A method for evaluating the appearance similarity of manta ray-like robots based on corner point extraction

Abstract

The invention relates to a method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction, wherein the appearance features of the manta ray-like robot and the manta ray are obtained, and the two-dimensional image of the manta ray is preprocessed; the topological structure of the appearance contours of the two is established and the deterministic similarity comparison of the topological structures is performed, and after the requirements are met, the CSS algorithm is used to extract the contour feature points of the manta ray-like robot and the manta ray, and the dynamic time warping algorithm is used to evaluate the similarity of the contour features of the manta ray-like robot and the manta ray. The invention establishes a static feature similarity evaluation method that progresses from topological structure similarity to precise appearance contour similarity. It has guiding significance for the optimization design of the appearance of the manta ray-like robot. By using the CSS corner point extraction method, the corner points have the advantages of not changing with the lighting conditions, being invariant to translation, rotation, and scaling, etc. By matching the time series curvature of the data points, it is unnecessary to screen the feature points when the manta ray-like robot calculates the appearance similarity with the manta ray.

Description

Technical Field

[0001] The invention belongs to the field of bionic underwater robots, and relates to a method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction, and specifically to a multi-viewpoint manta ray-like robot bionic similarity evaluation criterion based on a curvature scale space (CSS) corner point extraction method and a dynamic time warping (DTW) algorithm. Background Art

[0002] After natural selection and long-term evolution, fish have shown excellent swimming ability in water, and these abilities are closely related to their appearance and movement. With the deepening of research on bionic robots, the improvement of the similarity of bionic robots can greatly improve the stability and flexibility of bionic robots. Establishing a similarity evaluation method for the degree of "real" imitation of manta ray robots can provide a basis for the appearance design of bionic robots.

[0003] The process of recognizing an object is often a process of comparing it with the inherent objects in the human brain layer by layer, from coarse to fine, and such a comparison process makes the human brain's judgment of the object extremely uncertain. When judging the similarity between two objects, people usually compare them based on the static characteristics of the objects. Therefore, in order to scientifically and accurately judge the similarity between the manta ray robot and the manta ray, this paper analyzes the static characteristics of the manta ray robot and the manta ray, and proposes a basic method for evaluating the static appearance similarity of the manta ray robot and the manta ray. The progressive relationship between the appearance characteristics is used to judge the similarity between the manta ray robot and the manta ray. Summary of the invention

[0004] Technical issues to be solved

[0005] In order to avoid the shortcomings of the prior art and to improve the bionic similarity of a manta ray-like robot, the present invention proposes a method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction.

[0006] Technical Solution

[0007] A method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction is characterized by the following steps:

[0008] Step 1: Convert the three-dimensional appearance features of the manta ray into three two-dimensional views: front view, side view, and top view;

[0009] Step 2: Preprocess the three 2D images separately to make them consistent with the size of the manta ray robot;

[0010] Step 3: Establish the contour topological structures of all the two-dimensional views of the manta ray robot and all the two-dimensional views of the manta ray, and compare the contour topological structures of the two according to the three types of views: front view, side view, and top view. If the topological structures are similar, perform the similarity calculation in step 4. If not, it is characterized as fuzzy similarity and no similarity calculation is performed.

[0011] The establishment of the topological structure of the outline is as follows: the manta ray-like robot and the manta ray outline are formed into a collection by using geometric elements and sequential connection methods between the geometric elements. Each geometric element and each connection method is assigned a similar element attribute value, and the calculation of the topological similarity between the two is:

[0012]

[0013] Where: G is the number of geometric elements and their connection methods, H is the number of similar elements contained in each similar element, S MR is the topological similarity value between the manta ray and the manta ray-like robot, q ij Indicates similar element attribute values, p ij Represents the similarity coefficient of each similar element, with a value between [0 1]; A is the total number of similar elements, and B is the total number of similar elements;

[0014] When S MR =0 means the manta ray robot is different from the manta ray, or 0<S MR <1 means that the manta ray-like robot is fuzzily similar to the manta ray, and the similarity calculation of the appearance contour is no longer performed;

[0015] When the topological structure similarity S MR =1, it means that the manta ray-like robot is deterministically similar to the manta ray, and if the topological structures are deterministically similar, proceed to step 4;

[0016] Step 4, extracting contour feature points: using the CSS algorithm to extract contour feature points of three pairs of two-dimensional views of the manta ray-like robot and the manta ray, describing the contour as a data point set in a two-dimensional coordinate system, extracting the positive local maximum and negative local minimum points of the curvature on the evolved curve, and obtaining the corner points used to describe the image contour curve, and using the obtained corner points as contour feature points of the manta ray-like robot and the manta ray;

[0017] The two-dimensional view contour feature point set of the manta ray robot is Q = (q 1 ,q 2 ,...q n );

[0018] The two-dimensional view contour feature point set of the manta ray contour is M = (m 1 ,m 2 ,...m n);

[0019] Step 5: Similarity evaluation of the appearance contour: The dynamic time warping algorithm is used to evaluate the similarity of the manta ray robot and the manta ray contour features. The similarity calculation formula is as follows:

[0020]

[0021] Among them, k is the number of feature point sets of the contour;

[0022] Step 6: Multi-view appearance similarity: The appearance feature similarity of the manta ray robot is the weighted sum of the similarity values ​​of the three views. The total similarity value between the manta ray robot and the manta ray is calculated as follows:

[0023] S s =aS 0 +bS 1 +cS 2

[0024] Where: S s Represents the static similarity value, S 0 Represents the top view similarity value, S 1 is the side view similarity value, S 2 The front view similarity value, a, b, c are weights, and the weights satisfy; a+b+c=1.

[0025] The preprocessing of step 2 is to scale and rotate the two-dimensional view using the following formula:

[0026] τ(1)=s -1 FR(Ic)

[0027] Among them, s is the scaling factor, F is the flip matrix, R is the rotation matrix in the PCA transformation method, I is the coordinate of the original model, and c is the coordinate origin.

[0028] The number G of the similar elements is 2.

[0029] The geometric elements include but are not limited to straight lines, arcs, circles or spline curves.

[0030] The sequential connection modes between the geometric elements include, but are not limited to, vertical connection, sharp connection, obtuse connection A with an angle greater than 90° and less than 180°, obtuse connection B with an angle greater than 180° and less than 360°, or tangent connection.

[0031] The weight coefficients of each view are set to be a=b=0.5, c=0.

[0032] The geometric element type and attribute value:

[0033]

[0034] The geometric element connection mode and attribute value:

[0035]

[0036] Beneficial Effects

[0037] The present invention proposes a method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction, which obtains the appearance features of the manta ray-like robot and the manta ray, and pre-processes the two-dimensional image of the manta ray; establishes the appearance contour topological structure of the two and performs deterministic similarity comparison of the topological structure, and after meeting the requirements, uses a CSS algorithm to extract contour feature points of the manta ray-like robot and the manta ray, and uses a dynamic time warping algorithm to evaluate the similarity of contour features of the manta ray-like robot and the manta ray.

[0038] The specific beneficial effects of the present invention are:

[0039] (1) In order to quantitatively judge the similarity between the manta ray-like robot and the manta ray, the present invention establishes a static feature similarity evaluation method that progresses from topological structure similarity to precise similarity of appearance contours. This method has guiding significance for the optimal design of the appearance of the manta ray-like robot.

[0040] (2) The present invention adopts the CSS corner point extraction method for extracting the shape feature points of the manta ray-like robot. The corner points have the advantages of being invariant to changes in lighting conditions, translation, rotation, and scaling.

[0041] (3) Because the appearance feature point sets of manta ray and manta ray-like robot extracted by CSS algorithm have different data volumes, and distance measurement algorithms such as Euclidean distance and Manhattan distance are suitable for calculating two point sets with the same data volume, although Hausdorff distance can calculate point sets with different data volumes, the dissimilarity finally calculated is the maximum distance between the two nearest points in the two point sets, which cannot accurately describe the overall similarity between the manta ray-like robot and the manta ray. The present invention adopts dynamic time warping algorithm (DTW) to evaluate the similarity of the contour features of manta ray-like robot and manta ray. The algorithm matches the time series bending of data points, so that two data point sets with different data point volumes can be calculated. When the manta ray-like robot calculates the similarity with the manta ray, it is no longer necessary to screen the feature points. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of the manta ray-like robot appearance similarity evaluation method;

[0043] Figure 2 This is the three-view drawing of the manta ray-like robot;

[0044] Figure 3 This is a diagram of the manta ray-like robot and the manta ray topology;

[0045] Figure 4 This is a corner diagram of a manta ray from the side. DETAILED DESCRIPTION

[0046] The present invention will now be further described with reference to the embodiments and the accompanying drawings:

[0047] The purpose of the present invention is to provide a method for evaluating the appearance similarity of a manta ray-like robot in order to improve the bionic similarity of the manta ray-like robot. The technical solution of the present invention is as follows:

[0048] Figure 1 The flowchart of the method for evaluating the appearance similarity of a manta ray-like robot of the present invention is shown, and the method comprises the following steps:

[0049] S1: To evaluate the static similarity between the manta ray robot and the manta ray, we first need to obtain the appearance features of the two. The appearance features of the manta ray robot and the manta ray are complex and contain multiple dimensions, but the appearance features of the manta ray can mostly only be extracted through photos, so the three-dimensional appearance features of the manta ray are first converted into three two-dimensional images of the front view, side view, and top view.

[0050] The images of the manta ray robot and the manta ray are collected from the top, side and front views, and the following are obtained: Figure 2 Three views shown.

[0051] S2: When acquiring features of the manta ray robot and the manta ray, the difference in shooting distance and angle can easily affect the similarity evaluation between the two. Therefore, before acquiring features, the present invention performs scaling, rotation, and pre-processing of the three two-dimensional images of the manta ray in S1 to unify the size of the manta ray robot.

[0052] S3: Then use the two-dimensional image processed by S2 to analyze the topological structure of the manta ray robot and the manta ray. Establish a similarity evaluation method for the topological structure of the manta ray robot and the manta ray. The manta ray's outline is a collection of geometric elements in various parts and the connection methods between them. The basic elements of geometric composition include straight lines, arcs, circles, etc. The connection order of each geometric element includes vertical connection, sharp connection, obtuse connection A (angle greater than 90° and less than 180°), obtuse connection B (angle greater than 180° and less than 360°), tangent connection, etc. The specific attribute values ​​are shown in the following table.

[0053] Table 1 Geometric element types and attribute values

[0054]

[0055] Table 2 Geometric element connection methods and attribute values

[0056]

[0057] When calculating the topological structure similarity, it is assumed that the number of similar elements of the manta ray robot and the manta ray outline is G, and the number of similar elements in the present invention is 2, including two elements: geometric elements and geometric element connection methods; the number of similar elements contained in each similar element is H, and the similar elements are basic elements contained in the geometric basic elements and the connection methods, as shown in Tables 1 and 2, and the calculation method of the topological similarity between the two is as follows:

[0058]

[0059] Where: S MR is the topological similarity value between the manta ray and the manta ray-like robot, q ij Represents similar elements, p ij It represents the similarity coefficient of each similar element, and takes values ​​between [0 1]. A is the total number of similar elements, and B is the total number of similar elements.

[0060] If S MR =0 means the manta ray robot is different from the manta ray, 0<S MR <1 means that the manta ray robot is fuzzily similar to the manta ray, and the similarity calculation of the appearance contour is no longer performed. If the topological structure similarity S MR =1, it means that the manta ray-like robot is deterministically similar to the manta ray. If the topological structures are deterministically similar, the similarity calculation in Step 4 is further performed.

[0061] like Figure 3 As shown in the figure, the number of similar elements between the manta ray robot and the manta ray is 2, including basic geometric elements and connection methods; the number of similar elements in the side view is 6, and the number of similar elements in the top view is 15. By substituting formula (2) into the calculation, it can be obtained that the topological similarity value of each view is 1, and the topological structures of the two contours are completely similar, and the accurate similarity calculation of the outer contour can be performed.

[0062] S4: Extraction of contour feature points. The CSS algorithm is used to extract contour feature points of the manta ray robot and the manta ray. The manta ray side view feature point extraction diagram is shown in the figure. Figure 4 shown.

[0063] The CSS algorithm describes the contour of the manta ray robot as a set of data points in a two-dimensional coordinate system, extracts the positive local maximum and negative local minimum points of the curvature on the evolved curve, and obtains the corner points used to describe the image contour curve. The corner points obtained by the CSS algorithm are used as the feature points of the manta ray robot and the manta ray contour for similarity calculation. The feature point sets of the manta ray robot and the manta ray contour are Q = (q 1 ,q 2 ,...q n ), M=(m1 ,m 2 ,...m n ).

[0064] S5: The method for evaluating the similarity of the appearance contour is established. Among the similarity evaluation methods, the common distance measurement methods are Euclidean distance, Manhattan distance, Mahalanobis distance, and Hausdorff distance. Because the appearance feature point sets of manta ray and manta ray-like robot extracted by CSS algorithm have different data volumes, and distance measurement algorithms such as Euclidean distance and Manhattan distance are suitable for calculating two point sets with the same data volume, although Hausdorff distance can calculate point sets with different data volumes, the final calculated dissimilarity is the maximum distance between the two nearest points in the two point sets, which cannot accurately describe the overall similarity between the manta ray-like robot and the manta ray.

[0065] Therefore, the present invention adopts the dynamic time warping algorithm to evaluate the similarity of the manta ray-like robot and the manta ray contour features. The characteristic point set Q of the manta ray-like robot and the manta ray contour obtained by S4 is (q 1 ,q 2 ,...q n ), M=(m 1 ,m 2 ,...m n ), the similarity calculation formula based on the DTW algorithm is as follows:

[0066]

[0067] Among them, k is the number of feature point sets of the contour.

[0068] S6: Establishment of multi-view appearance similarity. The amount of information about the manta ray appearance features contained in each view is different. For the manta ray-like robot and the manta ray, the top view contains the largest and most important information. The amount of information in the side view and the front view is smaller than that in the top view, and they are mainly used to supplement the appearance contour features that cannot be described in the top view. Therefore, in the present invention, the appearance feature similarity of the manta ray-like robot is designed to be the weighted sum of the similarity values ​​of the three views. By assigning different weighted values ​​to each view, the importance of the similarity values ​​of each view is reflected. The total similarity value calculation formula between the manta ray-like robot and the manta ray is:

[0069] S s =aS 0 +bS 1 +cS 2 (4)

[0070] Where: S s Represents the static similarity value, S 0 Represents the top view similarity value, S 1 is the side view similarity value, S 2Front view similarity value, a, b, c are weights, and the weights satisfy;

[0071] a+b+c=1 (5)

[0072] How to reasonably set the weighted values ​​corresponding to the main, side, and top views is very important for calculating the similarity of the manta ray robot. In this study, it was found that the front view of the manta ray robot and the manta ray mainly contains information about the head, the thickness of the trunk and the pectoral fins. This information can be obtained from the top and side views, and the front view of the manta ray is difficult to obtain accurately, so this paper ignores it, so this paper sets the weight coefficients of each view to a=b=0.5 and c=0.

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, reductions, equivalent substitutions, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction, characterized in that Here are the steps: Step 1: Convert the three-dimensional appearance features of the manta ray into three two-dimensional views: front view, side view, and top view; Step 2: Preprocess the three 2D images separately to make them consistent with the size of the manta ray robot; Step 3: Establish the contour topological structures of all the two-dimensional views of the manta ray robot and all the two-dimensional views of the manta ray, and compare the contour topological structures of the two according to the three types of views: front view, side view, and top view. If the topological structures are similar, perform the similarity calculation in step 4. If not, it is characterized as fuzzy similarity and no similarity calculation is performed. The establishment of the topological structure of the outline is as follows: the manta ray-like robot and the manta ray outline are formed into a collection by using geometric elements and sequential connection methods between the geometric elements. Each geometric element and each connection method is assigned a similar element attribute value, and the calculation of the topological similarity between the two is: Where: G is the number of geometric elements and their connection methods, H is the number of similar elements contained in each similar element, S MR is the topological similarity value between the manta ray and the manta ray-like robot, q ij Indicates similar element attribute values, p ij Represents the similarity coefficient of each similar element, with a value between [0 1]; A is the total number of similar elements, and B is the total number of similar elements; When S MR =0 means the manta ray robot is different from the manta ray, or 0<S MR <1 means that the manta ray-like robot is fuzzily similar to the manta ray, and the similarity calculation of the appearance contour is no longer performed; When the topological structure similarity S MR =1, it means that the manta ray-like robot is deterministically similar to the manta ray, and if the topological structures are deterministically similar, proceed to step 4; Step 4, extracting contour feature points: using the CSS algorithm to extract contour feature points of three pairs of two-dimensional views of the manta ray-like robot and the manta ray, describing the contour as a data point set in a two-dimensional coordinate system, extracting the positive local maximum and negative local minimum points of the curvature on the evolved curve, and obtaining the corner points used to describe the image contour curve, and using the obtained corner points as contour feature points of the manta ray-like robot and the manta ray; The two-dimensional view contour feature point set of the manta ray robot is Q = (q1, q2, ...q n ); The two-dimensional view contour feature point set of the manta ray contour is M = (m1, m2, ...m n ); Step 5: Similarity evaluation of the appearance contour: The dynamic time warping algorithm is used to evaluate the similarity of the manta ray robot and the manta ray contour features. The similarity calculation formula is as follows: Among them, k is the number of feature point sets of the contour; Step 6: Multi-view appearance similarity: The appearance feature similarity of the manta ray robot is the weighted sum of the similarity values ​​of the three views. The total similarity value calculation formula between the manta ray robot and the manta ray is: S s =aS0+bS1+cS2 Where: S s represents the static similarity value, S0 represents the top view similarity value, S1 represents the side view similarity value, S2 represents the front view similarity value, a, b, c are weights, and the weights satisfy; a+b+c=1.

2. The method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction according to claim 1, characterized in that: The preprocessing of step 2 is to scale and rotate the two-dimensional view using the following formula: τ(1)=s -1 FR(Ic) Among them, s is the scaling factor, F is the flip matrix, R is the rotation matrix in the PCA transformation method, I is the coordinate of the original model, and c is the coordinate origin.

3. The method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction according to claim 1, characterized in that: The number G of the similar elements is 2.

4. The method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction according to claim 1, characterized in that: The geometric elements include but are not limited to straight lines, arcs, circles or spline curves.

5. The method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction according to claim 1, characterized in that: The sequential connection modes between the geometric elements include, but are not limited to, vertical connection, sharp connection, obtuse connection A with an angle greater than 90° and less than 180°, obtuse connection B with an angle greater than 180° and less than 360°, or tangent connection.

6. The method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction according to claim 1, characterized in that: The weight coefficients of each view are set to be a=b=0.5, c=0.

7. The method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction according to claim 1 or 4, characterized in that: The geometric element type and attribute value:

8. The method for evaluating the appearance similarity of a manta ray-like robot based on corner point extraction according to claim 1 or 5, characterized in that: The geometric element connection mode and attribute value:

Images