CN115546754A - Training method of lane line detection model - Google Patents

Abstract

The invention provides a training method for a lane line detection model, and relates to the technical field of neural network model training. In the present invention, the road sample image is first obtained, and the road sample image is marked with real lane lines, and then the road sample image is input into the pre-established initial model, and then the bottom edge of the road sample image is taken as the horizontal axis, and the anchor point is selected on the horizontal axis , and then generate multiple straight anchor lines with different preset angles starting from the anchor point, and generate curved anchor lines tangent to the straight anchor lines, and finally according to the distance between the real lane line and each straight anchor line and the real lane line The distance from each curve anchor line adjusts the parameters of the initial model to train the initial model. The above technical solution not only generates straight anchor lines but also curve anchor lines when training the initial model, which can take into account the detection of lane lines in normal scenes and the detection of lane lines in curve scenes, and improves the performance of curve lines in curve scenes. Accuracy of lane line detection.

Description

A training method for lane line detection model

technical field

The invention relates to the technical field of neural network model training, in particular to a training method for a lane line detection model.

Background technique

At present, the lane line detection method using deep learning is divided into the lane line detection method with anchor (anchor line or anchor box) and the lane line detection method without anchor. In the lane line detection method with anchor, the role of the anchor is to provide a position reference for the prediction of the lane line by presetting some fixed anchor boxes or anchor lines. During the training process of the convolutional neural network model, by calculating the marked The distance between the real lane line and the preset anchor frame or anchor line allows the neural network model to perform parameter learning, and then uses the trained model to predict the lane line. The setting and generation of anchor can greatly affect the effect and efficiency of lane line detection.

In the prior art, the detection accuracy of the lane line detection method in normal scenes can reach 92.14%, but the accuracy rate in curved road scenes is only 67.72%. Part of the reason why it is not good at curve detection is that the anchor lines used in the lane line detection method are all straight lines, which is not conducive to the model to learn the characteristics of curves during the training process.

Contents of the invention

The object of the present invention is to provide a training method for a lane line detection model to solve the technical problem in the prior art that the accuracy of lane line detection in a curved road scene is low.

According to the purpose of the present invention, the present invention provides a kind of training method of lane line detection model, comprises the following steps:

Obtaining a road sample image, where the road sample image is marked with real lane lines;

inputting the road sample image into a pre-established initial model;

Taking the bottom edge of the road sample image as the horizontal axis, selecting an anchor point on the horizontal axis; using the anchor point as a starting point to generate a plurality of straight anchor lines with different preset angles, and generating a line anchor line with the straight line anchor line tangent curve anchor line;

The parameters of the initial model are adjusted according to the distance between the real lane line and each straight anchor line and the distance between the real lane line and each curved anchor line, so as to train the initial model.

Optionally, in the step of generating a plurality of straight anchor lines with different preset angles starting from the anchor point, and generating a curved anchor line tangent to the straight anchor line, all of the curved anchor lines The curvatures are all preset curvatures, and except for the straight anchor line perpendicular to the horizontal axis, any straight anchor line in the other straight line anchor lines only generates one curved anchor line, and the straight line anchor lines perpendicular to the horizontal axis generate two A curved anchor line.

Optionally, the ordinate of the end point of the curved anchor line is less than or equal to the ordinate of the corresponding end point of the straight line anchor line;

The central angle of the curved anchor line is greater than 0° and less than 90°.

Optionally, the step of inputting the road sample image into a pre-established initial model includes the following steps:

Taking the bottom edge of the road sample image as the horizontal axis, selecting a plurality of anchor points arranged at intervals on the horizontal axis; using the plurality of anchor points as starting points to generate a plurality of linear anchors with different preset angles line, and generate curved anchors tangent to said straight anchors.

Optionally, after the step of generating a plurality of straight anchor lines with different preset angles starting from the anchor point, and generating a curved anchor line tangent to the straight anchor line, the following steps are further included:

The functional expression of the curved anchor line is calculated according to the first included angle of the straight anchor line and the radius of the curved anchor line, and the preset angle is the first angle between the straight anchor line and the horizontal axis. an angle.

Optionally, the step of calculating the functional expression of the curved anchor line according to the first included angle of the straight anchor line and the radius of the curved anchor line specifically includes the following steps:

Calculate the second included angle between the starting line of the curved anchor line and the horizontal axis according to the first included angle of the straight anchor line, and the starting line of the curved anchor line is the curved anchor line A straight line between the center of the line and said anchor point;

calculating the coordinates of the center of the curve anchor line according to the second angle and the radius of the curve anchor line;

The function expression of the curve anchor line is determined according to the center coordinates of the curve anchor line.

Optionally, after the step of calculating the functional expression of the curve anchor line according to the center coordinates of the curve anchor line, the following steps are further included:

Selecting multiple dividing lines, the multiple dividing lines are straight lines parallel to the horizontal axis and whose vertical coordinates are not zero;

According to the ordinates of the plurality of dividing lines, the functional expression of the straight anchor line and the functional expression of the curved anchor line, respectively calculate the plurality of dividing lines, the straight anchor line, and the curved anchor line A plurality of abscissas of the intersecting intersection points, the functional expression of the straight line anchor line is calculated according to the first included angle and the coordinates of the anchor point.

Optionally, adjust the parameters of the initial model according to the distance between the real lane line and each straight anchor line and the distance between the real lane line and each curved anchor line, so as to train the initial model The steps specifically include the following steps:

Adjusting the parameters of the initial model according to the multiple abscissas of intersection points of the plurality of dividing lines and the straight anchor line and the curved anchor line and the lateral distance between the lane lines in the lane line training image, The initial model is thus trained.

Optionally, calculate the coordinates of the center of the curve anchor line according to the following formula:

x ₁ =x ₀ +R*cos(π/2-θ);

y ₁ =-R*sin(π/2-θ);

Among them, x1 is the abscissa of the center _of the circle, and _y1 is the ordinate of the center of the circle;

x _s is the abscissa of the anchor point, R is the radius of the anchor line of the curve, θ is the first included angle, and the value of π/2-θ is the second included angle.

Optionally, the function expression of the curve anchor line is: (x−x ₀ ) ² +(yy ₀ ) ² =0.

In the present invention, the road sample image is obtained first, and the road sample image is marked with real lane lines, and then the road sample image is input into the pre-established initial model, and then the bottom edge of the road sample image is taken as the horizontal axis, and the anchor point is selected on the horizontal axis , and then use the anchor point as the starting point to generate multiple straight anchor lines with different preset angles, and generate curved anchor lines tangent to the straight anchor lines, and finally according to the distance between the real lane line and each straight anchor line and the real lane line The distance from each curve anchor line adjusts the parameters of the initial model to train the initial model. The above technical solution not only generates straight anchor lines but also curve anchor lines when training the initial model, which can take into account the detection of lane lines in normal scenes and the detection of lane lines in curve scenes, and improves the performance of curve lines in curve scenes. The accuracy of lane line detection.

Those skilled in the art will be more aware of the above and other objects, advantages and features of the present invention according to the following detailed description of specific embodiments of the present invention in conjunction with the accompanying drawings.

Description of drawings

Hereinafter, some specific embodiments of the present invention will be described in detail by way of illustration and not limitation with reference to the accompanying drawings. The same reference numerals in the drawings designate the same or similar parts or parts. Those skilled in the art will appreciate that the drawings are not necessarily drawn to scale. In the attached picture:

Fig. 1 is a schematic diagram of the anchor line in the lane line detection method in the prior art;

2 is a schematic diagram of an anchor line and a real lane line in a lane line detection method in the prior art;

3 is a schematic diagram of another anchor line and a real lane line in a lane line detection method in the prior art;

4 is a schematic flowchart of a training method for a lane line detection model according to an embodiment of the present invention;

Fig. 5 is a schematic diagram of the anchor line in the training method of the lane line detection model according to an embodiment of the present invention;

6 is a schematic diagram of an anchor line and a real lane line in a training method of a lane line detection model according to an embodiment of the present invention;

Fig. 7 is a schematic diagram of straight anchor lines and curved anchor lines in a method for training a lane line detection model according to an embodiment of the present invention.

detailed description

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar reference numerals designate the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

Fig. 1 is a schematic diagram of an anchor line in a lane line detection method in the prior art, Fig. 2 is a schematic diagram of an anchor line and a real lane line in a lane line detection method in the prior art, and Fig. 3 is a prior art Schematic diagram of another anchor line and real lane line in the middle lane line detection method. As shown in Figure 1, Figure 2 and Figure 3, in the existing anchor line generation method for lane line detection, the anchor line generation process is as follows: input an RGB image, on its left and right sides and bottom Select a certain number of points in equal parts on the side as the starting point of the anchor line, and a certain number of angles θ, and a straight line with an angle θ between the starting point and the horizontal line is used as the straight line anchor line 10'. Among them, the number of starting points on the left is 72, and the values of θ are respectively {θ1, θ2, ..., θ6}={72°, 60°, 49°, 39°, 30°, 22°}, each starting point is Generate 6 anchor lines whose angles are θ1, θ2, ..., θ6. The number of starting points on the right side is 72, the values of θ are respectively {θ1, θ2, ..., θ6}={108°, 120°, 131°, 141°, 150°, 158°}, and the straight anchor line is 10' Generated in the same way as on the left. The number of starting points of the base is 128, and the values of θ are {θ1, θ2, ..., θ15}={165°, 150°, 141°, 131°, 120°, 108°, 100°, 90°, 80 °, 72°, 60°, 49°, 39°, 30°, 15°}, the generation method of the straight anchor line 10' is the same as that of the left and right sides, and the number of generated lines is 15 for each starting point, corresponding to 15 different θ angles. For an RGB image, the total number of generated straight anchor lines 10' is 2*72*6+128*15=2784. The existing lane line detection method can achieve an average accuracy of 76.68% in all scenarios.

The principle of the algorithm is that in the training stage of the convolutional neural network model, an RGB image with the true value of the lane line is input, and the straight anchor line 10' is first generated according to the above method, see Figure 2 and Figure 3, and then the image Divide into 72 equal parts in the longitudinal direction, each dividing line is parallel to the bottom edge, the dividing line intersects with the straight anchor line 10' and the real lane line at several truncation points, the number of truncation points is determined by the actual length of the real lane line, The part of the picture that is higher or lower than the real lane line is not segmented. Since the ordinate of each truncation point is fixed, the difference between the abscissa of the straight anchor line 10' and the true lane line at the truncation point on the same horizontal line is the distance between the two lines in the horizontal section. Then generate the predicted lane line according to the position of the straight anchor line 10', let the network learn, iteratively update the parameters, shorten the distance between the predicted value and the real lane line at the truncation point, and finally use the trained convolutional neural network model to Carry out lane line detection.

Taking a starting point a of the base as an example, the generated straight anchor line 10' is shown in Fig. 1 . 15 straight anchor lines 10' of different angles generated for point a in Fig. 1 .

As shown in Figure 2 and Figure 3, the straight line with the arrow is the straight anchor line 10' generated in the lane line detection algorithm, the curve is the marked real lane line in the curve scene, and Figure 2 and Figure 3 are the anchor lines There are two situations that may exist between the label and the label.

Due to the large arc of the curve lane line, no matter which of the two cases, the distance between the straight anchor line 10' and the real lane line tends to increase at the truncation point, and the distance at the end is relatively large, so The straight anchor line 10' cannot fit the real lane line very well, which makes it more difficult to learn the model parameters of the convolutional neural network during the training process, and the loss function converges very slowly, so in the case of a curve, Lane line detection methods using straight lines as anchor lines have low detection accuracy.

Fig. 4 is a schematic flowchart of a method for training a lane line detection model according to an embodiment of the present invention. As shown in Figure 4, the training method of the lane line detection model includes the following steps:

Step S100, acquiring a road sample image, where the road sample image is marked with real lane lines;

Step S200, inputting the road sample image into the pre-established initial model;

Step S300, taking the bottom edge of the road sample image as the horizontal axis, and selecting an anchor point 50 on the horizontal axis;

Step S400, generating a plurality of straight anchor lines 10 with different preset angles starting from the anchor point 50, and generating a curved anchor line 20 tangent to the straight anchor line 10;

Step S500 , adjust the parameters of the initial model according to the distance between the real lane line and each straight anchor line 10 and the distance between the real lane line and each curved anchor line 20 , so as to train the initial model.

When training the initial model, this embodiment not only generates straight anchor lines 10, but also generates curved anchor lines 20, which can take into account the detection of lane lines in normal scenes and the detection of lane lines in curved scenes, and improves the performance of curved roads. The accuracy of lane line detection in the scene. That is to say, this embodiment can improve the problem that the lane line detection model is difficult to identify the lane line in a curve scene.

Fig. 5 is a schematic diagram of the anchor line in the training method of the lane line detection model according to one embodiment of the present invention, and Fig. 6 is the anchor line and the real lane line 30 in the training method of the lane line detection model according to one embodiment of the present invention schematic diagram of . As shown in FIG. 5 and FIG. 6 , the anchor line generation method proposed in this embodiment not only generates straight lines with different preset angles as the anchor lines, but also generates a curve with a fixed curvature tangent to the straight lines as the anchor lines. In the process of model training, the curve anchor line 20 is better and more quickly fitted to the marked real curve lane line. In Fig. 6, the curve with the arrow on the right side is the curved anchor line 20 with fixed curvature, and the left side is the actual marked curve lane line. It can be seen that since the curved anchor line 20 has a certain curvature, it can better fit the real The distance between the curve anchor line 20 and the real lane line 30 at each truncation point is small, which can better fit the real lane line 30 in the curve scene. In this embodiment, a curve is added in the process of generating the anchor line, so that the lane line detection model can better learn the characteristics of the curve, and the accuracy and efficiency of the lane line detection model in the curve scene are improved.

In this embodiment, in step S400, the curvatures of all the curved anchor lines 20 are preset curvatures, and except for the straight anchor lines 10 perpendicular to the horizontal axis, any straight anchor line 10 in the other straight anchor lines 10 only generates One curved anchor line 20 , the straight anchor line 10 perpendicular to the horizontal axis generates two curved anchor lines 20 . See Figure 5 for details. The preset curvature here is fixed curvature, which is set according to specific design requirements.

In this embodiment, the ordinate of the end point of the curved anchor line 20 is less than or equal to the ordinate of the end point of the corresponding straight anchor line 10 . It can be understood that the starting point of the curved anchor line 20 is the starting point of the previously selected straight anchor line 10 , the curved anchor line 20 extends along the direction of the arrow, and the end point of the curved anchor line 20 will not exceed the point of tangency with the straight anchor line 10 above it. In addition, the central angle of the curved anchor line 20 is greater than 0° and less than 90°. Therefore, except for the straight anchor line 10 perpendicular to the horizontal axis, the remaining straight anchor lines 10 generate only one corresponding curved anchor line 20 .

In this embodiment, after S200, the following steps are included:

Step 1: take the bottom edge of the road sample image as the horizontal axis, and select a plurality of anchor points 50 arranged at intervals on the horizontal axis;

Step 2: Using multiple anchor points 50 as starting points, generate a plurality of straight anchor lines 10 with different preset angles, and generate curved anchor lines 20 tangent to the straight anchor lines 10 .

Specifically, for an RGB image input with a width of w and a height of h, 128 points are equally spaced on its bottom as the starting point of the anchor line, and each point has 9 θ values, namely θ1, θ2, ... ..., θ9, generate 9 straight anchor lines 10, and then use each straight anchor line 10 as a tangent to generate 10 arcs with a curvature h as the curved anchor lines 20, wherein the straight anchor lines 10 perpendicular to the horizontal line are respectively Generate an arc.

The dividing line 40 still takes 72 equally divided points on the vertical axis, makes 72 horizontal lines parallel to the horizontal axis, and intersects the straight anchor line 10, the curved anchor line 20 and the real lane line at several truncation points. The straight line segment parallel to the horizontal line is the horizontal distance between each straight anchor line 10, curved anchor line 20 and the real lane line. It can be seen that the curved anchor line 20 can better fit the curved lane line, see FIG. 6 . Here, the number and coordinates of the starting points of the anchor line, the number and size of θ, the curvature of the curved anchor line 20 and the ordinate of the dividing line 40 are all selected based on experience, and can be adjusted according to the actual situation in practical applications. Choosing a different curvature or θ is also within the scope of the present invention. The curvature of all the curve anchor lines 20 in this embodiment is a single fixed value, and multiple different curvatures can also be selected when generating the curve anchor lines 20 to generate multiple curve anchor lines 20 with different curvatures. The present invention only selects the starting point of the anchor line at the bottom to generate the anchor line, and does not select points on the left and right sides. You can also select a certain number of points on the left and right sides as the starting point of the anchor line.

Fig. 7 is a schematic diagram of a straight anchor line 10 and a curved anchor line 20 in a method for training a lane line detection model according to an embodiment of the present invention. As shown in FIG. 7, in this embodiment, after step S400, the following steps are further included:

Step 3: Calculate the function expression of the curved anchor line 20 according to the first included angle θ of the straight anchor line 10 and the radius R of the curved anchor line 20, and the preset angle is the first included angle between the straight anchor line 10 and the horizontal axis theta.

Specifically, step three specifically includes the following steps:

，曲线锚线20的起始线为曲线锚线20的圆心与锚点50之间的直线；The first step is to calculate the second included angle between the starting line of the curved anchor line 20 and the horizontal axis according to the first included angle θ of the straight anchor line 10

, the starting line of the curved anchor line 20 is a straight line between the center of the curved anchor line 20 and the anchor point 50;

和曲线锚线20的半径R计算曲线锚线20的圆心坐标；The second step, according to the second included angle

Calculate the center coordinates of the curve anchor line 20 with the radius R of the curve anchor line 20;

The third step is to determine the function expression of the curve anchor line 20 according to the center coordinates of the curve anchor line 20 .

In this embodiment, after the third step, the following steps are also included:

The fourth step is to select a plurality of dividing lines 40, and the plurality of dividing lines 40 are straight lines parallel to the horizontal axis and whose vertical coordinate is not zero;

The fifth step, according to the ordinates of multiple dividing lines 40, the functional expression of the straight anchor line 10 and the functional expression of the curved anchor line 20, respectively calculate the intersection of the multiple dividing lines 40 with the straight anchor line 10 and the curved anchor line 20. The multiple abscissas of the intersection point and the functional expression of the straight anchor line 10 are calculated according to the first included angle and the coordinates of the anchor point 50 . Here, the distances between the plurality of dividing lines 40 are equal, and it can be understood that the distances between the vertical coordinates of the plurality of dividing lines 40 are equal.

In this embodiment, step S500 specifically includes the following steps:

Adjust the parameters of the initial model according to the horizontal distance between multiple abscissas of intersection points where the multiple dividing lines 40 intersect with the straight anchor line 10 and the curved anchor line 20 and the lane lines in the lane line training image, so as to train the initial model.

Specifically, the function of the straight anchor line 10 is: y=(xx _s )*tanθ;

Wherein, x _s is the abscissa of the anchor point 50, and θ is the first included angle;

Specifically, the center coordinates of the curved anchor line 20 are calculated according to the following formula:

x ₁ =x ₀ +R*cos(π/2-θ);

y ₁ =-R*sin(π/2-θ);

Among them, x1 is the abscissa of the center _of the circle, and _y1 is the ordinate of the center of the circle;

，第二夹角

是根据圆与切线的性质计算得出的。x _s is the abscissa of the anchor point 50, R is the radius of the curved anchor line 20, θ is the first included angle, and the value of π/2-θ is the second included angle, namely

, the second angle

It is calculated according to the properties of circles and tangents.

According to the mathematical properties, the center coordinates (x ₀ -y ₀ )=[(x _s +R*cos(π/2–θ),-R*sin(π/2 –θ)].

Therefore, the function expression of the curve anchor line 20 is: (x−x ₀ ) ² +(yy ₀ ) ² =0.

Assume that (x _l , y _l ), (x _c , y _c ) are the intersection coordinates of the dividing line 40, the straight line anchor line 10, and the curved anchor line 20, respectively, wherein, y _l = y _c , are respectively the longitudinal Coordinates, x _l and x _c are the abscissas of the two intersection points. Since y _l and y _c are known, according to the function of the straight anchor line 10 and the function of the curved anchor line 20, the abscissas x _l and x _c of the straight anchor line 10 and the curved anchor line 20 at the intersection point can be obtained respectively . Therefore, the distance between the abscissa x _l , x _c and the real lane line 40 can be calculated, so as to adjust the parameters of the initial model.

So far, those skilled in the art should appreciate that, although a number of exemplary embodiments of the present invention have been shown and described in detail herein, without departing from the spirit and scope of the present invention, the disclosed embodiments of the present invention can still be used. Many other variations or modifications consistent with the principles of the invention are directly identified or derived from the content. Accordingly, the scope of the present invention should be understood and deemed to cover all such other variations or modifications.

Claims (10)

1. a training method of lane line detection model, is characterized in that, comprises the following steps: Obtaining a road sample image, where the road sample image is marked with real lane lines; inputting the road sample image into a pre-established initial model; Taking the bottom edge of the road sample image as the horizontal axis, selecting an anchor point on the horizontal axis; generating a plurality of straight anchor lines with different preset angles starting from the anchor point, and generating a curved anchor line tangent to the straight anchor line; The parameters of the initial model are adjusted according to the distance between the real lane line and each straight anchor line and the distance between the real lane line and each curved anchor line, so as to train the initial model. 2. The training method according to claim 1, wherein the anchor point is used as a starting point to generate a plurality of straight anchor lines with different preset angles, and a curved anchor line tangent to the straight anchor line is generated. In the line step, the curvatures of all the curved anchor lines are preset curvatures, and except for the straight anchor lines perpendicular to the horizontal axis, any straight anchor line in the other straight line anchor lines only generates one curved anchor line, A straight anchor line perpendicular to the horizontal axis generates two curved anchor lines. 3. The training method according to claim 2, wherein the ordinate of the end point of the curved anchor line is less than or equal to the ordinate of the corresponding end point of the straight line anchor line; The central angle of the curved anchor line is greater than 0° and less than 90°. 4. The training method according to any one of claims 1-3, characterized in that, the step of inputting the road sample image into a pre-established initial model includes the following steps: Taking the bottom edge of the road sample image as the horizontal axis, selecting a plurality of anchor points arranged at intervals on the horizontal axis; using the plurality of anchor points as starting points to generate a plurality of linear anchors with different preset angles line, and generate curved anchors tangent to said straight anchors. 5. The training method according to any one of claims 1-3, characterized in that, taking the anchor point as a starting point to generate a plurality of straight anchor lines with different preset angles, and generate After the step of cutting the anchor line of the curve, the following steps are also included: The functional expression of the curved anchor line is calculated according to the first included angle of the straight anchor line and the radius of the curved anchor line, and the preset angle is the first angle between the straight anchor line and the horizontal axis. an angle. 6. The training method according to claim 5, characterized in that, the step of calculating the functional expression of the curved anchor line according to the first included angle of the straight line anchor line and the radius of the curved anchor line, Specifically include the following steps: Calculate the second included angle between the starting line of the curved anchor line and the horizontal axis according to the first included angle of the straight anchor line, and the starting line of the curved anchor line is the curved anchor line A straight line between the center of the line and said anchor point; calculating the coordinates of the center of the curve anchor line according to the second angle and the radius of the curve anchor line; The function expression of the curve anchor line is determined according to the center coordinates of the curve anchor line. 7. The training method according to claim 6, characterized in that, after the step of calculating the functional expression of the curve anchor line according to the center coordinates of the curve anchor line, the following steps are also included: Selecting multiple dividing lines, the multiple dividing lines are straight lines parallel to the horizontal axis and whose vertical coordinates are not zero; According to the ordinates of the plurality of dividing lines, the functional expression of the straight anchor line and the functional expression of the curved anchor line, respectively calculate the plurality of dividing lines, the straight anchor line, and the curved anchor line A plurality of abscissas of the intersecting intersection points, the functional expression of the straight line anchor line is calculated according to the first included angle and the coordinates of the anchor point. 8. The training method according to claim 7, characterized in that, according to the distance between the real lane line and each straight anchor line and the distance between the real lane line and each curved anchor line, the initial The parameters of the model are adjusted to train the steps of the initial model, which specifically includes the following steps: Adjusting the parameters of the initial model according to the multiple abscissas of intersection points of the plurality of dividing lines and the straight anchor line and the curved anchor line and the lateral distance between the lane lines in the lane line training image, The initial model is thus trained. 9. training method according to claim 6, is characterized in that, calculates the circle center coordinate of described curve anchor line according to following formula: x ₁ =x ₀ +R*cos(π/2-θ); y ₁ =-R*sin(π/2-θ); Among them, x1 is the abscissa of the center _of the circle, and _y1 is the ordinate of the center of the circle; x _s is the abscissa of the anchor point, R is the radius of the anchor line of the curve, θ is the first included angle, and the value of π/2-θ is the second included angle. 10. The training method according to claim 6, characterized in that, the functional expression of the curve anchor line is: (x−x ₀ ) ² +(yy ₀ ) ² =0.

Images