Cooperative target, target identification method, system and application

Abstract

The invention provides a planar passive cooperative target with robustness to shielding, and through the design of multiple mark points, the layout of the mark points is optimized, and the recognition capability of the system under the condition of partial shielding is effectively improved. Meanwhile, geometric constraints and coding information are combined, mark points are distinguished by utilizing affine invariance, edge extraction is performed by adopting a self-adaptive gray gradient threshold and a local optimal strategy, mark point edge screening is performed with the assistance of an improved non-maximum suppression method, and stable recognition can still be ensured in complex environments such as illumination variation and light reflection. In addition, through a target identification method based on ideal edge fitting and a sub-pixel-level edge positioning method, accurate positioning of the mark points is realized. According to the design, the robustness and stability of the visual servo system in complex and dynamic environments are remarkably enhanced, and reliable technical support is provided for high-precision tasks of the super-redundant mechanical arm in the fields of industry, spaceflight and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and image processing, and in particular to cooperative targets, target recognition methods, systems, and applications. Background Art

[0002] Compared with non-cooperative targets, cooperative targets are still a more commonly used method in fields such as visual measurement and visual servoing. As a carrier of visual feedback information, the structural form, number of landmarks and the way they are distinguished will directly affect the accuracy of subsequent landmark recognition and positioning, thereby indirectly affecting the effects of visual measurement and visual servoing. There are many ways to classify cooperative targets. Depending on whether they are equipped with additional modules such as power supplies and light-emitting devices, cooperative targets can be divided into active targets and passive targets; and according to the spatial distribution of landmarks, they can be divided into three-dimensional targets and planar targets.

[0003] Active cooperative targets usually have light-emitting devices and are suitable for scenes with a large working distance. Such active targets need to be equipped with additional power modules, which not only causes high energy consumption problems, but also limits the scope of application of the system. Passive targets are also widely used. Japan's ETS-VII satellite collects target images on the target satellite to solve the relative position and pose of the two and guide the capture action. The relative position and pose of the ETS-VII satellite and the target satellite can be obtained by using a target with three markers; while using a target with two markers, other relative position and pose parameters except yaw can be obtained. The docking of my country's Shenzhou 11 spacecraft with Tiangong 2 also uses visual guidance technology. The imager on the Shenzhou 11 spacecraft uses lasers to illuminate the cooperative target installed on Tiangong 2, and uses the target image obtained to guide the docking work. This cross-shaped design is very convenient for manual operation, but it is relatively difficult to automatically identify and locate it.

[0004] Wen Zhuoman and others designed a three-dimensional passive cooperative target for related research. In view of the problem that the graphic boundary of the target is not obvious, an improvement plan was proposed. Subsequently, further improvements were made to address the problem that the reference points on the column are easily confused with the landmark points near the straight line. Tweddle proposed a design plan for a planar passive target. It uses affine invariance to include numerical coding information in the geometric pattern to achieve rapid distinction between different landmark points.

[0005] The current marker design has obvious deficiencies in dealing with occlusion issues. When a marker is partially obscured, the recognition accuracy and robustness of traditional markers will be significantly reduced. In addition, under complex environmental conditions such as lighting changes and reflections, the performance of traditional markers is not stable enough to provide reliable visual feedback. Most markers are relatively simple in design and lack targeted optimization, and cannot provide efficient support in changing actual environments.

[0006] To address these problems, a planar passive cooperative marker that is robust to occlusion is proposed. The marker increases the number of markers and optimizes their layout through a multi-marker design scheme, thereby improving the system's recognition ability under partial occlusion. Combining geometric constraints and encoding information, affine invariance is used to distinguish each marker point, which significantly improves the recognition accuracy and robustness. Edge extraction is performed through adaptive grayscale gradient threshold and local optimal strategy, and the improved non-maximum suppression method is used to screen the edges of marker points, achieving accurate recognition and positioning of marker points. This design significantly improves the robustness and stability of the visual servo system in complex and dynamic environments, and provides strong support for high-precision tasks such as super-redundant robotic arms.

[0007] For example, application number CN201310639611.6 discloses a monocular vision posture measurement method based on point features, which can measure the relative position and posture between two objects and is applicable to complex and changing background and lighting conditions. However, in actual measurement, the overall measurement may be affected due to the occlusion of some landmark points, resulting in reduced accuracy.

[0008] In application number 202211376512.9, a non-planar point cooperative marker and measurement method are disclosed. The invention has the characteristics of rapid recognition and strong anti-interference ability, and can still accurately complete target recognition when individual marker points are lost. However, this method requires special point feature cooperative markers, which may increase the complexity and cost of the system, especially in large-scale applications, the cost of manufacturing and calibrating these special markers is high. Summary of the invention

[0009] The technical problem to be solved by the present invention is how to improve the recognition accuracy and robustness of visual markers in partially occluded and complex environments.

[0010] The present invention solves the above technical problems through the following technical means:

[0011] The cooperative target comprises 4 groups of circular rings and a circular E pattern, wherein the circular ring E has the same diameter as the outer circle of the circular ring; each group has at least 2 circular rings, the circular rings within the group have the same size, and the circular rings between groups have different sizes; the 4 groups of circular rings are radially distributed around the circular ring E, and the centers of the circular rings of two opposite groups are collinear with the center of the circular ring E, and the line connecting the centers of the circular rings of two adjacent groups is perpendicular.

[0012] Furthermore, based on the above cooperative target, the present invention also provides a target recognition method, comprising the following steps:

[0013] The steps of edge extraction are as follows: grayscale the RGB image and calculate the grayscale gradient of each pixel; select the pixel whose gradient grayscale is greater than the threshold as the candidate edge pixel; compare the surrounding pixels of the candidate edge pixel, and if the grayscale gradient meets the set conditions, select the pixel as the anchor point; connect the found anchor points to obtain complete edge information;

[0014] The step of identifying the landmark points: after obtaining the complete ellipse edge based on the edge information, the edges belonging to the landmark points are screened out according to the geometric constraints and numerical coding information set in the cooperative target design scheme; specifically:

[0015] First, the position of the ring is determined according to the center position constraints of the inner and outer rings, and then all the rings are grouped according to the calculated inner and outer circle area ratio of the ring; the straight lines passing through the center points of at least two groups of rings are calculated to obtain the coordinates of the corner points of the two straight lines; the circle E is determined according to the distance between the center of the circle and the coordinate, and the rings in each group of rings are identified based on the circle E. The rings closer and farther from the coordinate represent different positions respectively;

[0016] The steps of locating the landmark point are as follows: find the equation of the line passing through the candidate pixel point and the center of the ellipse, and find the intersection of the line and the ellipse. When there is an intersection point with a distance less than the threshold value, the pixel is regarded as the edge pixel of the landmark point for subsequent processing; otherwise, the pixel is discarded; then the sub-pixel position of the edge pixel is located; and the ellipse is fitted using the sub-pixel position of each edge pixel.

[0017] Furthermore, the step of edge extraction specifically includes:

[0018] Step 1.1 Identification of cooperative targets First, the color RGB image is grayed out and the horizontal and vertical gray gradients of each pixel are calculated using the weighted average method;

[0019] Step 1.2: Count the distribution of the grayscale gradient amplitude in the image, divide the grayscale gradient range into several intervals according to the maximum and minimum gradient amplitudes, and count the pixel ratios in each interval; then, define the pixels with larger gradient amplitudes as edge areas, and define the pixels with smaller gradient amplitudes as smooth areas, and calculate the weights according to the ratio of edge and smooth areas to determine the threshold of the grayscale gradient; finally, discard the pixels with grayscale gradients less than the threshold, and use the pixels with grayscale gradients greater than the threshold as candidate edge pixels for subsequent processing;

[0020] Step 1.3 obtains candidate edge pixel points, and obtains a continuous and smooth single-pixel wide edge by finding anchor points and connecting edges; when the grayscale gradient direction of the candidate edge pixel is horizontal, by comparing it with the left and right pixels, if its grayscale gradient meets a specific condition, that is, the grayscale gradient is greater than a threshold, then the pixel point is selected as the anchor point; when the grayscale gradient direction is vertical, by comparing it with the upper and lower pixels, if it meets the corresponding conditions, it is also selected as the anchor point;

[0021] Step 1.4 When the grayscale gradient direction in the image changes from horizontal to vertical or from vertical to horizontal, it is necessary to specifically analyze the trend of the gradient change and select the appropriate candidate edge pixel direction for connection; by comparing the gradient values ​​in the candidate edge pixel directions, select the point with the largest gradient for connection to ensure the continuity and smoothness of the edge; in addition, define the eight-neighborhood direction of the pixel, and connect the edge through the local optimal strategy. When it is connected to the processed pixel point, the direction search is terminated to finally obtain complete edge information.

[0022] Furthermore, the identification process of the edge of the marker point in the step of identifying the marker point is as follows:

[0023] Step 2.2.1: Based on the center position constraint, filter out all the circles in the image; specifically: when the center coordinates (x′ i ,y′ i ) and (x' j ,y' j ) satisfies the following formula:

[0024]

[0025] This means that it can form a ring, and ε3 is the set value;

[0026] Step 2.2.2 After filtering out the circles in the image, group the circles according to the ratio of the inner and outer ring areas:

[0027] r=S in / S out

[0028] Determine the ring group based on the r value;

[0029] Step 2.2.3: Distinguish the two circles in each group: Let the groups be named A, B, C, and D respectively; calculate the distance between the centers of the two circles in group A (x' A1 ,y' A1 ) and (x' A2 ,y' A2 ) A , and through the centers of the two rings of group B (x' B1 ,y' B1) and (x' B2 ,y' B2 ) B :

[0030]

[0031] Find the straight line l A and l B The intersection coordinates (x0, y0):

[0032]

[0033] in, and The straight lines l A and l B The slope of and are the intercepts of the two straight lines respectively;

[0034] Step 2.2.4 The ellipse whose center is near the intersection coordinates (x0, y0) is circle E. After obtaining the central circle E, using the collinearity of the three collinear points after the projective transformation, the circle in each group of circles that is smaller than E is identified as circle 1, and A1, B1, C1 and D1 are identified; the circle in each group of circles that is larger than E is identified as circle 2, and A2, B2, C2 and D2 are identified.

[0035] Furthermore, in the step of positioning the marker points, the sub-pixel position positioning method of the edge pixels is as follows:

[0036] The gray value I(u0,v0) of the edge pixel is affected by the position and direction of the edge in the pixel. Assume that:

[0037]

[0038] Among them, I A and I B is the gray value of the area on both sides of the edge of the pixel, S A and S B Then it represents the gray level as I A and I B The area occupied by the region, s represents the pixel size;

[0039] Step 3.2.1 For edge pixels with different grayscale gradient directions, different edge location models are used: First, consider G x G y ≥0 and |G x |≤|G y|, it is located in the normal direction of the edge; divide a 5×3 window with the pixel as the center, define the coordinate system xoy, and assume that the equation of the edge curve passing through the pixel is y=ax 2 +bx+c; At the same time, assume that the grayscale of the pixel on the upper side of the edge is I B , the gray value of the lower edge is I A , and define the sum of the grayscale values ​​of the pixels in the left, middle, and right columns of the window as I L ,I M and I R , which can be obtained by the following formula:

[0040]

[0041] Among them, S L , S M and S R Respectively represent the area occupied by the area under the edge in the left, middle and right columns; according to the curve equation of the edge, their expressions can be obtained as follows:

[0042]

[0043] Combining the above formulas, and taking s = 1 to simplify the calculation, we can get the following relationship:

[0044]

[0045] Step 3.2.2 Count the sum of the actual grayscale values ​​of the left, middle and right columns of the window I L ,I M and I R , and then according to the pixel grayscale values ​​of the diagonal areas on both sides of the edge of the window, A and I B The value of is estimated:

[0046]

[0047] At this time, the parameters a, b and c of the edge curve equation can be obtained, and the sub-pixel position of the edge is (u0, v0-c);

[0048] Step 3.2.3 Consider the case where the grayscale gradient is in other directions: when G x G y <0 and |G x |≤|G y |, divide the window into 5×3, and use the grayscale values ​​of the pixels in the lower left and upper right corners of the window to estimate I A and I B :

[0049]

[0050] When G x G y ≥0 and |G x |>|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the lower right and upper left corners of the window to estimate I A and I B :

[0051]

[0052] When G x G y <0 and |G x |>|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the upper right and lower left corners of the window to estimate I A and I B :

[0053]

[0054] Step 3.2.4 After obtaining I A and I B After obtaining the estimated value, the parameters a, b and c of the edge curve equation can be obtained respectively; the sub-pixel level positioning of edge pixels under different grayscale gradient directions can be obtained.

[0055] The present invention also provides a target recognition system, comprising:

[0056] Edge extraction module: grayscale the RGB image and calculate the grayscale gradient of each pixel; select the pixel points whose gradient grayscale is greater than the threshold as candidate edge pixels; compare the candidate edge pixels with the surrounding pixels, and if the grayscale gradient meets the set conditions, select the pixel point as the anchor point; connect the found anchor points to obtain complete edge information;

[0057] Marker point recognition module: after obtaining the complete ellipse edge based on the edge information, the edges belonging to the marker points are screened out according to the geometric constraints and numerical coding information set in the cooperative target design scheme; specifically:

[0058] First, the position of the ring is determined according to the center position constraints of the inner and outer rings, and then all the rings are grouped according to the calculated inner and outer circle area ratio of the ring; the straight lines passing through the center points of at least two groups of rings are calculated to obtain the coordinates of the corner points of the two straight lines; the circle E is determined according to the distance between the center of the circle and the coordinate, and the rings in each group of rings are identified based on the circle E. The rings closer and farther from the coordinate represent different positions respectively;

[0059] Landmark point positioning module: Calculate the equation of the line passing through the candidate pixel point and the center of the ellipse, and find the intersection of the line and the ellipse. When there is an intersection point with a distance less than the threshold value with the pixel point, the pixel is regarded as the edge pixel of the landmark point for subsequent processing; otherwise, the pixel point is discarded; then the sub-pixel position of the edge pixel is located; the ellipse is fitted using the sub-pixel position of each edge pixel.

[0060] Furthermore, the edge extraction module specifically includes:

[0061] Step 1.1 Identification of cooperative targets First, the color RGB image is grayed out and the horizontal and vertical gray gradients of each pixel are calculated using the weighted average method;

[0062] Step 1.2: Count the distribution of the grayscale gradient amplitude in the image, divide the grayscale gradient range into several intervals according to the maximum and minimum gradient amplitudes, and count the pixel ratios in each interval; then, define the pixels with larger gradient amplitudes as edge areas, and define the pixels with smaller gradient amplitudes as smooth areas, and calculate the weights according to the ratio of edge and smooth areas to determine the threshold of the grayscale gradient; finally, discard the pixels with grayscale gradients less than the threshold, and use the pixels with grayscale gradients greater than the threshold as candidate edge pixels for subsequent processing;

[0063] Step 1.3 obtains candidate edge pixel points, and obtains a continuous and smooth single-pixel wide edge by finding anchor points and connecting edges; when the grayscale gradient direction of the candidate edge pixel is horizontal, by comparing it with the left and right pixels, if its grayscale gradient meets a specific condition, that is, the grayscale gradient is greater than a threshold, then the pixel point is selected as the anchor point; when the grayscale gradient direction is vertical, by comparing it with the upper and lower pixels, if it meets the corresponding conditions, it is also selected as the anchor point;

[0064] Step 1.4 When the grayscale gradient direction in the image changes from horizontal to vertical or from vertical to horizontal, it is necessary to specifically analyze the trend of the gradient change and select the appropriate candidate edge pixel direction for connection; by comparing the gradient values ​​in the candidate edge pixel directions, select the point with the largest gradient for connection to ensure the continuity and smoothness of the edge; in addition, define the eight-neighborhood direction of the pixel, and connect the edge through the local optimal strategy. When it is connected to the processed pixel point, the direction search is terminated to finally obtain complete edge information.

[0065] Furthermore, the edge recognition process of the landmark point in the landmark point recognition module is as follows:

[0066] Step 2.2.1: Based on the center position constraint, filter out all the circles in the image; specifically: when the center coordinates (x′ i ,y′ i ) and (x'j ,y' j ) satisfies the following formula:

[0067]

[0068] This means that it can form a ring, and ε3 is the set value;

[0069] Step 2.2.2 After filtering out the circles in the image, group the circles according to the ratio of the inner and outer ring areas:

[0070] r=S in / S out

[0071] Determine the ring group based on the r value;

[0072] Step 2.2.3: Distinguish the two circles in each group: Let the groups be named A, B, C, and D respectively; calculate the distance between the centers of the two circles in group A (x' A1 ,y' A1 ) and (x' A2 ,y' A2 ) A , and through the centers of the two rings of group B (x' B1 ,y' B1 ) and (x' B2 ,y' B2 ) B :

[0073]

[0074] Find the straight line l A and l B The intersection coordinates (x0, y0):

[0075]

[0076] in, and The straight lines l A and l B The slope of and are the intercepts of the two straight lines respectively;

[0077] Step 2.2.4 The ellipse whose center is near the intersection coordinates (x0, y0) is circle E. After obtaining the central circle E, using the collinearity of the three collinear points after the projective transformation, the circle in each group of circles that is smaller than E is identified as circle 1, and A1, B1, C1 and D1 are identified; the circle in each group of circles that is larger than E is identified as circle 2, and A2, B2, C2 and D2 are identified.

[0078] Furthermore, in the landmark positioning module, the sub-pixel position positioning method of the edge pixel is as follows:

[0079] The gray value I(u0,v0) of the edge pixel is affected by the position and direction of the edge in the pixel. Assume that:

[0080]

[0081] Among them, I A and I B is the gray value of the area on both sides of the edge of the pixel, S A and S B Then it represents the gray level as I A and I B The area occupied by the region, s represents the pixel size;

[0082] Step 3.2.1 For edge pixels with different grayscale gradient directions, different edge location models are used: First, consider G x G y ≥0 and |G x |≤|G y |, it is located in the normal direction of the edge; divide a 5×3 window with the pixel as the center, define the coordinate system xoy, and assume that the equation of the edge curve passing through the pixel is y=ax 2 +bx+c; At the same time, assume that the grayscale of the pixel on the upper side of the edge is I B , the gray value of the lower edge is I A , and define the sum of the grayscale values ​​of the pixels in the left, middle, and right columns of the window as I L ,I M and I R , which can be obtained by the following formula:

[0083]

[0084] Among them, S L , S M and S R Respectively represent the area occupied by the area under the edge in the left, middle and right columns; according to the curve equation of the edge, their expressions can be obtained as follows:

[0085]

[0086] Combining the above formulas, and taking s = 1 to simplify the calculation, we can get the following relationship:

[0087]

[0088] Step 3.2.2 The sum of the actual grayscale values ​​of the left, middle and right columns of the statistics window IL ,I M and I R , and then according to the pixel grayscale values ​​of the diagonal areas on both sides of the edge of the window, A and I B The value of is estimated:

[0089]

[0090] At this time, the parameters a, b and c of the edge curve equation can be obtained, and the sub-pixel position of the edge is (u0, v0-c);

[0091] Step 3.2.3 Consider the case where the grayscale gradient is in other directions: when G x G y <0 and |G x |≤|G y |, divide the window into 5×3, and use the grayscale values ​​of the pixels in the lower left and upper right corners of the window to estimate I A and I B :

[0092]

[0093] When G x G y ≥0 and |G x |>|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the lower right and upper left corners of the window to estimate I A and I B :

[0094]

[0095] When G x G y <0 and |G x |>|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the upper right and lower left corners of the window to estimate I A and I B :

[0096]

[0097] Step 3.2.4 After obtaining I A and I B After obtaining the estimated value, the parameters a, b and c of the edge curve equation can be obtained respectively; the sub-pixel level positioning of edge pixels under different grayscale gradient directions can be obtained.

[0098] The invention also provides an application of a target recognition system.

[0099] The advantages of the present invention are:

[0100] The present invention belongs to the field of computer vision and image processing. Specifically, it relates to target detection and recognition in applications such as visual recognition and tracking technology, machine vision, robotics, augmented reality (AR), and autonomous driving. The core of the technology is to design a marker that can be accurately recognized and located by the visual system even when partially blocked, thereby improving the robustness and reliability of the system.

[0101] 1. The present invention ensures that the system can accurately identify the remaining markers even when some markers are blocked by multi-marker design and optimized layout. By increasing the number of markers and optimizing their layout, the robustness and recognition ability of the system in the case of blockage are effectively improved, avoiding large errors in the marker recognition process, and providing a guarantee for subsequent visual servoing operations.

[0102] 2. The present invention combines geometric constraints and coding information, and uses affine invariance to distinguish landmarks. Even in complex environments such as lighting changes and reflections, the system can still maintain high-precision recognition capabilities. Edge extraction is performed through adaptive grayscale gradient thresholds and local optimal strategies, and the improved non-maximum suppression method is used to screen edge points to ensure accurate positioning of landmarks, which significantly enhances the adaptability of the system in complex environments.

[0103] 3. The present invention adopts a sub-pixel edge positioning method to determine the exact position of each edge point by performing sub-pixel positioning on the edge points. Then, the edge points with too large residuals are eliminated by using the iterative fitting method of the ellipse equation to ensure that the fitting error is within a reasonable range, thereby obtaining accurate ellipse parameters, which significantly improves the accuracy of landmark positioning and the robustness of the system.

[0104] 4. The present invention has designed a fault-tolerant mechanism for partial recognition failure. Even if some landmarks are blocked or recognition fails, the system can still accurately locate and identify through other landmarks to ensure the stability and reliability of the system. In dynamic and complex environments, the system can still operate stably and provide reliable visual feedback information, providing strong support for the execution of high-precision tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0105] Figure 1 This is a physical picture of the cooperative target in Example 1 of the present invention;

[0106] Figure 2 This is the cooperative target design diagram in Example 1 of the present invention;

[0107] Figure 3 is the relationship between the pixel grayscale and the sub-pixel edge position in Example 2 of the present invention;

[0108] Figure 4 G in Example 2 of the present invention x G y ≥0,|G x |≤|G y | when the edge positioning model;

[0109] Figure 5 G in Example 2 of the present invention x G y ≥0,|G x |≤|G y |Sub-pixel edge position;

[0110] Figure 6 It is the sub-pixel edge positioning model under different grayscale gradient directions in Example 2 of the present invention;

[0111] Figure 7 are the sub-pixel edge positions in different grayscale gradient directions in Example 2 of the present invention. DETAILED DESCRIPTION

[0112] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0113] Example 1

[0114] like Figure 1 and Figure 2 As shown in the figure, the cooperative target adopts a planar passive design. On the basis of ensuring that it can meet the needs of different systems, it reduces the production cost and processing difficulty of the target as much as possible. The cooperative target uses white as the background color and black as the pattern, which enhances the contrast of the target pattern relative to the background and helps to improve the recognition accuracy. The target pattern is mainly composed of rings and circles, and the center of the circle is used as the mark point. Rings and circles have high recognition in actual working scenes and are not easily confused with the background. In addition, their images in the camera are elliptical. The position of the mark point can be obtained by directly locating the center of the elliptical image.

[0115] The cooperative target has a total of eight ring patterns and one circular pattern. Circle E has the same diameter of 23mm as the outer circles of the other rings, and its center is located at the center of the target. The eight rings are divided into four groups, A, B, C, and D. There are two rings in each group, of which ring 1 is closer to circle E, and ring 2 is farther away. The centers of the two rings in each group are collinear with the center of E, and the centers of the rings in group A and group D are collinear, and the centers of the rings in group B and group C are collinear. The horizontal and vertical distances from the center of E to the centers of A1, B1, C1, and D1 are all 37.5mm, and the horizontal and vertical distances between the centers of the two rings in each group are also 37.5mm. In addition, the center of any ring can form an ordered collinear four points with the centers of another group of two rings that are collinear with it and the center of E, and the intersection ratio between the four points is a constant. In each group of circles, the values ​​of the intersection ratios corresponding to the centers of circles 1 and 2 are different.

[0116] In the design, the two rings in each group have the same inner and outer diameters. The inner diameters of A1 and A2 are 9mm, B1 and B2 are 11mm, C1 and C2 are 14mm, and D1 and D2 are 17.5mm. This design ensures that the two rings in each group have the same inner and outer area ratio, and that the inner and outer area ratios of the rings in different groups are significantly different. It also ensures that the inner and outer edges of each ring can be clearly extracted. The area ratio of different closed areas is an affine invariant. After affine transformation, the inner and outer area ratio of the ring remains constant, which provides numerical coding information for the marker points, making it easier to distinguish the marker points. The length and width of the cooperative target are both 200mm, with a black border on the edge. All patterns on the target are located on the white background area surrounded by the black border, and are at a certain distance from the edge of the target. This design allows the target pattern to be located in a relatively independent area. When the target is mounted on a dark-colored target object or at the end of a robotic arm, a sharp contrast between the foreground and background can be maintained, which is beneficial to improving the accuracy of target identification.

[0117] Example 2

[0118] This embodiment is based on Embodiment 1, and the cooperative target is installed on the surface of the target to be tracked, and the image captured by the camera includes the target to be tracked and the cooperative target. This embodiment describes in detail how to identify the cooperative target, that is, the method of identifying the target to be tracked, which specifically includes the following steps:

[0119] Step 1: Edge extraction

[0120] Step 1.1 Identification of cooperative targets First, the color RGB image is grayed, the gray value is calculated using the weighted average method, and the image is preprocessed using the Gaussian smoothing method to reduce noise. Specifically, a 5×5 convolution kernel with a Gaussian convolution kernel parameter of σ=1 is applied to the image convolution to retain edge information and reduce the influence of electromagnetic noise, shot noise, and thermal noise. Subsequently, the smoothed image is convolved using the Sobel operator to calculate the horizontal and vertical gray gradients of each pixel.

[0121] Step 1.2 counts the distribution of grayscale gradient amplitude in the image, divides the grayscale gradient range into several intervals according to the maximum and minimum gradient amplitudes, and counts the pixel ratio of each interval. Then, define the pixels with larger gradient amplitude as edge areas, and define the pixels with smaller gradient amplitude as smooth areas. Calculate the weights according to the ratio of edge and smooth areas, and determine the threshold of grayscale gradient. Finally, discard the pixels with grayscale gradient less than the threshold, and take the pixels greater than the threshold as candidate edge pixels for subsequent processing.

[0122] After obtaining the candidate rough edge in step 1.3, find the anchor point and connect the edge to obtain a continuous and smooth single-pixel wide edge. When the grayscale gradient direction of the candidate edge pixel is horizontal, by comparing with the left and right pixels, if its grayscale gradient meets certain conditions, that is, the grayscale gradient is greater than the threshold, then the pixel point is selected as the anchor point; when the grayscale gradient direction is vertical, by comparing with the upper and lower pixels, if it meets the corresponding conditions, it is also selected as the anchor point.

[0123] Step 1.4: When the grayscale gradient direction in the image changes from horizontal to vertical or from vertical to horizontal, it is necessary to analyze the trend of the gradient change and select a suitable alternative direction for connection. By comparing the gradient values ​​in the alternative directions, the point with the largest gradient is selected for connection to ensure the continuity and smoothness of the edge. In addition, the eight-neighborhood direction of the pixel is defined, and the edge connection is performed through the local optimal strategy. When the direction is connected to the processed pixel point, the search is terminated, and the complete edge information is finally obtained.

[0124] Step 2: Landmark point recognition

[0125] Step 2.1 By extracting and connecting the edges, the complete elliptical edges in the image can be identified. When the parameters of the edge fitting meet specific constraints, the edge is considered to be a complete elliptical edge. These constraints include the position error between the center of the actual edge and the center of the fitted ellipse, and the relative error between the actual edge length and the perimeter of the fitted ellipse. By constructing a set of equations using more than five edge pixels and finding the least squares solution, the parameters of the ellipse equation can be obtained. These parameters are used to calculate the center coordinates, major and minor semi-axes, and perimeter of the ellipse to further determine the integrity and accuracy of the edge.

[0126] Step 2.2: After identifying the ellipse edge in the image, the edges belonging to the marker points are screened out according to the geometric constraints and numerical coding information set in the cooperative target design scheme. The marker point numbers and geometric constraints are as follows: Figure 2 shown.

[0127] Step 2.2.1: Based on the center position constraint, filter out all the circles in the image. i ',y i ') and (x' j ,y' j ) satisfies the following formula:

[0128]

[0129] This means that it can form a ring, and ε3=3 is generally set.

[0130] Step 2.2.2 After filtering out the circles in the image, group the circles according to step 2.1. Calculate the ratio of the inner and outer circle areas of each group of circles:

[0131] r=S in / S out

[0132] When r∈(0.13,0.18), the ring is A1 or A2; when r∈(0.20,0.24), the ring is B1 or B2; when r∈(0.30,0.39), the ring is C1 or C2; when r∈(0.45,0.6), the ring is D1 or D2.

[0133] Step 2.2.3 distinguish the two circles in each group. Calculate the distance (x') passing through the centers of the two circles in group A. A1 ,y' A1 ) and (x' A2 ,y' A2 ) A , and through the centers of the two rings of group B (x' B1 ,y' B1 ) and (x' B2 ,y' B2 ) B :

[0134]

[0135] Find the straight line l A and l B The intersection coordinates (x0, y0):

[0136]

[0137] in, and The straight lines l A and l B The slope of and are the intercepts of the two lines respectively.

[0138] Step 2.2.4 The ellipse whose center is near the intersection coordinates (x0, y0) is circle E. After obtaining the center circle E, the collinearity of the three collinear points after the projective transformation is used to identify the circle with the smaller distance from E in each group of circles as circle 1, and identify A1, B1, C1 and D1; the circle with the larger distance from E in each group of circles is identified as circle 2, and identify A2, B2, C2 and D2.

[0139] Step 2.2.5: When the cooperative target is partially blocked or part of the ring fails to be recognized, it still does not affect the distinction of other landmarks:

[0140] Since the area ratios of the inner and outer circles are different, the distinction of the circles in each group does not affect each other. Even if some circles fail to be identified, as long as the centers of any two identified circles in groups A and D, and the centers of any two circles in groups B and C are used to construct straight lines and find their intersection points, the center circle E of the target can still be identified.

[0141] When two circles in a group of circles cannot be identified, the two circles in other groups can still be distinguished based on the central circle E;

[0142] When only one circle in a group of circles is successfully identified, as long as the other group of circles collinear with it can be successfully identified, the center of the circle, the center of the central circle E and the two centers of the other group of circles collinear with the circle can be used to form four collinear points in order, which are defined as P1, ... P4 in sequence, and their intersection ratios can be calculated. Based on the property that the cross ratio between the four collinear points is a projective invariant, when the cross ratio value of these four points is in the interval [1.31,1.35], the ring is ring 1, and when the cross ratio value is in the interval [1.48,1.52], the ring is ring 2.

[0143] Step 3: Landmark point positioning

[0144] Step 3.1 Calculate the equation of the line passing through the candidate pixel and the center of the ellipse, and find the intersection of the line and the ellipse. If there is an intersection whose distance with the pixel is less than the threshold, the pixel is regarded as the edge pixel of the marker point for subsequent processing; otherwise, the pixel is discarded.

[0145] Step 3.2 The method based on local area effect assumes that the gray value I(u0,v0) of the edge pixel is affected by the position and direction of the edge in the pixel, and makes the following assumptions:

[0146]

[0147] Among them, I A and I B is the gray value of the area on both sides of the edge of the pixel, S A and S B Then it represents the gray level as I A and I B The area occupied by the region, s represents the pixel size, such as Figure 3 shown.

[0148] Step 3.2.1 For edge pixels with different grayscale gradient directions, different edge location models are used. First, consider G x G y ≥0 and |G x |≤|G y |, the gray gradient direction is as follows Figure 4 As shown, it is located in the normal direction of the edge. A 5×3 window is divided with the pixel as the center, and the coordinate system xoy is defined as shown in the figure, and the equation of the edge curve passing through the pixel is y=ax 2 +bx+c. At the same time, assume that the grayscale of the pixel on the upper side of the edge is I B , the gray value of the lower edge is I A , and define the sum of the grayscale values ​​of the pixels in the left, middle, and right columns of the window as I L ,I M and I R , which can be obtained by the following formula:

[0149]

[0150] Among them, S L , S M and S R Respectively represent the area occupied by the area under the edge in the left, middle and right columns. According to the curve equation of the edge, the expressions can be obtained as follows:

[0151]

[0152] Combining the above formulas, and taking s = 1 to simplify the calculation, we can get the following relationship:

[0153]

[0154] Step 3.2.2 The sum of the actual grayscale values ​​of the left, middle and right columns of the statistics window I L ,I M and I R , and then according to the pixel grayscale values ​​of the diagonal areas on both sides of the edge of the window, A and I B The value of is estimated:

[0155]

[0156] At this point, the parameters a, b and c of the edge curve equation can be obtained. Figure 5 As shown, the sub-pixel position of the edge is (u0, v0-c).

[0157] Step 3.2.3 Figure 6 As shown, consider the case where the gray gradient is in other directions. x G y <0 and |G x |≤|G y |, divide the window into 5×3, and use the grayscale values ​​of the pixels in the lower left and upper right corners of the window to estimate I A and I B :

[0158]

[0159] When G x G y ≥0 and |G x |>|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the lower right and upper left corners of the window to estimate I A and I B :

[0160]

[0161] When G x G y <0 and |G x |>|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the upper right and lower left corners of the window to estimate I A and I B :

[0162]

[0163] Step 3.2.4 After obtaining I A and I B After the estimated value of , the parameters a, b and c of the edge curve equation can be obtained respectively. Under different gray gradient directions, the sub-pixel positioning of the edge is as follows: Figure 7 shown.

[0164] Step 3.3 After determining the sub-pixel position of each edge point, use these points to perform ellipse fitting and preliminarily determine the parameters of the ellipse. In order to ensure the accuracy of the fitting, an optimization method with constraints is used to construct a Lagrangian function and calculate the eigenvalues ​​and eigenvectors to select the most suitable parameter vector as the initial fitting solution. During the fitting process, each edge point is substituted into the ellipse equation, the residual is calculated, and the standard deviation of the residual is counted. All edge points with residuals greater than three times the standard deviation are eliminated and refitted. Through repeated iterations, the fitting results are gradually optimized until all residuals are within a reasonable range.

[0165] This embodiment proposes a planar passive cooperative marker that is robust to occlusion. Through the design of multiple marker points and the optimization of the marker point layout, the recognition ability of the system in partial occlusion is effectively improved. At the same time, in combination with geometric constraints and encoded information, affine invariance is used to distinguish marker points, and adaptive grayscale gradient threshold and local optimal strategy are used for edge extraction, assisted by an improved non-maximum suppression method for marker point edge screening, to ensure stable recognition in complex environments such as lighting changes and reflections. In addition, precise positioning of marker points is achieved through target recognition methods based on ideal edge fitting and sub-pixel edge positioning methods. This design significantly enhances the robustness and stability of the visual servo system in complex and dynamic environments, and provides reliable technical support for high-precision tasks of ultra-redundant robotic arms in industries such as industry and aerospace.

[0166] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cooperative target, characterized in that It includes 4 groups of circular rings and a circular E pattern, wherein the diameter of the circular ring E is the same as that of the outer circle of the circular ring; each group has at least 2 circular rings, the sizes of the circular rings within the group are the same, and the sizes of the circular rings between groups are different; the 4 groups of circular rings are radially distributed around the circular ring E, and the centers of the circular rings of two opposite groups are collinear with the center of the circular ring E, and the line connecting the centers of the circular rings of two adjacent groups is perpendicular.

2. A target identification method, based on the cooperative target according to claim 1; characterized in that: The following steps are involved: The steps of edge extraction are as follows: grayscale the RGB image and calculate the grayscale gradient of each pixel; take the pixel points whose gradient grayscale is greater than the threshold as candidate edge pixel points; compare the surrounding pixels of the candidate edge pixel, and if the grayscale gradient meets the set conditions, the pixel point is selected as the anchor point; By connecting the found anchor points to the edges, you can get the complete edge information; The step of identifying the landmark points: after obtaining the complete ellipse edge based on the edge information, the edges belonging to the landmark points are screened out according to the geometric constraints and numerical coding information set in the cooperative target design scheme; specifically: First, the position of the ring is determined according to the center position constraints of the inner and outer rings, and then all the rings are grouped according to the calculated inner and outer circle area ratio of the ring; the straight lines passing through the center points of at least two groups of rings are calculated to obtain the coordinates of the corner points of the two straight lines; the circle E is determined according to the distance between the center of the circle and the coordinate, and the rings in each group of rings are identified based on the circle E. The rings closer and farther from the coordinate represent different positions respectively; The steps of locating the landmark point are as follows: find the equation of the line passing through the candidate pixel point and the center of the ellipse, and find the intersection of the line and the ellipse. When there is an intersection point with a distance less than the threshold value, the pixel is regarded as the edge pixel of the landmark point for subsequent processing; otherwise, the pixel is discarded; then the sub-pixel position of the edge pixel is located; and the ellipse is fitted using the sub-pixel position of each edge pixel.

3. The target recognition method according to claim 2, characterized in that: The step of edge extraction specifically includes: Step 1.1 Identification of cooperative targets First, the color RGB image is grayed out and the horizontal and vertical gray gradients of each pixel are calculated using the weighted average method; Step 1.2: Count the distribution of the grayscale gradient amplitude in the image, divide the grayscale gradient range into several intervals according to the maximum and minimum gradient amplitudes, and count the pixel ratios in each interval; then, define the pixels with larger gradient amplitudes as edge areas, and define the pixels with smaller gradient amplitudes as smooth areas, and calculate the weights according to the ratio of edge and smooth areas to determine the threshold of the grayscale gradient; finally, discard the pixels with grayscale gradients less than the threshold, and use the pixels with grayscale gradients greater than the threshold as candidate edge pixels for subsequent processing; Step 1.3 obtains candidate edge pixel points, and obtains a continuous and smooth single-pixel wide edge by finding anchor points and connecting edges; when the grayscale gradient direction of the candidate edge pixel is horizontal, by comparing it with the left and right pixels, if its grayscale gradient meets a specific condition, that is, the grayscale gradient is greater than a threshold, then the pixel point is selected as the anchor point; when the grayscale gradient direction is vertical, by comparing it with the upper and lower pixels, if it meets the corresponding conditions, it is also selected as the anchor point; Step 1.4 When the grayscale gradient direction in the image changes from horizontal to vertical or from vertical to horizontal, it is necessary to specifically analyze the trend of the gradient change and select the appropriate candidate edge pixel direction for connection; by comparing the gradient values ​​in the candidate edge pixel directions, select the point with the largest gradient for connection to ensure the continuity and smoothness of the edge; in addition, define the eight-neighborhood direction of the pixel, and connect the edge through the local optimal strategy. When it is connected to the processed pixel point, the direction search is terminated to finally obtain complete edge information.

4. The target recognition method according to claim 2, characterized in that: The identification process of the edge of the marker point in the step of identifying the marker point is as follows: Step 2.2.1: Based on the center position constraint, filter out all the circles in the image; specifically: when the center coordinates (x i ',y i ') and (x' j ,y' j ) satisfies the following formula: This means that it can form a ring, and ε3 is the set value; Step 2.2.2 After filtering out the circles in the image, group the circles according to the ratio of the inner and outer ring areas: r=S in / S out Determine the ring group based on the r value; Step 2.2.3: Distinguish the two circles in each group: Let the groups be named A, B, C, and D respectively; calculate the distance between the centers of the two circles in group A (x' A1 ,y' A1 ) and (x' A2 ,y' A2 ) A , and through the centers of the two rings of group B (x' B1 ,y' B1 ) and (x' B2 ,y' B2 ) B : Find the straight line l A and l B The intersection coordinates (x0, y0): in, and The straight lines l A and l B The slope of and are the intercepts of the two straight lines respectively; Step 2.2.4 The ellipse whose center is near the intersection coordinates (x0, y0) is circle E. After obtaining the central circle E, using the collinearity of the three collinear points after the projective transformation, the circle in each group of circles that is smaller than E is identified as circle 1, and A1, B1, C1 and D1 are identified; the circle in each group of circles that is larger than E is identified as circle 2, and A2, B2, C2 and D2 are identified.

5. The target recognition method according to claim 2, characterized in that: In the step of positioning the marker points, the sub-pixel position positioning method of the edge pixels is as follows: The gray value I(u0,v0) of the edge pixel is affected by the position and direction of the edge in the pixel. Assume that: Among them, I A and I B is the gray value of the area on both sides of the edge of the pixel, S A and S B Then it represents the gray level as I A and I B The area occupied by the region, s represents the pixel size; Step 3.2.1 For edge pixels with different grayscale gradient directions, different edge location models are used: First, consider G x G y ≥0 and | G x | ≤|G y |, it is located in the normal direction of the edge; divide a 5×3 window with the pixel as the center, define the coordinate system xoy, and assume that the equation of the edge curve passing through the pixel is y=ax 2 +bx+c; At the same time, assume that the grayscale of the pixel on the upper side of the edge is I B , the gray value of the lower edge is I A , and define the sum of the grayscale values ​​of the pixels in the left, middle, and right columns of the window as I L ,I M and I R , which can be obtained by the following formula: Among them, S L , S M and S R Respectively represent the area occupied by the area under the edge in the left, middle and right columns; according to the curve equation of the edge, their expressions can be obtained as follows: Combining the above formulas, and taking s = 1 to simplify the calculation, we can get the following relationship: Step 3.2.2 The sum of the actual grayscale values ​​of the left, middle and right columns of the statistics window I L ,I M and I R , and then according to the pixel grayscale values ​​of the diagonal areas on both sides of the edge of the window, A and I B The value of is estimated: At this time, the parameters a, b and c of the edge curve equation can be obtained, and the sub-pixel position of the edge is (u0, v0-c); Step 3.2.3 Consider the case where the grayscale gradient is in other directions: when G x G y <0 and | G x | ≤|G y |, divide the window into 5×3, and use the grayscale values ​​of the pixels in the lower left and upper right corners of the window to estimate I A and I B : When G x G y ≥0 and | G x | >|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the lower right and upper left corners of the window to estimate I A and I B : When G x G y <0 and |G x |>|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the upper right and lower left corners of the window to estimate I A and I B : Step 3.2.4 After obtaining I A and I B After obtaining the estimated value, the parameters a, b and c of the edge curve equation can be obtained respectively; the sub-pixel level positioning of edge pixels under different grayscale gradient directions can be obtained.

6. A target recognition system, based on the cooperative target according to claim 1; characterized in that: include: Edge extraction module: grayscale the RGB image and calculate the grayscale gradient of each pixel; Pixels whose gradient grayscale is greater than the threshold are taken as candidate edge pixels; Compare the candidate edge pixel with the surrounding pixels, and if the grayscale gradient meets the set conditions, the pixel point is selected as the anchor point; By connecting the found anchor points to the edges, you can get the complete edge information; Marker point recognition module: after obtaining the complete ellipse edge based on the edge information, the edges belonging to the marker points are screened out according to the geometric constraints and numerical coding information set in the cooperative target design scheme; specifically: First, the position of the ring is determined according to the center position constraints of the inner and outer rings, and then all the rings are grouped according to the calculated inner and outer circle area ratio of the ring; the straight lines passing through the center points of at least two groups of rings are calculated to obtain the coordinates of the corner points of the two straight lines; the circle E is determined according to the distance between the center of the circle and the coordinate, and the rings in each group of rings are identified based on the circle E. The rings closer and farther from the coordinate represent different positions respectively; Landmark point positioning module: Calculate the equation of the line passing through the candidate pixel point and the center of the ellipse, and find the intersection of the line and the ellipse. When there is an intersection point with a distance less than the threshold value with the pixel point, the pixel is regarded as the edge pixel of the landmark point for subsequent processing; otherwise, the pixel point is discarded; then the sub-pixel position of the edge pixel is located; the ellipse is fitted using the sub-pixel position of each edge pixel.

7. The target recognition system according to claim 6, characterized in that: The edge extraction module specifically includes: Step 1.1 Identification of cooperative targets First, the color RGB image is grayed out and the horizontal and vertical gray gradients of each pixel are calculated using the weighted average method; Step 1.2: Count the distribution of the grayscale gradient amplitude in the image, divide the grayscale gradient range into several intervals according to the maximum and minimum gradient amplitudes, and count the pixel ratios in each interval; then, define the pixels with larger gradient amplitudes as edge areas, and define the pixels with smaller gradient amplitudes as smooth areas, and calculate the weights according to the ratio of edge and smooth areas to determine the threshold of the grayscale gradient; finally, discard the pixels with grayscale gradients less than the threshold, and use the pixels with grayscale gradients greater than the threshold as candidate edge pixels for subsequent processing; Step 1.3 obtains candidate edge pixel points, and obtains a continuous and smooth single-pixel wide edge by finding anchor points and connecting edges; when the grayscale gradient direction of the candidate edge pixel is horizontal, by comparing it with the left and right pixels, if its grayscale gradient meets a specific condition, that is, the grayscale gradient is greater than a threshold, then the pixel point is selected as the anchor point; when the grayscale gradient direction is vertical, by comparing it with the upper and lower pixels, if it meets the corresponding conditions, it is also selected as the anchor point; Step 1.4 When the grayscale gradient direction in the image changes from horizontal to vertical or from vertical to horizontal, it is necessary to specifically analyze the trend of the gradient change and select the appropriate candidate edge pixel direction for connection; by comparing the gradient values ​​in the candidate edge pixel directions, select the point with the largest gradient for connection to ensure the continuity and smoothness of the edge; in addition, define the eight-neighborhood direction of the pixel, and connect the edge through the local optimal strategy. When it is connected to the processed pixel point, the direction search is terminated to finally obtain complete edge information.

8. The target recognition system according to claim 6, characterized in that: The recognition process of the edge of the landmark point in the landmark point recognition module is as follows: Step 2.2.1: Based on the center position constraint, filter out all the circles in the image; specifically: when the center coordinates (x i ',y i ') and (x' j ,y' j ) satisfies the following formula: This means that it can form a ring, and ε3 is the set value; Step 2.2.2 After filtering out the circles in the image, group the circles according to the ratio of the inner and outer ring areas: r=S in / S out Determine the ring group based on the r value; Step 2.2.3: Distinguish the two circles in each group: Let the groups be named A, B, C, and D respectively; calculate the distance between the centers of the two circles in group A (x' A1 ,y' A1 ) and (x' A2 ,y' A2 ) A , and through the centers of the two rings of group B (x' B1 ,y' B1 ) and (x' B2 ,y' B2 ) B : Find the straight line l A and l B The intersection coordinates (x0, y0): in, and The straight lines l A and l B The slope of and are the intercepts of the two straight lines respectively; Step 2.2.4 The ellipse whose center is near the intersection coordinates (x0, y0) is circle E. After obtaining the central circle E, using the collinearity of the three collinear points after the projective transformation, the circle in each group of circles that is smaller than E is identified as circle 1, and A1, B1, C1 and D1 are identified; the circle in each group of circles that is larger than E is identified as circle 2, and A2, B2, C2 and D2 are identified.

9. The target recognition system according to claim 6, characterized in that: In the landmark positioning module, the sub-pixel position positioning method of the edge pixel is as follows: The gray value I(u0,v0) of the edge pixel is affected by the position and direction of the edge in the pixel. Assume that: Among them, I A and I B is the gray value of the area on both sides of the edge of the pixel, S A and S B Then it represents the gray level as I A and I B The area occupied by the region, s represents the pixel size; Step 3.2.1 For edge pixels with different grayscale gradient directions, different edge location models are used: First, consider G x G y ≥0 and | G x | ≤|G y |, it is located in the normal direction of the edge; divide a 5×3 window with the pixel as the center, define the coordinate system xoy, and assume that the equation of the edge curve passing through the pixel is y=ax 2 +bx+c; At the same time, assume that the grayscale of the pixel on the upper side of the edge is I B , the gray value of the lower edge is I A , and define the sum of the grayscale values ​​of the pixels in the left, middle, and right columns of the window as I L ,I M and I R , which can be obtained by the following formula: Among them, S L , S M and S R Respectively represent the area occupied by the area under the edge in the left, middle and right columns; according to the curve equation of the edge, their expressions can be obtained as follows: Combining the above formulas, and taking s = 1 to simplify the calculation, we can get the following relationship: Step 3.2.2 The sum of the actual grayscale values ​​of the left, middle and right columns of the statistics window I L ,I M and I R , and then according to the pixel grayscale values ​​of the diagonal areas on both sides of the edge of the window, A and I B The value of is estimated: At this time, the parameters a, b and c of the edge curve equation can be obtained, and the sub-pixel position of the edge is (u0, v0-c); Step 3.2.3 Consider the case where the grayscale gradient is in other directions: when G x G y <0 and | G x | ≤|G y |, divide the window into 5×3, and use the grayscale values ​​of the pixels in the lower left and upper right corners of the window to estimate I A and I B : When G x G y ≥0 and | G x | >|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the lower right and upper left corners of the window to estimate I A and I B : When G x G y <0 and | G x | >|G y |, divide the window into 3×5, and use the grayscale values ​​of the pixels in the upper right and lower left corners of the window to estimate I A and I B : Step 3.2.4 After obtaining I A and I B After obtaining the estimated value, the parameters a, b and c of the edge curve equation can be obtained respectively; the sub-pixel level positioning of edge pixels under different grayscale gradient directions can be obtained.

10. Application of target recognition system.

Images