value is small. The embodiment of the present invention is more inclined to use the mean or median instead of the original pixel value. According to the calculated weights, the pixel values of the superpixel blocks are recalculated using weighted averaging. This method can reduce possible noise effects by using the mean and median while retaining the original image information.
[0076] Given a set of pixel values in a superpixel block {x 1 .x 2 , ..., x n}, where n is the number of pixel values. The reconstruction formula is shown in equation (8):
[0077] I R =(w std ×O)+(w mean ×M)+(w median ×N)#(8)
[0078] Among them, Wstd Represents the weight of the original pixel, w mean represents the weight of the mean, w median Represents the weight of the median, O is the original pixel value inside the superpixel block, M is the mean value inside the superpixel block, and N is the median value inside the superpixel block.
[0079] Among them, w std It is used to reflect the degree of change of pixel values in the superpixel block. When the standard deviation is large, w std The value of will be close to 1, indicating that the pixel values in the superpixel block are more dispersed, and more original pixel values will be retained. When the standard deviation σ is small, w std The value of will be close to 0, indicating that the pixel values in the superpixel block are more concentrated. At this time, the mean and median will be considered more to reduce the possible impact of noise.
[0080] In the embodiment of the present invention, in view of the high resolution of the image, four different segmentation scales are selected, namely 4000, 8000, 16000 and 32000. The image reconstruction operation is performed for the superpixel segmentation result at each scale. Finally, the reconstruction results at multiple scales are summed and averaged to obtain the final reconstructed image. This processing method can comprehensively consider the image details and features at different scales, thereby improving the quality and stability of image reconstruction. The formula for the final reconstruction result is:
[0081]
[0082] Wherein, S is the number of segmentation scales, which is set to 4 in the embodiment of the present invention. R t is the reconstructed image completed by formula (9) at different scales.
[0083] 3. Difference map fusion based on multi-scale singular value decomposition fusion
[0084] In order to solve the problem of insufficient robustness of a single difference map, the embodiment of the present invention uses a multi-scale singular value decomposition (MSVD) fusion method to fuse two difference maps. The MSVD algorithm is similar to wavelet transform. The basic idea is to use singular value decomposition instead of filtering at each layer of the smooth component. First, the original image is decomposed into images of multiple scales. In the embodiment of the present invention, the Laplace pyramid is used to decompose the difference map into images at three different scales. Then, singular value decomposition (SVD) is performed on the image at each scale. SVD is a matrix decomposition method that represents a matrix as the product of three matrices. The SVD decomposition process of the Laplace pyramid images at two different scales is shown in (10) and (11):
[0085] L 1(k) =U 1 (k) ∑ 1 (k) (V 1 k ) T #(10)
[0086] L 2 (k) =U 2 (k) ∑ 2 (k) (V 2 k ) T #(11)
[0087] Among them, L 1 (k) and L 2 (k) Yes DI 1 and DI 2 The image at the kth scale obtained by Laplacian pyramid sampling, K is the scale of the total sampling, where U 1 (k) and U 2 (k) are two orthogonal matrices, containing L 1 (k) and L 2 (k) The left singular vector of 1 (k) and∑ 2 (k) is a diagonal matrix, (V 1 k ) T and (V 2 k ) T is the transpose of two orthogonal matrices, which contain the image L 1 (k) and L 2 (k) The right singular vectors of can be decomposed by SVD to obtain their respective basic spatial features. Among them, the diagonal matrix saves the main information of the image, while the left and right singular matrices save the structural information of the image.
[0088] The selection rules for U and V of each layer are shown in equations (12) and (13). Since a larger singular value can better preserve the useful information of the original image, the embodiment of the present invention selects a larger singular value for fusion in the diagonal matrix. The selection rule is shown in equation (14):
[0089]
[0090]
[0091] ∑ fused =max(∑ 1 ,∑ 2 )#(14)
[0092] For Contrast1 and Contrast2, the embodiment of the present invention is obtained by calculating the global standard deviation of different layers of the Laplacian pyramid, μ 1 k and μ 2 k is the average value of the current layer image, m and n are the number of rows and columns of the image respectively.
[0093]
[0094]
[0095] The final fusion image expression obtained by MSVD is shown in equations (17) and (18):
[0096] Y k =U fused k ×∑ fused k ×(V fused k ) T #(17)
[0097]
[0098] Wherein, K is the number of decomposition layers of the Laplacian pyramid. In the embodiment of the present invention, K is 3, and Y is the final fused difference map.
[0099] Finally, the embodiment of the present invention uses the FCM clustering algorithm to segment the fused difference map to obtain the detection result.
[0100] experiment
[0101] In order to verify that this method is effective in detecting changes in nighttime large-field-of-view video images, experiments were conducted on nighttime large-field-of-view video image datasets at two resolutions. The datasets were shot multiple times in 25 different scenes at night using a camera fixed on a bracket. Finally, two groups were selected in each scene, each consisting of two frames of images, and 50 different image pairs were obtained. The initial resolution of each image pair was 1920×1080. Subsequently, the height and width of the 50 groups of images were reduced to half of the original by image downsampling, and another 50 groups of image pairs were obtained. The resolution of these 50 groups of image pairs was 960×540. A total of 100 groups of image pairs were obtained, of which 50 groups had a resolution of 1920×1080 and the other 50 groups had a resolution of 960×540.
[0102] Figure 1 In the paper, according to the description of the problem that small targets that change under a large field of view at night are difficult to detect, an adaptive superpixel reconstruction method is first used to reconstruct the two denoised difference maps by selecting different segmentation parameters. Then a multi-scale singular value fusion method is used to fuse the two difference maps. Multi-scale singular value fusion selects fusion rules at different scales and uses the complementary information of different difference maps to obtain a robust difference map. Finally, the FCM clustering algorithm is used to obtain the final changed image.
[0103] from Figure 4 and Figure 5 It can be seen that this method can not only effectively detect the changing target, but also will not cause blurring of moving objects. Calculate false negatives (FN) (undetected changed pixels) and false positives (FP) (unchanged pixels are mistakenly detected as changed). Overall error (OE), classification accuracy (PCC), Kappa coefficient, F1 value (F1). Among them, the classification accuracy (PCC) represents the accuracy of classification, the Kappa coefficient is to evaluate the consistency of the algorithm, and the F1 value (F1) is an important indicator for evaluating classification performance. This method measures the quality of the algorithm detection results through these six objective indicators. From the evaluation indicators obtained by averaging 100 groups of experiments at two resolutions in Table 1, it is not difficult to see that the detection results of this method have better performance in FN, FP, OE, PCC, Kappa and F1 indicators.
[0104] Table 1 Comparison of objective indicators
[0105]
[0106] A device for detecting changes in small targets in a video image at night with a large field of view, the device comprising: a processor and a memory, the memory storing program instructions, the processor calling the program instructions stored in the memory to enable the device to perform the following method steps:
[0107] Using comprehensive Anscomb e Root transformation and total variation regularization are used to remove the residual Poisson noise in the difference map, and the denoised image is obtained by inverse transformation;
[0108] The residual Gaussian noise in the denoised image is removed by using total variation regularization based on partial differential equations and minimizing the energy functional; the primal-dual algorithm is used to solve the energy functional;
[0109] The difference map is processed by a multi-scale superpixel segmentation and reconstruction method based on statistical information to enhance the edge features of the changed area and suppress the interference of noise;
[0110] The two difference images are fused based on multi-scale singular value decomposition fusion, and the FCM clustering algorithm is used to segment the fused difference image to obtain the detection result.
[0111] Among them, Anscomb e The forward transform and inverse transform of the root transform are:
[0112]
[0113]
[0114] Among them, DI k is the original image, X k is the image after the forward transformation of Anscombe transformation, X k - 1 is the image after the inverse Anscombe transform.
[0115] Among them, the primal-dual algorithm is used to solve the above energy functional:
[0116] Alternately update the original variable and the dual variable p, and update them according to the gradient descent principle of the regularized model. The original variable and the update formula are:
[0117]
[0118] Among them, I k+1 is the updated image, is the gradient of the image, λ is the Lagrange multiplier, I 0 is the original noise image, and the update formula of the Lagrange multiplier λ is:
[0119]
[0120]
[0121] Among them, I gradientis the size of the average gradient of the image, reflecting the overall gradient level of the image, p i,x and p i,y are the horizontal and vertical components of the gradient vector p, corresponding to the gradients in the x and y directions of the image, respectively;
[0122] Then the dual variable p is updated, and the update formula of the dual variable p is:
[0123]
[0124] Where σ is the step size parameter, is the updated image gradient.
[0125] Among them, the multi-scale superpixel segmentation and reconstruction method based on statistical information processes the difference map as follows:
[0126] Given a set of pixel values in a superpixel block {x 1 .x 2 , ..., x n}, n is the number of pixel values, and the reconstruction formula is:
[0127] I R =(w std ×O)+(w mean ×M)+(w median ×N)
[0128] Among them, W std Represents the weight of the original pixel, w mean represents the weight of the mean, w median Represents the weight of the median, O is the original pixel value inside the superpixel block, M is the mean value inside the superpixel block, and N is the median value inside the superpixel block.
[0129] Among them, the two difference images are fused based on multi-scale singular value decomposition fusion as follows:
[0130] The selection rules for U and V for each layer are as follows:
[0131]
[0132]
[0133] For diagonal matrices, select larger singular values for fusion. The selection rule is:
[0134] ∑ fused =ma(∑ 1 ,∑ 2 )
[0135] For Contrast1 and Contrast2, the global standard deviation of different layers of the Laplacian pyramid is calculated to obtain μ 1 k and μ 2 k is the average value of the current layer image, m and n are the number of rows and columns of the image respectively;
[0136]
[0137]
[0138] The final fusion image expression obtained by MSVD is:
[0139] Y k =U fused k ×∑ fused k ×(V fused k ) T
[0140]
[0141] Among them, K is the number of decomposition layers of the Laplacian pyramid, and Y is the final fused difference map.
[0142] It should be pointed out here that the device description in the above embodiment corresponds to the method description in the embodiment, and the embodiment of the present invention will not be described in detail here.
[0143] The execution subjects of the above-mentioned processor and memory can be devices with computing functions such as computers, single-chip microcomputers, and microcontrollers. In specific implementation, the embodiments of the present invention do not limit the execution subjects and are selected according to the needs of actual applications.
[0144] The data signal is transmitted between the memory and the processor via a bus, which is not described in detail in the embodiment of the present invention.
[0145] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, the storage medium includes a stored program, and when the program is running, the device where the storage medium is located is controlled to execute the method steps in the above embodiment.
[0146] The computer-readable storage medium includes but is not limited to a flash memory, a hard disk, a solid-state drive, and the like.
[0147] It should be pointed out here that the description of the readable storage medium in the above embodiment corresponds to the description of the method in the embodiment, and the embodiment of the present invention will not be described in detail here.
[0148] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. The 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 processes or functions according to the embodiments of the present invention are generated.
[0149] The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions may be stored in a computer-readable storage medium or transmitted via a computer-readable storage medium. The computer-readable storage medium may be any available medium that can be accessed by the computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium or a semiconductor medium, etc.
[0150] Unless otherwise specified, the models of the components in the embodiments of the present invention are not limited, and any device that can perform the above functions may be used.
[0151] Those skilled in the art will appreciate that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.
[0152] 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 principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for detecting small target changes in video images at night with a large field of view, characterized in that: The method comprises: The comprehensive Anscombe root transform and total variation regularization are used to remove the residual Poisson noise in the difference map, and the denoised image is obtained by inverse transformation; The residual Gaussian noise in the denoised image is removed by using total variation regularization based on partial differential equations and minimizing the energy functional; the primal-dual algorithm is used to solve the energy functional; The difference map is processed by a multi-scale superpixel segmentation and reconstruction method based on statistical information to enhance the edge features of the changed area and suppress the interference of noise; The two difference images are fused based on multi-scale singular value decomposition fusion, and the FCM clustering algorithm is used to segment the fused difference image to obtain the detection result.
2. The method for detecting small target changes in video images at night with a large field of view according to claim 1, characterized in that: The forward transform and inverse transform of the Anscombe root transform are: Among them, DI k is the original image, X k is the image after the forward transformation of Anscombe transformation, X k -1 is the image after the inverse Anscombe transform.
3. The method for detecting small target changes in video images at night with a large field of view according to claim 1, characterized in that: The primal-dual algorithm is used to solve the above energy functional: The original variable I and the dual variable p are updated alternately and updated according to the gradient descent principle of the regularized model. The formula for updating the original variable I is: Among them, I k+1 is the updated image, is the gradient of image I, λ is the Lagrange multiplier, I0 is the original noise image, and the update formula of Lagrange multiplier λ is: Among them, I gradinent is the size of the average gradient of the image, reflecting the overall gradient level of the image, p i,x and p i,y are the horizontal and vertical components of the gradient vector p, corresponding to the gradients in the x and y directions of the image, respectively; Then the dual variable p is updated, and the update formula of the dual variable p is: Where σ is the step size parameter, is the updated image gradient.
4. The method for detecting small target changes in video images at night with a large field of view according to claim 1, characterized in that: The multi-scale superpixel segmentation and reconstruction method based on statistical information processes the difference map as follows: Given a set of pixel values in a superpixel block {x1.x2,...,x n }, n is the number of pixel values, and the reconstruction formula is: I R =(w std ×O)+(w mean ×M)+(w median ×N) Among them, w std Represents the weight of the original pixel, w mean represents the weight of the mean, w median Represents the weight of the median, O is the original pixel value inside the superpixel block, M is the mean value inside the superpixel block, and N is the median value inside the superpixel block.
5. The method for detecting small target changes in video images at night with a large field of view according to claim 1, characterized in that: The fusion of the two difference images based on multi-scale singular value decomposition fusion is as follows: The selection rules for U and V for each layer are as follows: For diagonal matrices, select larger singular values for fusion. The selection rule is: ∑ fused =max(∑1,∑2) For Contrast1 and Contrast2, the global standard deviation of different layers of the Laplacian pyramid is calculated to obtain μ1 k and μ2 k is the average value of the current layer image, m and n are the number of rows and columns of the image respectively; The final fusion image expression obtained by MSVD is: Y k =U fused k ×∑ fused k ×(V fused k ) T Among them, K is the number of decomposition layers of the Laplacian pyramid, and Y is the final fused difference map.
6. A device for detecting small target changes in video images at night with a large field of view, characterized in that: The device comprises: a processor and a memory, wherein program instructions are stored in the memory, and the processor calls the program instructions stored in the memory to enable the device to execute any one of the methods of claims 1-5.
7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 5.
Citation Information
Patent Citations
Low-illumination wide-view-field video image change detection method and device
CN118279829A
Video image change detection method and device under conditions of low illumination and wide field of view
CN118333966A
Cited By
Method and system for removing instrument shielding from spinal endoscope image
CN119477747A
