A video image change detection method and device under low-illumination wide field of view conditions

Abstract

The application discloses a video image change detection method and device under low-illumination wide-field conditions, and the method comprises the following steps: introducing a Gaussian kernel quadratic smoothing algorithm, pre-processing an input image video frame, and obtaining a more anti-noise image through quadratic convolution smoothing; adopting a joint difference map change detection method, designing a hyperbolic cotangent logarithmic ratio operator conforming to low-illumination image characteristics and a Chebyshev accelerated multi-scale dense optical flow algorithm; adopting a self-similarity fusion method based on fractal theory to integrate difference maps containing different characteristics, taking a self-similarity coefficient as a fusion parameter, making the difference map retain the change part and suppress noise; and adopting an adaptive threshold denoising to obtain a final change detection result. The device comprises a processor and a memory. The application can resist random noise under a low-illumination environment, detect dark changes and small changes submerged in darkness which are difficult to detect, and quickly and accurately improve the detection rate.

Description

Technical Field

[0001] This invention relates to the field of video image change detection, and more particularly to a method and apparatus for video image change detection under low-light wide field of view conditions. Background Technology

[0002] In recent years, high-definition surveillance equipment has been continuously upgraded and iterated. Surveillance equipment deployed at high altitudes has strengthened urban grid management, effectively preventing unstable events and reducing and controlling the risk of accidents. However, when the monitored area is in a low-light environment, the area of ​​change in the wide-field-of-view surveillance image is small and the intensity is weak. This makes it difficult for security personnel to observe changes in the darkness in a timely manner, undoubtedly creating significant security risks. Therefore, research on change detection in low-light wide-field-of-view environments is particularly important. It is widely used in smart city construction, traffic management, public environment anomaly detection, and detection of illegal buildings.

[0003] In recent years, many change detection methods have been proposed, including traditional methods such as pixel statistics and transformation of the change domain to determine whether a change has occurred, such as mean ratio, logarithmic ratio, and principal component analysis. Meanwhile, deep learning-based change detection methods are also constantly evolving. These methods classify pixels into changed or unchanged pixels through network model training. Although deep learning methods can demonstrate excellent detection performance on labeled training sets, they require significant costs for annotation and hardware and lack real-time performance. Especially in low-light, wide-field-of-view scenes lacking annotations, factors such as local atmospheric conditions, diverse weather, and lighting can lead to high noise in the acquired image data, making it difficult for existing change detection algorithms to identify weak changes or detect more subtle changes in multi-temporal video images.

[0004] In the process of realizing this invention, the inventors discovered the following drawbacks and shortcomings of the above method:

[0005] First: Due to the small scale and low brightness of changes in low-light wide-field scenes, existing change detection methods can only acquire few features in this scene, making it difficult to detect dark changes.

[0006] Second: Due to interference from complex light sources and atmospheric scattering particles, there is often significant random noise in surveillance video images under low illumination. Existing change detection algorithms are difficult to effectively solve the problem of weak change detection under significant noise.

[0007] Third: The wide field of view makes the existing methods have high runtime complexity and lack real-time performance. Summary of the Invention

[0008] This invention provides a method and apparatus for detecting changes in video images under low-light wide field-of-view conditions. This invention can combat random noise in low-light environments and detect dark changes and small changes that are difficult to detect in darkness, thus rapidly and accurately improving the detection rate. See the description below for details:

[0009] A first aspect: a method for detecting changes in video images under low-light, wide-field-of-view conditions, the method comprising:

[0010] A Gaussian kernel quadratic smoothing algorithm is introduced to preprocess the input image and video frames, and a more noise-resistant image is obtained through quadratic convolution smoothing;

[0011] A joint difference map change detection method was adopted, and an inverse hyperbolic cotangent log ratio operator and a Chebyshev-accelerated multi-scale dense optical flow algorithm that conform to the characteristics of low-light images were designed.

[0012] A self-similarity fusion method based on fractal theory is adopted to integrate difference maps containing different features. The self-similarity coefficient is used as a fusion parameter to preserve the variable parts of the difference map and suppress noise.

[0013] The final change detection result is obtained by using adaptive threshold denoising.

[0014] The inverse hyperbolic cotangent logarithmic ratio operator is:

[0015]

[0016] Where acoth is the inverse hyperbolic cotangent function. , , and These are two video frames, ( Here, G represents the spatial coordinates of each frame of the image, and T represents the transpose of the Gaussian kernel. It is 1e-6.

[0017] The Chebyshev-accelerated multi-scale dense optical flow algorithm is as follows:

[0018]

[0019] in, Let be the horizontal component of the optical flow field at image position (i,j) in the nth outer iteration and the (r+1)th inner iteration. Let (i,j) be the vertical component of the optical flow field at image position (i,j) in the nth outer iteration and the (r+1)th inner iteration. It is the acceleration correction factor generated by the Chebyshev polynomial, denoted as... , and ... Let k denote the Chebyshev polynomial, where k is the order of the Chebyshev polynomial. The first frame of the video image, This is the second frame of the video image. This represents the horizontal displacement of the second frame of the video image. This represents the vertical displacement of the second frame of the video image. For displacement vectors, and These are the means of u and v, respectively. is the regularization parameter. The self-similarity fusion method based on fractal theory is as follows:

[0020]

[0021] Where a is the scaling parameter of the sigmoid function, and b is the adjustment factor introduced. It is the bias parameter of the sigmoid function. and These are two images to be merged, where c represents the image inversion operation. and These are the weight coefficients of the two images.

[0022] A second aspect is a video image change detection device under low-light wide field of view conditions, 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.

[0023] 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.

[0024] The beneficial effects of the technical solution provided by this invention are:

[0025] 1. This invention uses a joint difference map approach to integrate difference maps generated by different methods, which can not only retain the changing pixels of different features, but also improve the detection accuracy and reduce the time complexity.

[0026] 2. To detect weak changes in a wide field of view under low illumination, this invention proposes for the first time an inverse hyperbolic cotangent function logarithmic ratio operator, combining enhanced pixel contrast with a change detection operator. Compared to existing algorithms, this invention truly achieves the ability to detect weak changes in a wide field of view under low illumination. At the same time, leveraging the advantages of multi-scale optical flow, this invention proposes an optical flow difference map detection method suitable for a wide field of view. Compared to existing algorithms, this invention can avoid interference from complex light sources and random noise.

[0027] 3. This invention is the first to propose a self-similar fusion algorithm that fuses only the changed parts based on the change detection characteristics, which meets the difference map fusion requirements in different scenarios; experimental analysis has proven that this invention can detect weak changes in different low-light wide field-of-view scenarios, meeting various needs in practical applications. Attached Figure Description

[0028] Figure 1 A flowchart of a video image change detection method under low-light wide field-of-view conditions;

[0029] Figure 2 A flowchart for the improved multi-scale dense optical flow method;

[0030] Figure 3 Schematic diagram of a low-light wide field-of-view video image;

[0031] Figure 4 Schematic diagram of low-light wide field-of-view video image 2;

[0032] Figure 5 This is a schematic diagram showing the results of change detection on video image 1 and video image 2;

[0033] Figure 6 Schematic diagram of a wide field-of-view video image with high illumination;

[0034] Figure 7 Schematic diagram of low-light wide field-of-view video image 4;

[0035] Figure 8 This is a schematic diagram showing the results of change detection on video images 3 and 4;

[0036] Figure 9 A schematic diagram for validating the effectiveness of the joint difference map method;

[0037] Among them, (a) is the log-ratio + mean ratio framework; (b) is the log-ratio + pixel ratio framework; (c) is the log-ratio + extreme value ratio framework; and (d) is the inverse hyperbolic cotangent log-ratio operator + improved multi-scale dense optical flow framework.

[0038] Figure 10 This is a schematic diagram illustrating the effectiveness verification of the self-similar fusion method based on fractal theory.

[0039] Figure 11 For validity verification.

[0040] Among them, (a) PCA-KMEANS; (b) CWNN; (c) INLPG; (d) FDA-RMG; (e) NR-ELM; (f) MSFNF; (g) RSACD; (h) IST-CRF; (i) embodiments of the present invention. Detailed Implementation

[0041] 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.

[0042] Example 1

[0043] This invention provides a method for detecting changes in video images under low-light, wide-field-of-view conditions. (See also...) Figure 1 and Figure 2 The method includes the following steps:

[0044] Step 101: Introduce the Gaussian kernel quadratic smoothing algorithm to preprocess the input image and video frames, and obtain a more noise-resistant image through quadratic convolution smoothing;

[0045] The embodiments of the present invention achieve image smoothing through the above design, reducing noise interference while maintaining image details, and to a certain extent eliminating the influence of noise on the grayscale changes of image edges.

[0046] Step 102: Using the joint difference map change detection method, an inverse hyperbolic cotangent logarithmic ratio operator and a noise-resistant multi-scale dense optical flow algorithm were designed that conform to the characteristics of low-light images.

[0047] This method employs an inverse hyperbolic cotangent logarithmic ratio operator to enhance the contrast of difference maps in low-light images, highlighting changing targets in low-light scenes. A Chebyshev-accelerated multi-scale dense optical flow algorithm is introduced to address detection environments with wide fields of view. This joint approach allows the generated difference maps to complement each other's strengths. The inverse cosine logarithmic ratio operator compensates for the inability of dense optical flow to detect changing targets in dark areas under low-light conditions, while dense optical flow reduces noise interference introduced by the inverse cosine logarithmic ratio operator.

[0048] Step 103: The self-similarity fusion method based on fractal theory is used to integrate the difference maps containing different features. The integration process uses the self-similarity coefficient as the fusion parameter to preserve the variable parts of the difference map and suppress noise.

[0049] The fusion algorithm is based on the sigmoid function and introduces image inversion technology. It uses the self-similarity characteristics of adjacent frames of video images as fusion weight coefficients, effectively fusing the changed parts and setting the unchanged parts to zero.

[0050] Step 104: Use adaptive threshold denoising to obtain the final change detection result;

[0051] In this embodiment of the invention, a threshold denoising method based on area is set in combination with the magnitude of target changes in real-world scenarios to remove unnecessary interference and obtain the final change detection result.

[0052] Step 105: To verify the feasibility of the experiment, low-light wide-field-of-view video images from real-world scenarios were used for verification.

[0053] Among them, the embodiments of the present invention can accurately and efficiently detect weak changes in a wide field of view under low illumination.

[0054] In summary, the embodiments of the present invention, through steps 101-105, achieve both the ability to combat random noise in low-light environments and the ability to detect dark changes and small changes that are difficult to detect in darkness, thereby rapidly and accurately improving the detection rate.

[0055] Example 2

[0056] The following is combined with Figure 2 The solution in Example 1 will be further described below:

[0057] Step 201: Introduce the Gaussian kernel double smoothing algorithm to preprocess the input image and video frames, and obtain a more noise-resistant image through double convolution smoothing;

[0058] To reduce noise interference while preserving image details, this embodiment of the invention employs a Gaussian smoothing algorithm, which follows a Gaussian distribution. Window size .

[0059] The embodiments of the present invention, through the above design, enable image smoothing and reduce noise interference while preserving image details, and to a certain extent eliminate the influence of noise on the grayscale changes of image edges. Here, [the following will be discussed]. Set to 1 to preserve image details. The secondary smoothing operation involves converting two video frames... and Convolving the image with a Gaussian filter yields an image smoothed in both the x and y directions. Then, the smoothed image is combined with the transpose of the Gaussian filter. Convolution, or double smoothing of the function, yields a smoother image, expressed by the formula: , Low-light video images are affected by the complex lighting conditions at night and contain a lot of random noise. Gaussian smoothing can avoid false detections caused by noise.

[0060] Step 202: In the inverse hyperbolic cotangent logarithmic ratio operator, the inverse hyperbolic cotangent function is added to the original logarithmic ratio operator to process the smoothed image;

[0061] (1)

[0062] Here, acoth is the inverse hyperbolic cotangent function, defined here with a relatively small value. This prevents the occurrence of pixels with a value of 0. The inverse hyperbolic cotangent function is a non-linear function with good gradient properties. Compared to the logarithmic ratio operator, which can only detect changes, the inverse hyperbolic cotangent logarithmic ratio operator has both enhancement and change detection functions. This means that in an image after inverse hyperbolic cotangent transformation, low-brightness parts will be enhanced, while high-brightness parts will be weakened, thereby improving image contrast. When used in combination with the logarithmic ratio operator, this difference is further amplified. Therefore, compared to general difference map generation algorithms, the inverse hyperbolic cotangent logarithmic ratio operator proposed in this embodiment can also detect changing targets submerged in darkness to a certain extent, improving detection accuracy while having lower time complexity.

[0063] Step 203: In the Chebyshev accelerated multi-scale dense optical flow algorithm, it is based on the HS optical flow method, and based on the wide field of view characteristics, it is extended into a large displacement optical flow field by taking advantage of the multi-scale pyramid.

[0064] The basic optical flow constraint equation, extended to adapt to large displacements, can be described as follows: ;in Represents the velocity at a certain moment Represents light intensity. Substituting the above equation into the energy functional of the HS optical flow, we obtain equation (2):

[0065] (2)

[0066] After minimizing the energy functional in the above equation, linearizing the Euler-Lagrange equation using a first-order Taylor expansion yields equation (3):

[0067] (3)

[0068] The smoothed two-phase images were then scaled down proportionally to different scales, for a set of scales. , The set scale number, the pyramid construction expression for the image can be represented by formula (4). It is the image at the s-th scale.

[0069] (4)

[0070] in, (0,1) represents the scaling factor, which can be set to 0.5. Starting from the coarsest scale, the equations are solved at each scale to obtain a successive approximation of the optical flow. Each intermediate solution is used as the initialization for the next scale, and numerical propagation is performed from the coarser scale. The flow field update is expressed as follows: .

[0071] In this embodiment of the invention, the Chebyshev-accelerated SOR method is used to solve the Euler-Lagrange equation at each scale. To accelerate the convergence speed of the SOR iteration, Chebyshev polynomial coefficients are introduced to generate a correction factor. First, use a function. Define the Chebyshev polynomial: Where k is the order of the polynomial and x is the independent variable, in this embodiment of the invention, the order k is set to [1, 4]. Then the correction factor for the k-th order can be expressed by the formula: Indicated. In the formula and and represent the scaling factors of the Chebyshev polynomial, used to map the polynomial's independent variable from [-1, 1] to the range of matrix eigenvalues, respectively. and It means that, among them, and This sets the minimum and maximum eigenvalues ​​of the Chebyshev polynomial.

[0072] Since the SOR method requires an additional fixed-point iteration level, namely the inner iteration r, to achieve its convergence, this embodiment of the invention embeds r into the outer iteration number n of the Taylor expansion, while setting a maximum number of inner iterations. At iteration (n, r+1), this embodiment of the invention groups and represents the Euler-Lagrange equations with respect to the flow field, and substitutes the correction factor formula, thereby obtaining the following fixed-point iterative form:

[0073] (5)

[0074] The correction factor in the above formula It is generated using Chebyshev polynomials, which can optimize the direction and step size of updates, allowing the SOR method to converge to the optimal solution more quickly. and These are the means of u and v, respectively. , , , .

[0075] At each scale of the pyramid, the SOR method with accelerated convergence, exemplified by Chebyshev, is used iteratively until convergence to a steady-state solution or the external maximum number of iterations is reached. Once the SOR converges, this embodiment of the invention proceeds to the next outer iteration n+1. The stopping criterion from the HS method is still used here:

[0076] (6)

[0077] Therefore, the optical flow field can be obtained through this algorithm. In order to map it as a change, this embodiment of the invention uses an interpolation mapping operation to represent the optical flow field as a changing field, such as the movement, deformation, and brightness change of a changing target. The flowchart is as follows. Figure 2 As shown.

[0078] Step 204: Propose a self-similar weight fusion based on fractal theory, using the sigmoid function as a foundation, and introduce image inversion technology to... The image after color inversion is The self-similarity of an image is used as a weight to enhance the image's features.

[0079] The self-similarity weight is described by the fractal dimension (FD) in fractal theory. This embodiment of the invention uses a morphological covering algorithm to quickly solve for the fractal dimension of an image. For two-dimensional discrete image signals... in The corresponding expansion process is as follows: In the formula, This represents a structuring element with a coverage area of ​​d. Set to the maximum coverage area, and also set... One pixel The value is used express. ,use The morphological coverage is Pixel value greater than area The quantity. The fractal dimension can be expressed by the formula:

[0080] (7)

[0081] Use the least squares method and By fitting the equation into a linear equation with two variables, the slope of which is an approximation of FD. The above formula is then used to calculate... fractal dimension As The self-similarity weights are used because the complexity of difference maps in the same scene is quite similar over a short period of time. To reduce computational complexity, The self-similarity weight is set to N times. N is determined based on the fusion threshold. Experimental testing showed that N=20 allows the fusion algorithm to achieve maximum efficiency. The self-similar weighted fusion method based on fractal theory can be expressed as:

[0082] (8)

[0083] Where a is the scaling parameter of the sigmoid function, and b is the adjustment factor introduced. These are the bias parameters of the sigmoid function. In this embodiment of the invention, a=2 and b=1. =0, and These are two images to be merged. and These are the weight coefficients of the two images.

[0084] Therefore, by using self-similar weight fusion, it is possible to retain the changing parts and suppress noise based on different features.

[0085] Step 205: Use an adaptive thresholding method for denoising.

[0086] Threshold denoising comprises two main parts. First, the mean of the fused results is used as the threshold; values ​​less than the mean are set to zero, while values ​​greater than the mean are retained. Second, based on the different characteristics of noise points and changing targets, a denoising algorithm is designed using the area of ​​connected components to effectively remove isolated noise interference.

[0087] Experimental Verification 1: Effectiveness Analysis of the Joint Difference Map Method

[0088] To verify the effectiveness of the inverse hyperbolic cotangent logarithmic ratio operator and the improved multi-scale dense optical flow jointly generating difference maps in the embodiments of the present invention, three different joint operator frameworks—logarithmic ratio and mean ratio operator, logarithmic ratio and pixel ratio operator, and logarithmic ratio and extremum ratio operator—were used as comparative algorithms, and four experimental control groups were selected for experimental analysis. The experimental results are as follows: Figure 9 As shown, the other three joint frameworks suffer from missed detections, false detections, and significant noise interference during the detection process. However, the joint difference map algorithm proposed in this embodiment of the invention has strong anti-interference capabilities and can preserve the edge details of changing targets more completely.

[0089] Experimental Verification 2: Effectiveness Analysis of Self-Similarity Fusion Based on Fractal Theory

[0090] To verify the effectiveness of the self-similar fusion method based on fractal theory in this embodiment of the invention, the equal-weight fusion method, PCA fusion algorithm, LRD fusion algorithm, and EMD fusion algorithm are used as comparative algorithms, and one experimental control group is selected for experimental analysis. The experimental results are as follows: Figure 10As shown, the equal-weight fusion, PCA, LRD, and EMD fusion algorithms retain a considerable amount of residual noise during the fusion process, causing the changing target to be masked by noise. In contrast, the proposed self-similar fusion algorithm based on fractal theory effectively preserves the changing parts and suppresses noise and background interference to the greatest extent.

[0091] Experimental Verification 3: Effectiveness Analysis of the Embodiments of the Invention

[0092] To verify the accuracy of the algorithm in this embodiment of the invention, this method was compared experimentally with eight other change detection algorithms: PCA-KMEANS, CWNN, INLPG, FDA-RMG, NR-ELM, MSFNF, RSACD, and ISTCRF, and the experimental results were analyzed. The experimental results are as follows: Figure 11 As shown. Six scenes were selected for visualization experiments. From these experimental control groups, it can be observed that other algorithms have poor detection performance in low-light wide field-of-view change detection scenes. However, the algorithm proposed in this embodiment of the invention, which combines multiple difference maps and self-similarity fusion, can easily cope with this type of challenge. This is because the difference map generated by the combination of the inverse hyperbolic cotangent logarithm ratio algorithm and the improved multi-scale dense optical flow can highlight the changing target under low-light conditions, while the self-similarity weighted fusion based on inverse color technology can retain the changing target while suppressing noise, effectively enhancing the robustness of the overall algorithm. (a) The PCA-Kmeans algorithm performs well in simple scenes, but detects a lot of noise under complex lighting conditions. (b) The CWNN algorithm exhibits a lot of false changes because it cannot effectively eliminate random noise interference. (c) The INLPG algorithm almost fails in low-light environments because its effectiveness is limited under highly structurally similar conditions. (d) The FDA-RMG algorithm significantly improves the detection accuracy through a two-stage detection method, but may miss small changes. (e) The NR-ELM algorithm contains a lot of false changes because the binary classification of changes is not accurate enough. (f) The MSFNF algorithm performs well in simple challenges, but due to its powerful denoising algorithm, it tends to remove subtle changes as noise. (g) The multi-scale superpixel reconstruction method of the RSACD algorithm causes the disappearance of changed targets, resulting in poor detection performance. (h) The ISTCRF algorithm's structural feature extraction for change detection is inaccurate, leading to the loss of edge structures of changed targets and reducing detection accuracy.

[0093] A video image change detection device under low-light wide field of view conditions, the device includes: a processor and a memory, the memory storing program instructions, and the processor calling the program instructions stored in the memory to cause the device to execute the following method steps in Embodiment 1:

[0094] A Gaussian kernel quadratic smoothing algorithm is introduced to preprocess the input image and video frames, and a more noise-resistant image is obtained through quadratic convolution smoothing;

[0095] A joint difference map change detection method was adopted, and an inverse hyperbolic cotangent log ratio operator and a Chebyshev-accelerated multi-scale dense optical flow algorithm that conform to the characteristics of low-light images were designed.

[0096] A self-similarity fusion method based on fractal theory is adopted to integrate difference maps containing different features. The self-similarity coefficient is used as a fusion parameter to preserve the variable parts of the difference map and suppress noise.

[0097] The final change detection result is obtained by using adaptive threshold denoising.

[0098] The inverse hyperbolic cotangent logarithmic ratio operator is:

[0099]

[0100] Where acoth is the inverse hyperbolic cotangent function. , , and These are two video frames, ( The coordinates of each frame are given by G, G is the Gaussian kernel, and T is the transpose. It is 1e-6.

[0101] The Chebyshev-accelerated multi-scale dense optical flow algorithm is as follows:

[0102]

[0103] in, Let be the horizontal component of the optical flow field at image position (i,j) in the nth outer iteration and the (r+1)th inner iteration. Let (i,j) be the vertical component of the optical flow field at image position (i,j) in the nth outer iteration and the (r+1)th inner iteration. It is the acceleration correction factor generated by the Chebyshev polynomial, denoted as... , and ... Let k denote the Chebyshev polynomial, where k is the order of the Chebyshev polynomial. The first frame of the video image, This is the second frame of the video image. This represents the horizontal displacement of the second frame of the video image. This represents the vertical displacement of the second frame of the video image. For displacement vectors, and These are the means of u and v, respectively. Here is the regularization parameter. The self-similarity fusion method based on fractal theory is as follows:

[0104]

[0105] Where a is the scaling parameter of the sigmoid function, and b is the adjustment factor introduced. It is the bias parameter of the sigmoid function. and These are two images to be merged, where c represents the image inversion operation. and These are the weight coefficients of the two images.

[0106] 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.

[0107] 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.

[0108] 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.

[0109] 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.

[0110] The computer-readable storage medium includes, but is not limited to, flash memory, hard disk, solid-state drive, etc.

[0111] 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.

[0112] 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.

[0113] 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.

[0114] 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.

[0115] 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 video images under low-light, wide-field-of-view conditions, characterized in that, The method includes: A Gaussian kernel quadratic smoothing algorithm is introduced to preprocess the input image and video frames, and a more noise-resistant image is obtained through quadratic convolution smoothing; A joint difference map change detection method was adopted, and an inverse hyperbolic cotangent log ratio operator and a Chebyshev-accelerated multi-scale dense optical flow algorithm that conform to the characteristics of low-light images were designed. A self-similarity fusion method based on fractal theory is adopted to integrate difference maps containing different features. The self-similarity coefficient is used as a fusion parameter to preserve the variable parts of the difference map and suppress noise. The final change detection result is obtained by using adaptive threshold denoising. The inverse hyperbolic cotangent logarithmic ratio operator is: ； Where acoth is the inverse hyperbolic cotangent function. , , and These are two video frames, Here, G represents the spatial coordinates of each frame of the image, and T represents the transpose of the Gaussian kernel. It is 1e-6; The Chebyshev-accelerated multi-scale dense optical flow algorithm is as follows: ； in, Let be the horizontal component of the optical flow field at image position (i,j) in the nth outer iteration and the (r+1)th inner iteration. Let (i,j) be the vertical component of the optical flow field at image position (i,j) in the nth outer iteration and the (r+1)th inner iteration. It is the acceleration correction factor generated by the Chebyshev polynomial, denoted as... , and ... Let k denote the Chebyshev polynomial, where k is the order of the Chebyshev polynomial. The first frame of the video image, This is the second frame of the video image. This represents the horizontal displacement of the second frame of the video image. This represents the vertical displacement of the second frame of the video image. For displacement vectors, and These are the means of u and v, respectively. For regularization parameters; The self-similarity fusion method based on fractal theory is as follows: ； Where a is the scaling parameter of the sigmoid function, and b is the adjustment factor introduced. It is the bias parameter of the sigmoid function. and These are two images to be merged, where c represents the image inversion operation. and These are the weight coefficients of the two images.

2. A video image change detection device under low-light wide field-of-view conditions, characterized in that, The device includes a processor and a memory, the memory storing program instructions, and the processor calling the program instructions stored in the memory to cause the device to perform the method of claim 1.

3. 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 of claim 1.

Images