CN110210295B - High-precision target identification detection method under grid background - Google Patents
High-precision target identification detection method under grid background Download PDFInfo
- Publication number
- CN110210295B CN110210295B CN201910334915.9A CN201910334915A CN110210295B CN 110210295 B CN110210295 B CN 110210295B CN 201910334915 A CN201910334915 A CN 201910334915A CN 110210295 B CN110210295 B CN 110210295B
- Authority
- CN
- China
- Prior art keywords
- target
- region
- center
- interest
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 11
- 238000000034 method Methods 0.000 claims abstract description 44
- 238000001914 filtration Methods 0.000 claims abstract description 31
- 238000012545 processing Methods 0.000 claims abstract description 14
- 230000011218 segmentation Effects 0.000 claims description 22
- 238000003708 edge detection Methods 0.000 claims description 11
- 230000000875 corresponding effect Effects 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 5
- 238000013461 design Methods 0.000 claims description 5
- 125000004122 cyclic group Chemical group 0.000 claims description 4
- 230000005587 bubbling Effects 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000002596 correlated effect Effects 0.000 claims description 2
- 238000003384 imaging method Methods 0.000 claims description 2
- 239000002904 solvent Substances 0.000 claims description 2
- 238000011084 recovery Methods 0.000 abstract description 14
- 230000008569 process Effects 0.000 abstract description 8
- 238000012216 screening Methods 0.000 abstract description 7
- 230000000877 morphologic effect Effects 0.000 abstract description 5
- 238000005070 sampling Methods 0.000 abstract description 5
- 230000000007 visual effect Effects 0.000 abstract description 5
- 230000001133 acceleration Effects 0.000 abstract description 3
- 238000011897 real-time detection Methods 0.000 abstract description 2
- 230000007797 corrosion Effects 0.000 description 7
- 238000005260 corrosion Methods 0.000 description 7
- 230000003628 erosive effect Effects 0.000 description 6
- 230000010339 dilation Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000008030 elimination Effects 0.000 description 3
- 238000003379 elimination reaction Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013016 damping Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000005530 etching Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000008961 swelling Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Abstract
The invention discloses a high-precision target identification and detection method under a grid background. The method can realize real-time and high-precision identification of the target under the grid background, and provides reliable visual navigation information for recovery of unmanned aerial vehicle network collision. The method comprises the steps of firstly designing a target, obtaining a proper binary threshold value based on an algorithm for updating a gradient search threshold value and an optimal threshold value, realizing target identification under a mesh background by using a morphological filtering algorithm and a coarse and fine contour screening algorithm, realizing algorithm acceleration by adopting a method of generating an area of interest and reducing sampling in order to realize real-time detection, and providing position resolving information for navigation by adopting an image processing mode in the process of recovering the unmanned aerial vehicle from a network so that the unmanned aerial vehicle can realize the function of autonomous recovery.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to a high-precision target identification and detection method under a grid background.
Background
With the continuous development of the unmanned aerial vehicle technology, the self-service recovery mode of the fixed-wing unmanned aerial vehicle mainly comprises recovery modes such as overhead hooking, parachute recovery, collision net collision rope recovery, air bag damping recovery and recoil rocket. Wherein hit the net and retrieve and to make unmanned aerial vehicle reduce the damage of organism and airborne equipment that causes because of gust and ground striking, simultaneously, hit the net and retrieve and can also retrieve the umbrella with unmanned aerial vehicle and remove to increase the task load on the machine or increase the oil mass, prolong the time of endurance. At present, the recovery of the collision net is realized mainly by adopting a visual guide mode, the reliability and the accuracy of a visual guide part have a decisive effect on the recovery of the collision net, and the following problems exist in the existing visual identification method in the recovery process of the collision net:
(1) the distance between the unmanned aerial vehicle and the target landing net cannot be accurately calculated through a pure center alignment algorithm, and high requirements are provided for a collision net recovery strategy.
(2) The target identification algorithm based on the color has poor sunlight resistance, and the color is easily interfered by similar targets.
(3) The recognition algorithm based on two-dimensional codes such as Aruco Marker and the like cannot solve the problem of influence of the grids of the recovery net on the target contour.
In summary, the following steps: the prior art cannot realize the remote high-precision identification under the grid background.
Disclosure of Invention
In view of the above, the invention provides a high-precision target identification and detection method under a grid background, which can realize real-time and high-precision identification of a target under the grid background and provide reliable visual navigation information for recovery of network collision of an unmanned aerial vehicle.
The invention provides a high-precision target identification and detection method under a grid background, which comprises the following steps:
step 2.1, performing low-pass filtering on the current frame image and then performing binarization processing; the binary segmentation threshold value is changed in a cyclic gradient mode within a set threshold value range, and one threshold value is selected for calculation in each frame; setting a binary segmentation threshold of the initial frame as the maximum value or the minimum value of the threshold range;
step 2.2, expanding and corroding the image after binarization processing;
step 2.3, extracting the outline of the image processed in the step 2.2, and judging whether the extracted outline meets requirements a) to c), if so, judging that the target is identified, and executing the step 3; otherwise, judging that the target is not identified, extracting the next frame of image, executing the step 2.1, and setting a binary segmentation threshold of the next frame according to the gradient change rule of the step 2.1;
wherein a) to c) are required:
a) the extracted outer contour is consistent with the outer contour shape of the designed target baseplate;
b) the extracted outer contour is internally provided with N inner contours, and the ratio of the sum of the areas of all the inner contours to the area of the outer contour is 0.7-1.3 times of the ratio of the sum of the areas of all icons in the design target to the area of the bottom plate;
c) designing inner outlines corresponding to icons positioned on the same straight line on the target and still being collinear;
converting the target area identified by the current frame into a gray histogram, and then updating a binarization threshold value according to the following formula:
wherein the threshold is an updated binarization threshold; the value range of i is 0-255; p (t) is the number of pixels with the gray value t; sblackIs the sum of the areas of the N inner contours; sallIs the area of the outer contour of the target;
and (5) repeatedly executing the step 3 to the step 4 until the aircraft successfully hits the net.
Preferably, in step 3, the size of the target dimension K is determined according to the pixel area of the target outer contour identified in the previous frame, and the region of interest of the next frame image is determined according to the value K:
if S is more than 0 and less than or equal to 4000, K is 1, and the region of interest for extracting the next frame image is: the center of the region of interest is the same as the center of the target region identified in the previous frame, the height of the region of interest is seven times of the height of the target region in the previous frame, and the width of the region of interest is three times of the width of the target region in the previous frame;
if S is more than 4000 and less than or equal to 30000, K is 2, and the region of interest for extracting the next frame image is: the center of the region of interest is the same as the center of the target region identified in the previous frame, the height of the region of interest is five times of the height of the target region in the previous frame, and the width of the region of interest is three times of the width of the target region in the previous frame;
if 30000 < S ≦ 80000, K ≦ 3, and the region of interest for extracting the next frame image is: the center of the region of interest is the same as the center of the target region identified in the previous frame, the height of the region of interest is three times of the height of the target region in the previous frame, and the width of the region of interest is three times of the width of the target region in the previous frame;
if S > 80000, K is 4, the next frame image is down-sampled, and target recognition is performed on the down-sampled image.
Preferably, after the modified binarization threshold is obtained by calculation using formula (4), the modified threshold and the binarization threshold of the current frame are subjected to smooth filtering to obtain an optimal threshold, and the optimal threshold is used for target identification according to the method in step 2.
Preferably, in the step 1, a black frame is arranged on the periphery of the bottom plate.
Preferably, in step 2.1, low-pass filtering is performed by using a gaussian filtering method.
Preferably, in the step 2.1, a gaussian filtering method is adopted for low-pass filtering; after the target is identified, the size of a Gaussian kernel of the next frame of filtering is set according to the target scale; wherein the gaussian kernel size is positively correlated with the target scale.
Preferably, in step 2.2, the expanded nuclei are smaller than the eroded nuclei.
Preferably, in the step 2.3, a Canny edge detection operator is adopted to perform contour extraction.
Preferably, in the step 2.3, the corresponding relationship between the inner contour and the icon on the target is determined by the following method:
s1, respectively calculating the distance between the center of the N inner contours and the center of the outer contour of the target, wherein the inner contour with the shortest distance to the center of the outer contour corresponds to the icon at the center position in the target and is marked as a center icon;
s2, performing polygon fitting on the outer contour of the target, wherein the two nearest vertexes in the pentagon are the unfilled corners; in the N inner outlines, the inner outline with the shortest distance from the center of the inner outline to the two vertexes is the icon closest to the vacancy in the corresponding target and marked as a No. 0 icon;
and S3, turning to a polar coordinate system by taking the center icon as the center and the ray from the center of the center icon to the center of the No. 0 icon as a polar axis, and carrying out bubbling sequencing on polar angles of the centers of the rest inner contours to finish the correspondence between the rest inner contours and the target icons.
Has the advantages that:
the invention designs a target, obtains a proper binary threshold value based on an algorithm for updating a gradient search threshold value and an optimal threshold value, realizes target identification under a mesh background by using morphological filtering and a coarse and fine contour screening algorithm, realizes algorithm acceleration by adopting a mode of generating an area of interest and reducing sampling in order to realize real-time detection, and provides position resolving information for navigation by adopting an image processing mode in the process of recovering an unmanned aerial vehicle from a network so that the unmanned aerial vehicle can realize the function of autonomous recovery.
The centers of the N icons on the bottom plate in the target can be used when the target position is calculated, and high-precision calculation of the target position can be ensured.
Drawings
FIG. 1 is a schematic target.
FIG. 2 is an image comparison before and after Gaussian filtering; wherein, (a) an original image; (b) a gaussian filtered image.
FIG. 3 is an image comparison before and after OTSU binarization processing; wherein, (a) the original image; (b) and (5) binarizing the image.
FIG. 4 is a schematic representation of binarization of a target portion; wherein, (a) a gaussian filtered image; (b) and (4) binarizing the image.
Fig. 5 is a schematic view of the expansion principle.
FIG. 6 is a schematic view of the corrosion principle.
FIG. 7 is a schematic view of the expansion operation; wherein (a) the image before inflation; (b) the expanded image.
FIG. 8 is a schematic view of an etching operation; wherein, (a) the pre-etch image; (b) and (4) an etched image.
FIG. 9 is a schematic diagram of edge extraction; wherein, (a) the original image; (b) a Canny edge; (c) an expanded edge.
FIG. 10 is a schematic view of collinear detection.
FIG. 11 is a schematic view of a region of interest; wherein, (a) the target region of the previous frame of image is shown schematically; (b) and the region of interest of the next frame image is shown schematically.
Fig. 12 is a region of interest extraction with the target at the boundary.
FIG. 13 is a flow chart of the present invention.
Detailed Description
The invention is described in detail below by way of example with reference to the accompanying drawings.
The invention provides a high-precision target identification and detection method under a grid background, which comprises the following steps:
the method comprises the following steps: design target
The target is a white bottom plate with one corner cut out; the bottom plate is provided with an icon array formed by regularly arranging a plurality of same icons; the size of the icons, the spacing between the icons is known. The icons are regular figures such as circles, rectangles, hexagons or stars and are black.
Wherein, in order to improve the reliability of extracting the outline, the black outer frame is additionally arranged on the periphery of the bottom plate, thereby being more beneficial to extracting the outline of the bottom plate.
As shown in fig. 1, target parameters: the bottom plate is a white square without a corner, the icon array is a standard rectangular array consisting of 9 black circular icons, the diameter of each circular icon is 15cm, the distance between the centers of circles of adjacent icons is 15cm, the distance between the center of the circle of the outermost circular icon and the inner boundary of the black outer frame is 15cm, and the length of the corner is 15 cm.
After the target is designed, the designed target is used to realize target identification of the aircraft in the flight process, the identification process is shown in fig. 13, and the target shown in fig. 1 is taken as an example in this embodiment, which is specifically described as follows:
step two: the method specifically comprises the following substeps of identifying a target of a shot image by an aircraft in the flying process:
step 2.1, low-pass filtering is carried out on the image, the influence of noise on identification is reduced, and fuzzy processing on grids is achieved; then, carrying out binarization processing on the image subjected to low-pass filtering;
(1) low pass filtering
The fuzzy processing of the grid can be realized by adopting methods such as Gaussian filtering, median filtering and the like. The gaussian blur is an advantage of a low-pass filter having a smoother characteristic, and is widely used for image noise reduction. Gaussian filtering is often used to remove detail prior to edge detection. The present embodiment employs gaussian filtering, which can be performed by the function gaussian blur in OpenCV. The result of the gaussian filtering is shown in fig. 2, wherein, after the target has been identified, a suitable gaussian kernel can be selected according to the scale of the target (target) in each frame of image, and the criteria are: the size of the Gaussian kernel is in positive correlation with the target scale, so that the extraction of the circle center is not influenced while the line is blurred. And because the initial flying height is higher, the initial Gaussian kernel can be selected to be 3 x 3, and the oversized Gaussian kernel can be blurred, so that the identification effect is poor.
(2) Binarization processing
Performing binary segmentation on the low-pass filtered image by using a set segmentation threshold value; among them, the most common threshold selection algorithm at present is the maximum inter-class variance method, which was proposed by the scholars of japan (Nobuyuki Otsu) in 1979 and is a method for determining a self-adaptive threshold, which is also called the atsu method, for short. However, the algorithm has certain limitations, and cannot segment small objects in a large background scene, for example, when a target in the scene is to be identified, after binarization is performed by using a threshold value obtained by the Otsu method, only a road surface and a sidewalk can be distinguished, and an area of the target cannot be accurately segmented, as shown in fig. 3.
Therefore, the invention provides an algorithm for circularly searching the threshold value to adapt to different scenes. The cyclic search algorithm is to determine a threshold variation range at first, test according to a certain gray scale interval in the range, because the camera resolution is higher, the calculated amount is larger in the process of filtering and eliminating noise, and therefore, only one threshold test is performed on each frame. As shown in fig. 4. For example, if the threshold variation range is set to [120,200] and the threshold variation gradient is set to 20, the binarization threshold of the next frame may be set to be increased by 20 from 120 compared with the previous frame, and the binarization threshold of the next frame may be reset to 120 if the binarization threshold of the previous frame has reached 200. It can also be varied from 200 gradients to 120. By using the method of cyclic gradient change of the threshold, the segmentation threshold of the optimal binary segmentation effect is determined. In the selection of the variation gradient, it can be judged according to specific situations that if the flight speed of the aircraft is slow, a smaller interval can be selected to ensure the identification accuracy, and if the flight speed of the aircraft is fast, a larger interval is selected, and the selection of the general interval is 5, 10, 15 or 20.
In addition, in the actual flight process, the situation that the individual frames cannot be identified due to complex environmental factors such as occlusion and the like may exist, so a threshold value is set, and when the number of times of continuous identification exceeds the threshold value, the loop search for the appropriate threshold value is restarted in a mode of searching for the threshold value from the beginning.
And 2.2, finishing the elimination of the grids by using morphological filtering.
Morphological filtering consists mainly of erosion and dilation.
First, the concept of erosion and expansion is introduced: erosion and dilation are areas of high brightness where the dilated image has a larger area of high brightness than the original image, and erosion is the opposite. Mathematically, the dilation or erosion operation is the convolution of an image (or a portion of an image, referred to as a) with a kernel (referred to as B). A kernel can be any shape and size that has a separately defined reference point, called anchor point (anchor). In most cases, the kernel is a small middle with reference points and a solid square or disk, and thus, the kernel can be considered as a template or mask.
The expansion is an operation of solving a local maximum, the kernel B is convoluted with the graph A, namely, the maximum value of the pixel points of the area A covered by the kernel B is calculated, and the maximum value is assigned to the pixel specified by the reference point. This causes the highlight region in the image a to grow gradually, which is the original intention of the expansion operation, as shown in fig. 5. And the corrosion is an operation of solving a local minimum value, the kernel B is convoluted with the graph A, namely the minimum value of pixel points of the area A covered by the kernel B is calculated, and the minimum value is assigned to a pixel specified by a reference point.
Mathematical expression for dilation:
mathematical expression for corrosion:
the corrosion principle is schematically shown in fig. 6.
In the embodiment, the method of expansion first and corrosion second is adopted to carry out morphological filtering processing on the binary image, and the method of expansion first and corrosion second is adopted to ensure that the target edge is not ablated during remote identification. In addition, in this embodiment, a method of small-kernel swelling and large-kernel erosion is adopted, in which a 3 × 3 kernel is firstly swelled, and then a 5 × 5 kernel is used to eliminate the remaining points of the grid, so that it is possible to avoid the situation that too much elimination is likely to occur if the 5 × 5 kernel is directly used to perform grid elimination in the case of a small target.
The expansion operation is schematically shown in fig. 7, and the corrosion operation is schematically shown in fig. 8.
Step 2.3, contour extraction is carried out
After the processing of step 2.2, the background grids are eliminated, and on the basis, the outlines of the target baseplate and the icons on the baseplate are extracted. Wherein, the Canny edge detection operator, the Laplacian operator, the Sobels operator and other methods can be used for extracting the edge; the Canny edge detection operator is a multi-stage edge detection algorithm developed by John f. More importantly, Canny has created the edge detection computing theory (Computational and edge detection), explaining how this technique works. The Canny edge detection algorithm is named by the name of Canny and is favored by many people as the best edge detection algorithm at present. The present embodiment employs a Canny edge detection operator.
Since the extracted edge may be discontinuous due to the problem of non-maximum suppression, the contour is expanded and then extracted before the contour extraction, and only the outer contour of the figure is considered in the contour extraction process. The edge extraction diagram is shown in fig. 9.
And 2.4, screening the extracted contour to extract a target. The invention designs coarse screening and fine screening, and solves the problem of efficient screening of the candidate target profile.
(1) And (3) performing coarse screening on the extracted images:
firstly, detecting whether the extracted outer contour is consistent with the outer contour shape of a designed target baseplate; and then, detecting whether N inner contours (N is the number of the graphics in the target) exist inside the outer contour, wherein the ratio of the sum of the areas of all the inner contours to the area of the outer contour is 0.7-1.3 times of the ratio of the sum of the areas of all the icons in the designed target to the area of the bottom plate. According to the target, the proportion of each circle in the target is constant, and if m% is assumed, if the areas of the N inner contours to be detected all satisfy: and judging that a reliable outline is extracted when the outline area of the whole target is 0.7 m% -1.3 m%. And fitting the circle center coordinates of the N inner contours, and determining the corresponding relation between the N inner contours and the icon in the target.
The method comprises the following steps of:
i) and searching the circle center coordinate with the shortest distance from the nine circle center coordinates to the center coordinate of the outer contour according to the distance between the nine circle center coordinates and the center of the outer contour, wherein the circle center is the circle center corresponding to the No. 4 icon in the target.
ii) performing polygon fitting on the external contour, wherein two nearest vertexes in the pentagon vertexes are unfilled corners, and a point which is closest to the sum of the distances between the two vertexes is the circle center corresponding to the icon No. 0.
iii) ranking nine points using bubble ranking: and (3) taking the circle center of No. 4 as the center, taking the ray from the circle center of No. 4 to the circle center of No. 0 as a polar axis, turning to a polar coordinate system, carrying out bubbling sequencing on polar angles of the rest circle centers, and arranging the polar angles from small to large in a clockwise positive mode, so that the rest circle centers correspond to the icons No. 1, 2, 3, 5, 6, 7 and 8 in the icon array.
(2) And (3) performing fine detection on the extracted image:
the method utilizes the security of perspective transformation, namely three points in a straight line in space are still on a line in an image, so as to circularly detect collinear central points, thereby increasing the robustness of the system. I.e. the algorithm detects if the eight lines (three horizontal, three vertical and two diagonal) as in fig. 10 are still straight lines.
If the collinear relation corresponding to each circle center is not changed, judging and identifying the target, and extracting the next frame to execute the step three; otherwise, judging that the target is not identified, extracting the next frame of image, executing the step 2.1, and setting the binarization threshold of the next frame according to the gradient change rule of the step 2.1.
In addition, if the target is identified in the previous frame of image, the region of interest of the next frame can be extracted by utilizing the interframe information, and the target identification is carried out in the region of interest in the next frame by utilizing the mode of the step two until the aircraft hits the net, so that the target identification is finished; therefore, the target identification can be directly carried out on the key attention area (the interested area) in the next frame of image, the identification speed of the target identification is improved, and the real-time performance of the algorithm is ensured. In addition, after the target is identified in the current frame, the binary segmentation threshold of the previous frame can be updated by utilizing the black-white ratio of the target in the next frame, and the updated binary segmentation threshold is adopted for image segmentation during the identification of the next frame, so that the binary segmentation effect is better, and the environment change interference resistance is improved; namely, after the target is identified, the next frame of image is processed by adopting the mode of the third step to the fourth step:
step three, after the detection is successful, updating the threshold value:
because the shape and size of the target are known, the optimal binary segmentation threshold value is the target and the actual target in the binarized image, the proportion of black and white areas is consistent, then the target area identified in the previous frame is converted into a gray histogram, a more appropriate threshold value is searched by using the proportion of nine circle centers occupying the whole target, the threshold value updating algorithm is shown as a formula (3), and the threshold value updated by the formula (3) is used as the binary segmentation threshold value of the next frame. In addition, in order to avoid too violent threshold oscillation, the corrected threshold is subjected to smooth filtering to obtain an optimal threshold, and the optimal threshold is used for carrying out binarization segmentation on the next frame.
Wherein threshold is a corrected threshold; i is a gray value between 0 and 255; p (t) is the number of pixels with the gray value t; sblackIs the area of the black area in the target; sallIs the target area.
Step four: and generating an interested area by utilizing the interframe information and ensuring the instantaneity by using a down-sampling algorithm.
For 2048 × 2048 high-resolution images, performing global image filtering processing is extremely computationally expensive, and thus the real-time performance of the system cannot be guaranteed.
Therefore, if the target is identified in the previous frame, the region of interest of the frame is defined according to the target region detected in the previous frame and the interframe information, the target is identified in the region of interest of the frame according to the second step, the identification speed is improved, the real-time identification is ensured, and the binary segmentation threshold in the identification process of the frame is the updated binary segmentation threshold.
And repeating the third step and the fourth step until the aircraft successfully hits the net.
In addition, in the fourth step, the target pixel area of the previous frame is scaled, and the regions of interest with different sizes are respectively extracted according to different scales, so that the identification speed can be further increased. The scale can be divided as follows:
wherein K is a target scale; and S is the area of the target pixel of the previous frame, namely the area of the outer contour pixel of the target in the previous frame of image.
The next frame generates different regions of interest according to different scales, as shown in fig. 11: the black area is the minimum circumscribed quadrangle of the target extracted from the previous frame; the gray area is the region of interest (ROI) of the next frame;
when the target dimension is 1:
the target is in the non-boundary condition, and the center of the black area is ensured to be the same as the center of the gray area, the height of the gray area is seven times of that of the black area, and the width of the gray area is three times of that of the black area.
When the target dimension is 2:
in the case of a non-boundary target, it is ensured that the center of the black area is the same as the center of the gray area, the height of the gray area is five times that of the black area, and the width of the gray area is three times that of the black area.
When the target dimension is 3:
in the case of a non-boundary target, it is ensured that the center of the black region is the same as the center of the gray region, the height of the gray region is three times that of the black region, and the width of the gray region is three times that of the black region.
When the target dimension is 4:
the size of the target in the image can still be identified under the condition of using down sampling, and meanwhile, the method of continuously using the region of interest cannot ensure real-time performance, so that the horizontal and vertical axis coordinates of the image acquired by the next frame are reduced to one fourth by using the down sampling method, and then the image is identified.
For the case that the target is on the boundary, the region of interest is generated according to the case that the target is not on the boundary, and then the region beyond the boundary of the image is completed in the acquired image, the portion beyond the image, i.e., the short line surrounding region in fig. 12(b), ensures that the height and width of the region of interest are consistent with the case that the target is not on the boundary, and the white region in the image is the supplemented portion as shown in fig. 12 (c). In fig. 12(a), black represents the target region of the previous frame.
For the real-time testing, the average time per frame of operation after successful recognition was 30ms using an Intel Corei5-6400HQ processor without any GPU acceleration.
In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. A high-precision target identification and detection method under a grid background is characterized by comprising the following steps:
step 1, designing a target; the target is a white bottom plate with one corner cut out; the bottom plate is provided with an icon array formed by regularly arranging a plurality of same icons;
step 2, the aircraft carries out imaging identification on the target; in particular, the amount of the solvent to be used,
step 2.1, performing low-pass filtering on the current frame image and then performing binarization processing; the binary segmentation threshold value is changed in a cyclic gradient mode within a set threshold value range, and one threshold value is selected for calculation in each frame; setting a binary segmentation threshold of the initial frame as the maximum value or the minimum value of the threshold range;
step 2.2, expanding and corroding the image after binarization processing;
step 2.3, extracting the outline of the image processed in the step 2.2, and judging whether the extracted outline meets requirements a) to c), if so, judging that the target is identified, and executing the step 3; otherwise, judging that the target is not identified, extracting the next frame of image, executing the step 2.1, and setting a binary segmentation threshold of the next frame according to the gradient change rule of the step 2.1;
wherein a) to c) are required:
a) the extracted outer contour is consistent with the outer contour shape of the designed target baseplate;
b) the extracted outer contour is internally provided with N inner contours, and the ratio of the sum of the areas of all the inner contours to the area of the outer contour is 0.7-1.3 times of the ratio of the sum of the areas of all icons in the design target to the area of the bottom plate;
c) designing inner outlines corresponding to icons positioned on the same straight line on the target and still being collinear;
step 3, updating a binary segmentation threshold value:
converting the target area identified by the current frame into a gray histogram, and then updating a binarization threshold value according to the following formula:
wherein the threshold is an updated binarization threshold; the value range of i is 0-255; p (t) is the number of pixels with the gray value t; sblackIs the sum of the areas of the N inner contours; sallIs the area of the outer contour of the target;
step 4, determining and extracting the region of interest of the next frame image by taking the target position identified by the previous frame as the center, setting the binary segmentation threshold of the next frame as threshold, and identifying the target in the region of interest according to the mode of the step 2;
and (5) repeatedly executing the step 3 to the step 4 until the aircraft successfully hits the net.
2. The method for identifying and detecting a target with high accuracy in a grid background as claimed in claim 1, wherein in step 3, the size of the target dimension K is determined according to the pixel area of the target outline identified in the previous frame, and the region of interest of the image in the next frame is determined according to the value of K:
if S is more than 0 and less than or equal to 4000, K is 1, and the region of interest for extracting the next frame image is: the center of the region of interest is the same as the center of the target region identified in the previous frame, the height of the region of interest is seven times of the height of the target region in the previous frame, and the width of the region of interest is three times of the width of the target region in the previous frame;
if S is more than 4000 and less than or equal to 30000, K is 2, and the region of interest for extracting the next frame image is: the center of the region of interest is the same as the center of the target region identified in the previous frame, the height of the region of interest is five times of the height of the target region in the previous frame, and the width of the region of interest is three times of the width of the target region in the previous frame;
if 30000 < S ≦ 80000, K ≦ 3, and the region of interest for extracting the next frame image is: the center of the region of interest is the same as the center of the target region identified in the previous frame, the height of the region of interest is three times of the height of the target region in the previous frame, and the width of the region of interest is three times of the width of the target region in the previous frame;
if S > 80000, K is 4, the next frame image is down-sampled, and target recognition is performed on the down-sampled image.
3. The method for identifying and detecting the target with high precision under the grid background as claimed in claim 1, wherein after the modified binarization threshold is obtained by calculation using formula (3), the modified threshold and the binarization threshold of the current frame are subjected to smooth filtering to obtain an optimal threshold, and the target is identified by using the optimal threshold according to the method of step 2.
4. The method for identifying and detecting the target with high precision under the grid background as claimed in claim 1, wherein in the step 1, a black border is arranged on the periphery of the bottom plate.
5. The method for detecting target recognition under grid background according to claim 1, wherein in step 2.1, the low-pass filtering is performed by using gaussian filtering.
6. The method for identifying and detecting a target with high precision in a grid background according to claim 2, wherein in the step 2.1, low-pass filtering is performed by using a Gaussian filtering method; after the target is identified, the size of a Gaussian kernel of the next frame of filtering is set according to the target scale; wherein the gaussian kernel size is positively correlated with the target scale.
7. The method for detecting high-precision target recognition in a grid background as claimed in claim 1, wherein in step 2.2, the expanded nuclei are smaller than the eroded nuclei.
8. The method for identifying and detecting the target with high precision in the grid context according to claim 1, wherein in the step 2.3, a Canny edge detection operator is adopted for contour extraction.
9. The method for detecting target recognition under grid background as claimed in claim 1, wherein in step 2.3, the correspondence between the inner contour and the icon on the target is determined by the following method:
s1, respectively calculating the distance between the center of the N inner contours and the center of the outer contour of the target, wherein the inner contour with the shortest distance to the center of the outer contour corresponds to the icon at the center position in the target and is marked as a center icon;
s2, performing polygon fitting on the outer contour of the target, wherein the two nearest vertexes in the pentagon are the unfilled corners; in the N inner outlines, the inner outline with the shortest distance from the center of the inner outline to the two vertexes is the icon closest to the vacancy in the corresponding target and marked as a No. 0 icon;
and S3, turning to a polar coordinate system by taking the center icon as the center and the ray from the center of the center icon to the center of the No. 0 icon as a polar axis, and carrying out bubbling sequencing on polar angles of the centers of the rest inner contours to finish the correspondence between the rest inner contours and the target icons.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910334915.9A CN110210295B (en) | 2019-04-24 | 2019-04-24 | High-precision target identification detection method under grid background |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910334915.9A CN110210295B (en) | 2019-04-24 | 2019-04-24 | High-precision target identification detection method under grid background |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110210295A CN110210295A (en) | 2019-09-06 |
| CN110210295B true CN110210295B (en) | 2021-04-27 |
Family
ID=67786247
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910334915.9A Expired - Fee Related CN110210295B (en) | 2019-04-24 | 2019-04-24 | High-precision target identification detection method under grid background |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110210295B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111488847B (en) * | 2020-04-17 | 2024-02-02 | 上海媒智科技有限公司 | Sports game video ball-feeding segment acquisition system, method and terminal |
| CN112648998A (en) * | 2020-08-06 | 2021-04-13 | 成都道克科技有限公司 | Unmanned aerial vehicle cooperative target autonomous guidance measurement method based on shape and color |
| CN112233076B (en) * | 2020-09-30 | 2022-12-13 | 石家庄铁道大学 | Structural vibration displacement measurement method and device based on red round target image processing |
| CN112907490B (en) * | 2021-01-29 | 2023-01-20 | 合肥工业大学 | Pixel point extraction method, target information acquisition method and push bench control system |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100954494B1 (en) * | 2009-05-25 | 2010-04-22 | 한국항공우주산업 주식회사 | Net recovery system for unmanned aerial vehicle |
| CN105427284A (en) * | 2015-11-06 | 2016-03-23 | 西北工业大学 | Fixed target marking method based on airborne android platform |
| CN105469084A (en) * | 2015-11-20 | 2016-04-06 | 中国科学院苏州生物医学工程技术研究所 | Rapid extraction method and system for target central point |
| CN109229405B (en) * | 2018-10-23 | 2022-03-25 | 西北工业大学 | Air net collision type unmanned aerial vehicle continuous recovery method and system |
-
2019
- 2019-04-24 CN CN201910334915.9A patent/CN110210295B/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN110210295A (en) | 2019-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110210295B (en) | High-precision target identification detection method under grid background | |
| CN110717489B (en) | Method, device and storage medium for identifying text region of OSD (on Screen display) | |
| CN108510467B (en) | SAR image target identification method based on depth deformable convolution neural network | |
| EP3401842B1 (en) | Text image processing method and apparatus | |
| CN104134209B (en) | A kind of feature extracting and matching method and system in vision guided navigation | |
| CN108229342B (en) | Automatic sea surface ship target detection method | |
| CN108446707B (en) | Remote sensing image airplane detection method based on key point screening and DPM confirmation | |
| CN110807473A (en) | Target detection method, device and computer storage medium | |
| CN112699967B (en) | Remote airport target detection method based on improved deep neural network | |
| CN111915583B (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
| US11804025B2 (en) | Methods and systems for identifying topographic features | |
| CN111598952A (en) | Multi-scale cooperative target design and online detection and identification method and system | |
| Li et al. | Road markings extraction based on threshold segmentation | |
| CN109961065B (en) | Sea surface ship target detection method | |
| CN108319961B (en) | Image ROI rapid detection method based on local feature points | |
| JP6110174B2 (en) | Image detection apparatus, control program, and image detection method | |
| CN116543171A (en) | Target detection method and device and electronic equipment | |
| US8891879B2 (en) | Image processing apparatus, image processing method, and program | |
| CN111428624B (en) | Optical remote sensing image on-orbit ship target detection method | |
| Raikar et al. | Automatic building detection from satellite images using internal gray variance and digital surface model | |
| CN110310263B (en) | SAR image residential area detection method based on significance analysis and background prior | |
| CN112907574B (en) | Landing point searching method, device and system of aircraft and storage medium | |
| CN115797374A (en) | Airport runway extraction method based on image processing | |
| CN112418210B (en) | Intelligent classification method for tower inspection information | |
| CN111382632A (en) | Target detection method, terminal device and computer-readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210427 |