A method for detecting moving object shadows using road information

Abstract

The present invention discloses a method for detecting the shadow of a moving target using road information. The method uses edge detection and line detection to extract road edge information, excludes clutter areas outside the road, and obtains the ROI area of the SAR image. Edge detection is used to remove clutter and partially connected areas, and the partially connected areas are merged to obtain shadow detection results. In specific scenarios, the method can efficiently detect the shadow of a moving target, reduce false alarms in shadow detection, and improve detection efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and in particular relates to a target detection technology. Background Art

[0002] Moving target detection, tracking, and localization are crucial tasks in video SAR (Spectral Aperture Recognition) (SAR). Target detection is a prerequisite for both tracking and localization. Moving targets in SAR images suffer from defocus and positional offsets, making them difficult to detect and accurately locate. Shadows can reveal the true location of targets, making detecting shadows of moving targets extremely valuable.

[0003] Methods for detecting moving targets in SAR images include:

[0004] The paper "Moving Target Detection Method in a Single High-Resolution SAR Image Based on Shadow Detection" describes a moving target detection method based on a single SAR image. This method detects moving targets by detecting shadow areas and, combined with road auxiliary information, completes target detection, target positioning, and parameter estimation.

[0005] In the paper "Video SAR Imaging and Moving Target Shadow Detection Technology," a target shadow detection method is disclosed. This method analyzes the formation mechanism of video SAR shadows and their impact on moving target detection performance, and uses machine learning methods to detect shadowed targets in video SAR.

[0006] In "A VideoSAR Moving Target Shadow Detection Method," a video SAR shadow detection method is disclosed. This method first registers the SAR images to suppress speckle, then performs background modeling on multiple adjacent frames and uses a three-frame difference method to extract the target shadow.

[0007] Judging from the existing technology, there is an urgent need for a simple and efficient shadow detection method that does not require a large number of samples for machine learning training. Summary of the Invention

[0008] To address the problems of the prior art, the present invention proposes a moving target shadow detection method using road information. The method identifies road edges, selects an ROI region in a SAR image, and uses a shadow detection method to detect the ROI region. The method utilizes road edge information to eliminate the influence of stationary target shadows, thereby improving detection efficiency and accuracy. To achieve efficient detection of target shadows in the above-mentioned specific scenarios, the present invention adopts the following technical solutions.

[0009] Edge detection and line detection are used to extract road edge information, exclude clutter areas outside the road, and obtain the ROI area of the SAR image; edge detection is used to remove clutter and partially connected areas, and partially connected areas are merged to obtain shadow detection results.

[0010] Furthermore, obtaining the ROI region of the SAR image includes: Gaussian smoothing filtering the SAR image to reduce the influence of speckle noise on edge detection, using the Canny operator to detect edges in the SAR image, and binarizing the detection results; using Radon transform to calculate the slope of a straight line in the binary image, and using the slope to traverse the straight line to search and determine the specific location of the road edge; setting the pixel values of the area outside the road edge of the SAR image to 0 to delete the information outside the road edge of the SAR image, thereby generating the ROI region of the SAR image.

[0011] Specifically, determining the specific position of the road edge includes: assuming that the horizontal direction of the image is the x-axis of the coordinate system, the vertical direction is the y-axis of the coordinate system, the SAR image is I, and the row and column sizes are N respectively. r and N c , the size of the Gaussian filter is [N,N], the standard deviation is σ, the smoothed SAR image is I′, and the SAR image after edge detection is I e ′, the binarization threshold is T bw1 , binary SAR image is I b ' w The angle θ corresponding to the maximum point is the straight line inclination angle of the SAR image, and the slope of the straight line is The edge value is 1, the non-edge value is 0, and the search image I is traversed. b ' w Among all the lines with a slope of k, the two lines that overlap with the pixel with the value of 1 the most are selected as the road edges.

[0012] In particular, if |k|≤1, let the equation of the line y=k·(x-x0)+y0, let x0=N c / 2, the value range of y0 is [1,N r ], calculate the number of overlaps between the straight line corresponding to each integer value of y0 and the pixel with a value of 1, select the two y0 values with the largest number of overlaps as the parameter values for solving the straight line equation, calculate the specific positions of the two straight lines as the road edge positions; if 1<|k|<∞, set the straight line equation x=k·(y-y0)+x0, let y0=N r / 2, the value range of x0 is [1,N c ], calculate the number of overlaps between the straight line corresponding to each integer value of x0 and the pixel with a value of 1, select the two x0 values with the largest number of overlaps as the parameter values for solving the straight line equation, calculate the specific positions of the two straight lines as the road edge positions; if |k| = ∞, set the straight line equation x = x0, let the value range of x0 be [1, N c], calculate the number of overlaps between the straight line corresponding to each integer value of x0 and the pixels with a value of 1, select the two x0 values with the largest number of overlaps as the parameter values for solving the straight line equation, and calculate the specific positions of the two straight lines as the road edge positions.

[0013] Furthermore, obtaining shadow detection results includes: Gaussian smoothing filtering of the ROI area of the SAR image to reduce the influence of speckle noise on edge detection, using the Canny operator to detect the edge of the ROI area, binarizing the detection results, indenting the road edge of the binary image inward by 10 pixels, and eliminating the influence of the edge of the ROI area on edge detection; removing clutter from the binary image, setting a shadow size threshold, and deleting connected areas of the binary image with a size smaller than the threshold; calculating the distance between any two connected areas of the binary image, setting a distance threshold, using the image dilation method, merging connected areas with a distance smaller than the threshold, and marking the SAR image.

[0014] Specifically, shadow detection includes: setting the ROI area as I ROI , the size of the Gaussian filter is [N,N], the standard deviation is σ, and the smoothed ROI area is I′ ROI , the ROI area after edge detection is I′ ROI_e , the binarization threshold is T bw2 , the binary ROI area is I′ ROI_bw The process of generating the ROI region deletes part of the SAR image content, resulting in a large amount of false information with a value of 1 at the edge of the road in the binary image. The road edge of the ROI region is indented inward by several pixels to eliminate the binary edge effect caused by the deletion of the SAR image content; let the length and width of the target shadow be S, set the threshold by S, and delete I′ ROI_bw Less than The connected area is set to 0, and the pixel values of the connected area are updated and merged to obtain the connected area I″ ROI_bw , taking S as the side length of the square, use a box to select the shadow in the SAR image I as the shadow detection result of the SAR image.

[0015] In particular, updating the connected regions includes: using the CFAR algorithm, setting the center coordinates of each connected region to [x c ,y c ], select the image block I with row and column size S blk_c , calculate the mean M of the image block c , respectively [x c -S / 2,y c ]、[x c +S / 2,y c ]、[x c ,y c -S / 2] and [x c ,yc +S / 2] as the center, select four image blocks with row and column size S, and calculate the total mean M4 of the four image blocks. If M c If it is >2M4 / 3, the connected area is retained; otherwise, it is deleted and the pixel values of the connected area are set to 0.

[0016] In particular, merging connected regions includes: the same shadow may be split into multiple connected regions in some cases, setting the threshold T d , if I′ ROI_bw The distance between any two connected regions is less than T d , then the two connected regions are determined to be the same target and merged.

[0017] The beneficial effects of the present invention are as follows: edge detection and line detection are used to obtain road edge information, interference of clutter outside the road on shadow detection is eliminated, and the edge detection method is used to detect the shadow of a moving target. In specific scenarios, the shadow of a moving target can be efficiently detected, false alarms in shadow detection can be reduced, and detection efficiency can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is the original SAR image, Figure 2 is the SAR image after Gaussian smoothing filtering, Figure 3 is the edge detection result of the SAR image, Figure 4 is a binary image of the SAR image edge detection result, Figure 5 is the Radon transform result, Figure 6 is the ROI area, Figure 7 is the edge detection result of the ROI area, Figure 8 is the ROI edge detection result after removing clutter, Figure 9 is the shadow detection result. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0020] SAR images such as Figure 1 As shown in , the size of the Gaussian filter is [40, 40], the standard deviation is 7, and the Gaussian smoothing filter SAR image I is used to reduce the impact of speckle noise on edge detection. The smoothed SAR image I′ is as follows Figure 2 shown.

[0021] Use the Canny operator to perform edge detection on I′ and obtain the SAR edge information image I′ e ,like Figure 3 As shown, the threshold T is selected bw1 =10, binary image I′ e , get the binary image I′ bw ,like Figure 4 shown.

[0022] The row and column sizes of image I are N r =1680 and N c =1350, for binary image I′ bw Perform Radon transform to obtain a two-dimensional map of angles and distances, such as Figure 5 As shown, the row coordinates are angles, the column coordinates are distances, and the calculated slope k = 1.7321.

[0023] According to 1<|k|<∞, the number of overlaps between the straight line and the binary image value of 1 pixel is calculated for different x0 conditions. The two x0 values with the largest overlap are found, and the equations of the two straight lines are determined as the positions of the two edges of the road in the SAR image.

[0024] The ROI area in the SAR image is the road area. After determining the road edge information, the information outside the road edge in the SAR image is deleted to obtain the SAR image ROI area I ROI ,like Figure 6 shown.

[0025] Gaussian filter on I ROI , do Canny edge detection and binarization processing, binarization threshold T bw2 =15, indent the road edge 10 pixels into the road in the binary image to eliminate the influence of the ROI area edge on edge detection, and obtain the ROI area edge detection binary image I′ ROI_bw ,like Figure 7 shown.

[0026] Assume that the length and width of the target shadow are both S=15, extract I′ ROI_bw All connected domains in the image are first screened by size, and false alarms of connected domains with smaller sizes are deleted. Then, preliminary detection results of shadows are obtained by CFAR detection, such as Figure 8 shown.

[0027] In some cases, the same shadow may detect multiple connected domains, and I′ is calculated. ROI_bw The distance between any two connected domains in the d =10, then the two connected domains are considered to be the same target, and the two connected domains are merged to obtain I″ ROI_bw .

[0028] Extract I″ ROI_bw The location information of all connected domains is centered on the centroid of the connected domain and the side length of the square is S = 15. The shadows are marked with a box in the SAR image to complete the shadow detection of the SAR image, as shown in the following example: Figure 9 shown.

[0029] The above are embodiments of the present invention and do not limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention are included in the protection scope of the present invention.

Claims

1. A moving target shadow detection method using road information, characterized in that: include: Edge detection and line detection are used to extract road edge information and exclude clutter areas outside the road; Gaussian smoothing filter is used to filter the SAR image, Canny operator is used to detect the edge of the SAR image, and the detection results are binarized; Radon transform is used to calculate the slope of the line of the binary image, and the line is searched based on the slope; let the horizontal axis of the image be the x-axis of the coordinate system, the vertical axis be the y-axis of the coordinate system, the SAR image be I, and the row and column sizes be N respectively. r and N c , the size of the Gaussian filter is [N,N], the standard deviation is σ, the smoothed SAR image is I′, and the SAR image after edge detection is I e ′, the binarization threshold is T bw1 , the binary SAR image is I′ bw The angle θ corresponding to the maximum point is the straight line inclination angle of the SAR image, and the slope of the straight line is The edge value is 1, the non-edge value is 0, and the search image I is traversed. b ' w For all straight lines with a slope of k, the two straight lines with the largest number of overlaps with pixels with a value of 1 are selected as road edges; the pixel values of the area outside the road edge of the SAR image are set to 0 to delete the information outside the road edge of the SAR image and generate the ROI area of the SAR image; edge detection is used to remove clutter and partially connected areas, and partially connected areas are merged; Gaussian smoothing filter is used to filter the ROI area of the SAR image to reduce the influence of speckle noise on edge detection, Canny operator is used to detect the ROI area, the detection result is binarized, the road edge of the binary image is indented 10 pixels inward, and the influence of the edge of the ROI area on edge detection is eliminated; clutter of the binary image is removed, a shadow size threshold is set, and connected areas of the binary image with a size smaller than the threshold are deleted; the distance between any two connected areas of the binary image is calculated, a distance threshold is set, and the image dilation method is used to merge connected areas with a distance smaller than the threshold, and the SAR image is marked; let the ROI area be I ROI , the size of the Gaussian filter is [N,N], the standard deviation is σ, and the smoothed ROI area is I′ ROI , the ROI area after edge detection is I′ ROI_e , the binarization threshold is T bw2 , the binary ROI area is I′ ROI_bw The process of generating the ROI region deletes part of the SAR image content, resulting in a large amount of false information with a value of 1 at the edge of the road in the binary image. The road edge of the ROI region is indented inward by several pixels to eliminate the binary edge effect caused by the deletion of the SAR image content; let the length and width of the target shadow be S, set the threshold by S, and delete I′ ROI_bw Less than The connected area is set to 0, and the pixel values of the connected area are updated and merged to obtain the connected area I″ ROI_bw , taking S as the side length of the square, use a box to select the shadow in the SAR image I as the shadow detection result of the SAR image.

2. The method for detecting moving target shadows using road information according to claim 1, characterized in that: Determine the specific location of the road edge, including: if |k|≤1, set the straight line equation y=k·(x-x0)+y0, let x0=N c / 2, the value range of y0 is [1,N r ], calculate the number of overlaps between the straight line corresponding to each integer value of y0 and the pixel with a value of 1, select the two y0 values with the largest number of overlaps as the parameter values for solving the straight line equation, calculate the specific positions of the two straight lines as the road edge positions; if 1<|k|<∞, set the straight line equation x=k·(y-y0)+x0, let y0=N r / 2, the value range of x0 is [1,N c ], calculate the number of overlaps between the straight line corresponding to each integer value of x0 and the pixel with a value of 1, select the two x0 values with the largest number of overlaps as the parameter values for solving the straight line equation, calculate the specific positions of the two straight lines as the road edge positions; if |k| = ∞, set the straight line equation x = x0, let the value range of x0 be [1, N c ], calculate the number of overlaps between the straight line corresponding to each integer value of x0 and the pixels with a value of 1, select the two x0 values with the largest number of overlaps as the parameter values for solving the straight line equation, and calculate the specific positions of the two straight lines as the road edge positions.

3. The method for detecting moving target shadows using road information according to claim 1, wherein: The updating of connected regions includes: using the CFAR algorithm, setting the center coordinates of each connected region to [x c ,y c ], select the image block I with row and column size S blk_c , calculate the mean M of the image block c , respectively [x c -S / 2,y c ]、[x c +S / 2,y c ]、[x c ,y c -S / 2] and [x c ,y c +S / 2] as the center, select four image blocks with row and column size S, and calculate the total mean M4 of the four image blocks. If M c If it is >2M4 / 3, the connected area is retained; otherwise, it is deleted and the pixel values of the connected area are set to 0.

4. The method for detecting moving target shadows using road information according to claim 1, wherein: The merging of connected regions includes: the same shadow may be split into multiple connected regions in some cases, and the threshold T is set. d , if I′ ROI_bw The distance between any two connected regions is less than T d , then the two connected regions are determined to be the same target and merged.

Images