A method for detecting moving targets against a sea-sky background

Abstract

The present invention relates to a method for detecting moving targets in a sea-sky background, including: collecting consecutive frames of field-of-view images in a sea-sky background; performing visibility grading on each obtained field-of-view image in a sea-sky background, and selecting consecutive frames of field-of-view images whose visibility grading meets the requirements for moving target detection for subsequent moving target detection; for consecutive frames of field-of-view images whose visibility grading meets the requirements for moving target detection, extracting consecutive frame information; after processing the front and rear frame images including differential operation and centroid search, determining candidate target regions in each frame of the field-of-view image, and detecting moving targets according to the relationship between the change in gray-scale features in the candidate target regions and the overall average gray scale of the field-of-view image. The present invention performs visibility grading on each obtained field-of-view image in a sea-sky background, and selects consecutive frames of field-of-view images whose visibility grading meets the requirements for moving target detection for subsequent moving target detection, thereby improving the detection probability.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing and moving target detection, and particularly relates to a moving target detection method under a sea-sky background. Background Art

[0002] In recent years, the research on ship moving target detection has always been a research hotspot in the field of moving target detection. As the most important transportation mode in international logistics, with the rapid development of China's maritime trade, the number of various ships entering and leaving ports is increasing day by day. The moving target detection of ships is of increasing importance for improving maritime traffic order and maintaining marine security. In particular, the detection of moving ship targets at a long distance that is difficult to distinguish by the naked eye and needs to be recorded by an optical detector becomes necessary. However, when the sea clutter or fish-scale light is strong, the gray-scale intensity of a large number of wave crests in the field-of-view image is close to or even equal to or greater than the maximum gray-scale value of the target pixel points, which greatly reduces the performance of the detection algorithm. Moreover, the visibility of the sea-sky background has a great influence on the target detection, and it is necessary to comprehensively consider the detection of moving targets according to the visibility conditions. Summary of the Invention

[0003] In view of the above analysis, the present invention aims to disclose a moving target detection method under a sea-sky background to solve the problem of detecting moving targets that are difficult to distinguish by the naked eye and need to be recorded by an optical detector.

[0004] The present invention discloses a moving target detection method under a sea-sky background, including:

[0005] Collecting consecutive frames of field-of-view images under a sea-sky background;

[0006] Performing visibility grading on the obtained field-of-view images of each frame under a sea-sky background, and selecting the field-of-view images of consecutive frames whose visibility grading meets the requirements of moving target detection for subsequent moving target detection;

[0007] For the field-of-view images of consecutive frames whose visibility grading meets the requirements of moving target detection, extracting consecutive frame information; after processing the front and rear frame images including differential operation and centroid search, determining candidate target regions in the field-of-view image of each frame, and detecting moving targets according to the relationship between the change of gray-scale features in the candidate target regions and the overall average gray-scale of the field-of-view image.

[0008] Further, the moving target detection for the field-of-view images of consecutive frames whose visibility meets the requirements of moving target detection includes:

[0009] Extracting consecutive frame information, and performing differential operation on the front and rear frame images to extract a binary image;

[0010] Filtering the binary image to eliminate the influence of the sea surface and weather;

[0011] Determine the candidate target area in the binary image through centroid search;

[0012] Obtain the change value of the gray-scale feature of the image in the candidate target area of the field-of-view image;

[0013] When the change value of the gray-scale feature of the image is greater than the overall average gray-scale of the field-of-view image, it is determined that the candidate target area includes a moving target.

[0014] Furthermore, the moving target detection for the field-of-view images of consecutive frames with visibility satisfying the moving target detection includes:

[0015] Extract the consecutive frame information, and establish a target search area near the sea horizon on the sea horizons in the front and rear frame images for image interception;

[0016] Perform a differential operation on the target search areas intercepted from the front and rear frame images to extract a binary image;

[0017] Filter the binary image to eliminate the influence of the sea surface and weather;

[0018] Determine the candidate target area in the binary image through centroid search;

[0019] Obtain the change value of the gray-scale feature of the image in the candidate target area of the field-of-view image;

[0020] When the change value of the gray-scale feature of the image is greater than the overall average gray-scale of the field-of-view image, it is determined that the candidate target area includes a moving target.

[0021] Furthermore, perform visibility grading on the field-of-view images of each frame under the sea-sky background, and select the field-of-view images of consecutive frames with visibility grading satisfying the moving target detection for subsequent moving target detection; including:

[0022] Perform preprocessing on the input field-of-view image, including downsampling, target gradient feature extraction, and gradient normalization;

[0023] Extract the edge points in the image according to the result of the gradient normalization;

[0024] Judge whether the sea horizon in the image can be extracted according to the number and length information of the edge points;

[0025] For the field-of-view images of consecutive frames that can extract the sea horizon, grade the visibility of the image according to the magnitude of the maximum gradient;

[0026] Select the field-of-view images of consecutive frames with visibility grading satisfying the moving target detection for subsequent moving target detection.

[0027] Furthermore, the preprocessing includes:

[0028] Downsample the input field of view image to obtain a downsampled image; and classify the visibility of the image with a gray mean value less than the gray threshold in the downsampled image as a poor level;

[0029] Perform adaptive Gaussian filtering on the downsampled image with a gray mean value not less than the gray threshold to calculate the target gradient feature set of the image including the horizontal, vertical gradients and the maximum gradient of the image;

[0030] Classify the visibility of the image with a maximum gradient value less than the first gradient threshold as a poor level;

[0031] For the image with a maximum gradient value not less than the first gradient threshold, obtain the gradient normalization result according to the gradient normalization operation.

[0032] Furthermore, calculate the horizontal and vertical gradients of the image through adaptive Gaussian filtering;

[0033] The size of the Gaussian filter is an n*n matrix, and the Gaussian filter is obtained by combining the sigma parameter of the canny operator: dgau2D=-x.*exp(-(x.*x+x T ·*x T ) / (2*canny_sigma)) / (2*π*canny_sigma 2 );

[0034] The horizontal gradient set and the vertical gradient sets dx and dy are:

[0035] The median value of the image gradient is: value = sqrt(dx.*dx + dy.*dy);

[0036] The maximum gradient is: max_grad = max(value);

[0037] I is the pixel matrix of the downsampled image.

[0038] Furthermore, perform gradient normalization operations according to the maximum gradient binning, specifically including:

[0039] Perform gradient normalization by binning the maximum gradient into 5 normalization factors. The normalization factors are shown in the following formula: In the formula, the binning constants va1 < va2 < va3 < va4 < va5;

[0040] When the maximum gradient max_grad is greater than th1, the gradient is normalized to value1; otherwise, when the maximum gradient is greater than th2, the gradient is normalized to value2; otherwise, when the maximum gradient is greater than th3, the gradient is normalized to value3; otherwise, when the maximum gradient is greater than th4, the gradient is normalized to value4; when none of the above conditions are met, the gradient is directly normalized to value5.

[0041] Further, the edge points of the image are extracted using the canny algorithm; when extracting the edge points of the image, non-maximum suppression in the canny algorithm is performed on the horizontal and vertical gradient sets calculated by Gaussian filtering;

[0042] When extracting the edge points of the image, the gradient normalization result is used for high and low threshold detection of the canny algorithm; the high threshold is detected to reduce false edges in the image; the low threshold is used to collect edge points whose contours meet the requirements to form new edges, thereby realizing the closure of the image edge.

[0043] Further, the method for judging whether the sea horizon line in the image can be extracted according to the number and length information of the edge points includes:

[0044] The number and length information of each edge are obtained through eight-neighborhood labeling. When the number of edges is greater than the threshold, the hough transform is used to extract the sea horizon line; the images whose number and range of the sea horizon line do not meet the conditions are determined as images that cannot extract the sea horizon line, and the visibility classification is a poor level, and they are classified as difficult samples without performing moving target detection.

[0045] Further, the images whose number and range of the sea horizon line meet the conditions are determined as images that can extract the sea horizon line; for the images that can extract the sea horizon line, threshold comparison judgment is performed using the maximum gradient of the image. When the maximum gradient of the image is less than the second gradient threshold, the visibility classification of the image is a poor level; when the maximum gradient of the image is between the second gradient threshold and the third gradient threshold, the visibility classification of the image is a general level; when the maximum gradient of the image is greater than the third gradient threshold, the visibility classification of the image is a good level;

[0046] Among them, the images with a poor visibility classification are classified as difficult samples without performing moving target detection; the consecutive frame field-of-view images with a general and good visibility classification are used as images for moving target detection.

[0047] The present invention can at least achieve one of the following beneficial effects:

[0048] The present invention performs visibility classification on the field-of-view images in the sea-sky background for each frame, and selects the consecutive frame field-of-view images whose visibility classification meets the requirements for moving target detection for subsequent moving target detection, thereby improving the detection probability.

[0049] Moreover, in the present invention, by improving the Canny algorithm, a gradient adaptive planning method based on the maximum gradient calculation of the acquired image and weighted texture is established to complete the visibility judgment of the sea-sky background, realizing a more stable and effective visibility judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings are only for the purpose of showing specific embodiments and are not considered as limitations of the present invention. Throughout the drawings, the same reference signs denote the same components.

[0051] Figure 1 It is a flowchart of the moving target detection method under the sea-sky background in the embodiment of the present invention;

[0052] Figure 2 It is a flowchart of the field-of-view image selection method for satisfying moving target detection in the embodiment of the present invention;

[0053] Figure 3 It is a flowchart of the preprocessing method in the embodiment of the present invention;

[0054] Figure 4 It is a flowchart of the gradient normalization method in the embodiment of the present invention;

[0055] Figure 5 It is a flowchart of the image visibility grading method in the embodiment of the present invention;

[0056] Figure 6 It is a flowchart of the moving target detection method for the field-of-view image satisfying moving target detection in the embodiment of the present invention;

[0057] Figure 7 It is a flowchart of the moving target detection method for the field-of-view image satisfying moving target detection in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] The following will specifically describe the preferred embodiments of the present invention with reference to the drawings, wherein the drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention.

[0059] An embodiment of the present invention discloses a moving target detection method under a sea-sky background, as Figure 1 shown, including the following steps:

[0060] Step S101, acquiring consecutive frames of field-of-view images under the sea-sky background;

[0061] Step S102, performing visibility grading on each acquired frame of the field-of-view image under the sea-sky background, and selecting consecutive frames of field-of-view images whose visibility grading meets the requirements for moving target detection for subsequent moving target detection;

[0062] Step S103: For the field-of-view images of consecutive frames whose visibility grading meets the requirements for moving target detection, extract the information of consecutive frames; after processing the front and rear frame images including differential operation and centroid search, determine the candidate target areas in the field-of-view image of each frame, and detect the moving targets according to the relationship between the change of gray-scale features in the candidate target areas and the overall average gray-scale of the field-of-view image.

[0063] Specifically, in step S101, an optical detector for long-distance infrared imaging is used to observe the sea, and the field-of-view images under the sky-sea background of consecutive frames are collected. The field-of-view images include the sky, the sea surface, and the sea horizon that divides the sky and the sea; in this application, the moving targets to be detected are long-distance targets and appear near the sea horizon in the field-of-view image.

[0064] Specifically, as Figure 2 shown, step S102 includes:

[0065] Step S201: Perform preprocessing on the input field-of-view image including downsampling, target gradient feature extraction, and gradient normalization;

[0066] Step S202: Extract the edge points in the image according to the result of the gradient normalization;

[0067] Step S203: Judge whether the sea horizon in the image can be extracted according to the number and length information of the edge points;

[0068] Step S204: For the field-of-view images of consecutive frames for which the sea horizon can be extracted, grade the visibility of the images according to the magnitude of the maximum gradient;

[0069] Step S205: Select the field-of-view images of consecutive frames whose visibility grading meets the requirements for moving target detection for subsequent moving target detection.

[0070] Specifically, as Figure 3 shown, the preprocessing includes:

[0071] Step S301: Perform downsampling on the input field-of-view image to obtain a downsampled image; and grade the visibility of the image with a gray-scale mean less than the gray-scale threshold in the downsampled image as a poor level;

[0072] Perform downsampling on the input original image (image resolution 640*480) to obtain a downsampled image (image resolution 320*240), thereby reducing the computational amount. At the same time, when the gray-scale mean of the image is less than the gray-scale threshold Gray_TH, grade the visibility of the image as the "poor" level.

[0073] Step S302: Perform adaptive Gaussian filtering on the downsampled image with a grayscale mean not less than the grayscale threshold to calculate the target gradient feature set of the image, including the horizontal, vertical gradients, and maximum gradient of the image;

[0074] The size of the Gaussian filter is an n*n matrix. Combining with the sigma parameter of the canny operator, the Gaussian filter is: dgau2D = -x.*exp(-(x.*x + x T .*x T ) / (2*canny_sigma)) / (2*π*canny_sigma 2 );

[0075] where x is the input data of the Gaussian filter, a two-dimensional n*n vector.

[0076] The horizontal gradient set and vertical gradient set dx and dy are:

[0077] The intermediate value of the image gradient is: value = sqrt(dx.*dx + dy.*dy);

[0078] The maximum gradient is: max_grad = max(value);

[0079] I is the pixel matrix of the downsampled image.

[0080] Step S303: Classify the visibility of the image with a maximum gradient value less than the first gradient threshold grad1 as a poor level;

[0081] Step S304: For the image with a maximum gradient value not less than the first gradient threshold grad1, obtain the gradient normalization result according to the gradient normalization operation.

[0082] In this embodiment, gradient normalization operation is performed according to the maximum gradient binning to achieve fine classification of the sea-sky background image; specifically including:

[0083] Five normalization factors are binned from the maximum gradient to complete the gradient normalization. The normalization factors are In the formula, the binning constants va1 < va2 < va3 < va4 < va5;

[0084] Specifically, the normalization process is as Figure 4As shown, when the maximum gradient max_grad is greater than th1, the gradient is normalized to value1; otherwise, when the maximum gradient is greater than th2, the gradient is normalized to value2; otherwise, when the maximum gradient is greater than th3, the gradient is normalized to value3; otherwise, when the maximum gradient is greater than th4, the gradient is normalized to value4; when none of the above conditions are met, the gradient is directly normalized to value5.

[0085] The th1, th2, th3, and th4 are four thresholds for gradient binning, which are set according to specific application scenarios.

[0086] Normalization operations through maximum gradient value binning can achieve fine classification of sea-sky background images, thereby completing visibility grading. This reduces misjudgment of image quality and improves the accuracy of visibility judgment.

[0087] Specifically, when extracting edge points of the image according to the result of the gradient normalization in step S202, the canny algorithm is used to extract image edge points. Moreover, the canny algorithm is improved through the obtained gradient normalization result, and non-maximum suppression and double-threshold detection included in the canny adaptive algorithm can be completed.

[0088] More specifically, when extracting edge points of the image, horizontal and vertical gradient sets are calculated through adaptive Gaussian filtering for non-maximum suppression in the canny algorithm, so as to retain points with the largest local gradient, and the purpose of thinning the edge can be achieved.

[0089] When extracting edge points of the image, the gradient normalization result is used for high and low threshold detection of the canny algorithm; detecting the high threshold is used to reduce false edges of the image, but at the same time, the phenomenon of unclosed image edges occurs; then the low threshold is used to collect edge points whose contours meet the requirements to form new edges, thereby realizing the closure of the image edge.

[0090] Preferably, when performing high and low threshold detection, a histogram is established according to the normalized gradient according to the 16-equal division and ratio principle, and the gradient high threshold and gradient low threshold that meet the conditions are extracted using the histogram;

[0091] Gradient high threshold high_thresold = find(add(counts)>canny_highth*Δrow*Δline, 1, first);

[0092] Gradient low threshold low_thresold = canny_lowth*high_thresold;

[0093] Among them, counts represents the normalized histogram of gradients, canny_higth and canny_lowth respectively represent the preset high and low gradient thresholds, and Δrow and Δline respectively refer to the pixel width and height of the image after removing the boundaries such as black edges and white edges. find() is a function in matlab used to implement the search function; add() is a summation function in matlab.

[0094] Specifically, in step S203, the method for determining whether the sea horizon in the image can be extracted according to the number of edge points and the length information includes:

[0095] Obtain the number of points and the length information of each edge through eight-neighborhood labeling. When the number of edges is greater than the threshold, use the hough transform to extract the sea horizon; determine the images whose number and range of the sea horizon do not meet the conditions as the images where the sea horizon cannot be extracted, and the visibility classification is the poor level.

[0096] Specifically, in step S204, determine the images whose number and range of the sea horizon meet the conditions as the images where the sea horizon can be extracted; for the images where the sea horizon can be extracted, use the maximum gradient of the image for threshold comparison and judgment. When the maximum gradient of the image is less than the second gradient threshold grad2, the visibility classification of the image is the poor level; when the maximum gradient of the image is between the second gradient threshold grad2 and the third gradient threshold grad3, the visibility classification of the image is the general level; when the maximum gradient of the image is greater than the third gradient threshold grad3, the visibility classification of the image is the good level.

[0097] Specifically, in step S205, classify the images with the poor visibility classification as difficult samples and do not perform moving target detection; use the consecutive frame field-of-view images with the general and good visibility classifications as the images for moving target detection.

[0098] As Figure 5 shown, it is a more specific flowchart of the method for classifying the visibility of an image including a sea-sky background according to this embodiment.

[0099] Specifically, as Figure 6 shown, the moving target detection for the consecutive frame field-of-view images with the visibility meeting the moving target detection in step S103 includes:

[0100] Step S601: Extract the consecutive frame information, and perform differential operation on the front and rear frame images to extract the binary image;

[0101] Step S602: Filter the binary image to eliminate the influence of the sea surface and weather;

[0102] Let the filtered binary image I(x, y) be of size m×n, where the target area is A and the background area is B, i.e.:

[0103]

[0104] Step S603: Determine the candidate target area in the binary image through centroid search;

[0105] The target area in the binary image is composed of many pixel points. Through centroid search, the pixel points of the same target are classified as one target, and the center of the target area is determined; the candidate target area is determined with this center; the candidate target area contains all the pixel points classified as one target.

[0106] Through the processing of centroid search, the suspected targets formed by the scattered pixels of the same target in the next detection process are reduced, thereby reducing the processing complexity and improving the processing speed.

[0107] Step S604: Obtain the change value of the gray feature of the image in the candidate target area of the field-of-view image;

[0108] Restore the candidate target area to the field-of-view image and obtain the change value of the gray feature of the front and back frame images.

[0109] Step S605: When the change value of the gray feature of the image is greater than the overall average gray of the field-of-view image, it is determined that the candidate target area includes a moving target.

[0110] In a more preferred moving target detection method in this embodiment, as Figure 7 shown, it includes:

[0111] Step S701: Extract the continuous frame information, and establish a target search area near the sea horizon on the sea horizon in the front and back frame images for image interception;

[0112] Since the target in this embodiment is a long-distance target, its position in the field-of-view image is the part above the sea horizon and will not be confused with the sea surface. Therefore, after establishing a target search area for image interception and then performing the detection of moving targets, the interference caused by sea clutter or other targets to the detection can be avoided.

[0113] Preferably, for the detection of ship targets moving on the sea surface, the target search area is a rectangular area near the sea horizon, that is, the width direction is greater than the height direction, so as to facilitate the search for ship targets and reduce the calculation amount of the search.

[0114] Step S702: Perform a differential operation on the target search areas intercepted from the front and back frame images to extract a binary image;

[0115] Step S703: Filter the binary image to eliminate the influence of the sea surface and weather;

[0116] Let the filtered m×n - dimensional binary image be I(x, y), where the target area is A and the background area is B, that is:

[0117]

[0118] Step S704: Determine the candidate target area in the binary image through centroid search;

[0119] The target area in the binary image is composed of many pixel points. Through centroid search, the pixel points of the same target are classified as one target, and the center of the target area is determined; the candidate target area is determined with this center; the candidate target area contains all the pixel points classified as one target.

[0120] Through the processing of centroid search, the suspected targets formed by the relatively scattered pixels of the same target in the next detection process are reduced, thereby reducing the processing complexity and improving the processing speed.

[0121] Step S705: Obtain the change value of the gray - level feature of the image in the candidate target area of the field - of - view image;

[0122] Restore the candidate target area to the field - of - view image, and obtain the change value of the gray - level feature of the front and back frame images.

[0123] Step S706: When the change value of the gray - level feature of the image is greater than the overall average gray level of the field - of - view image, it is determined that there is a moving target in the candidate target area.

[0124] In summary, for the moving target detection method under the sea - sky background in the embodiments of the present invention, the visibility of the field - of - view image under the sea - sky background for each frame is graded, and the field - of - view images of consecutive frames whose visibility grading meets the requirements of moving target detection are selected for subsequent moving target detection, improving the detection probability.

[0125] Moreover, in the present invention, by improving the canny algorithm, a gradient adaptive planning method based on the maximum gradient calculation and weighted texture of the acquired image is established to complete the sea - sky background visibility judgment, realizing a more stable and effective visibility judgment.

[0126] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A moving target detection method under a sea-sky background, characterized in that, Including: Collecting the field-of-view images of the sea-sky background with continuous frames; Performing visibility grading on the obtained field-of-view images of the sea-sky background for each frame, and selecting the field-of-view images of continuous frames whose visibility grading meets the requirements for moving target detection for subsequent moving target detection; For the field-of-view images of continuous frames whose visibility grading meets the requirements for moving target detection, extracting the continuous frame information; After processing the front and back frame images including differential operation and centroid search, determining the candidate target areas in the field-of-view image for each frame, and detecting the moving targets according to the relationship between the change of gray-scale features in the candidate target areas and the overall average gray-scale of the field-of-view image; Performing visibility grading on the obtained field-of-view images of the sea-sky background for each frame, and selecting the field-of-view images of continuous frames whose visibility grading meets the requirements for moving target detection for subsequent moving target detection; including: Performing preprocessing on the input field-of-view images including downsampling, target gradient feature extraction, and gradient normalization; Extracting the edge points in the image according to the result of the gradient normalization; Judging whether the sea-sky line in the image can be extracted according to the number and length information of the edge points; For the field-of-view images of continuous frames for which the sea-sky line can be extracted, grading the visibility of the images according to the magnitude of the maximum gradient; Selecting the field-of-view images of continuous frames whose visibility grading meets the requirements for moving target detection for subsequent moving target detection.

2. The moving target detection method according to claim 1, characterized in that, The moving target detection for the field-of-view images of continuous frames whose visibility meets the requirements for moving target detection includes: Extracting the continuous frame information, and performing differential operation on the front and back frame images to extract the binary image; Filtering the binary image to eliminate the influence of the sea surface and weather; Determining the candidate target areas in the binary image through centroid search; Obtaining the change value of the gray-scale features of the image in the candidate target areas of the field-of-view image; When the change value of the gray-scale features of the image is greater than the overall average gray-scale of the field-of-view image, it is judged that the candidate target areas include moving targets.

3. The moving target detection method according to claim 1, characterized in that, The moving target detection for the field-of-view images of continuous frames whose visibility meets the requirements for moving target detection includes: Extracting the continuous frame information, and establishing a target search area near the sea-sky line in the front and back frame images to intercept the images; Performing differential operation on the target search areas intercepted from the front and back frame images to extract the binary image; Filtering the binary image to eliminate the influence of the sea surface and weather; Determining the candidate target areas in the binary image through centroid search; Obtaining the change value of the gray-scale features of the image in the candidate target areas of the field-of-view image; When the change value of the gray-scale features of the image is greater than the overall average gray-scale of the field-of-view image, it is judged that the candidate target areas include moving targets.

4. The moving target detection method according to claim 1, characterized in that, The preprocessing includes: Performing downsampling on the input field-of-view images to obtain the downsampled images; and grading the visibility of the images with gray-scale means less than the gray-scale threshold in the downsampled images as a poor level; Performing adaptive Gaussian filtering on the downsampled images with gray-scale means not less than the gray-scale threshold to calculate the target gradient feature set of the images including the horizontal, vertical gradients and the maximum gradient of the images; Grading the visibility of the images with the maximum gradient value less than the first gradient threshold as a poor level; For an image with a maximum gradient value not less than the first gradient threshold, perform gradient normalization operation to obtain the gradient normalization result.

5. The moving target detection method according to claim 1, characterized in that, Calculate the horizontal and vertical gradients of the image through adaptive Gaussian filtering; The size of the Gaussian filter is an n*n matrix. Combining with the sigma parameter of the Canny operator, the Gaussian filter is: dgau2D = -x · * exp(-(x · * x + x T · * x T ) / (2 * canny_sigma)) / (2 * π * canny_sigma 2 ); The horizontal and vertical gradient sets dx and dy are as follows: The intermediate value of the image gradient is: value = sqrt(dx.*dx + dy.*dy); The maximum gradient is: max_grad = max(value); I is the pixel matrix of the image after downsampling the resolution.

6. The motion target detection method according to claim 5, wherein, Perform gradient normalization operation according to the maximum gradient binning, specifically including: Perform gradient normalization on the maximum gradient by dividing it into 5 normalization factors. The normalization factors are shown in the following formula: In the formula, the binning constants va1 < va2 < va3 < va4 < va5; When the maximum gradient max_grad is greater than th1, the gradient is normalized to value1; otherwise, when the maximum gradient is greater than th2, the gradient is normalized to value2; otherwise, when the maximum gradient is greater than th3, the gradient is normalized to value3; otherwise, when the maximum gradient is greater than th4, the gradient is normalized to value4; when none of the above conditions are met, the gradient is directly normalized to value5.

7. The motion target detection method according to claim 1, wherein, The edge points of the image are extracted using the canny algorithm; when extracting the edge points of the image, perform non-maximum suppression in the canny algorithm using the horizontal and vertical gradient sets calculated through Gaussian filtering; When extracting the edge points of the image, use the gradient normalization result to perform high and low threshold detection in the canny algorithm; the high threshold is used to reduce the false edges of the image; the low threshold is used to collect the edge points whose contours meet the requirements to form new edges, thereby realizing the closure of the image edges.

8. The motion target detection method according to claim 1, wherein, The method for judging whether the sea horizon can be extracted from the image according to the number and length information of the edge points includes: Obtain the number and length information of each edge through eight-neighborhood labeling. When the number of edges is greater than the threshold, use the hough transform to extract the sea horizon; determine the images whose number and range of the sea horizon do not meet the conditions as images that cannot extract the sea horizon, grade the visibility as a poor level, and classify them as difficult samples without performing moving target detection.

9. The motion target detection method according to claim 8, wherein, Determine the images whose number and range of the sea horizon meet the conditions as images that can extract the sea horizon; for the images that can extract the sea horizon, use the maximum gradient of the image to perform threshold comparison and judgment. When the maximum gradient of the image is less than the second gradient threshold, the visibility of the image is graded as a poor level; when the maximum gradient of the image is between the second gradient threshold and the third gradient threshold, the visibility of the image is graded as a general level; when the maximum gradient of the image is greater than the third gradient threshold, the visibility of the image is graded as a good level; Among them, the images with a poor visibility grade are classified as difficult samples without performing moving target detection; the consecutive frame field-of-view images with a general and good visibility grade are used as images for moving target detection.

Images