CN111292346A - Method for detecting contour of casting box body in noise environment - Google Patents

Abstract

The invention discloses a method for detecting the outline of a casting box body in a noise environment, which belongs to the technical field of image edge detection and comprises the following steps: step 1, inputting a to-be-detected image containing noise; step 2, using bilateral filtering to perform noise reduction processing on the input noise image; step 3, constructing a random structure forest; step 4, using the trained random structure forest to perform preliminary contour detection on the denoised image; step 5, carrying out binarization processing on the preliminary contour detection result; step 6, fitting a pouring gate of the casting box body through Hough circle transformation; and 7, outputting a final detection result image. The method has the main purposes of accurately fitting the circular pouring gate of the casting box body while accurately detecting the linear profile of the casting box body and accurately positioning the circle center of the circular pouring gate.

Description

A method for detecting the contour of a casting box in a noisy environment

technical field

The invention belongs to the technical field of image edge detection, and more particularly relates to a method for detecting the contour of a casting box in a noise environment.

Background technique

In the field of image edge detection technology, the edge information of a noisy image can be accurately detected, a clear edge image of an object can be obtained, and an image edge detection method can be used to provide favorable conditions for subsequent operations. In the past few decades, predecessors have done a lot of work on object edge detection. Traditional image-based categories can be divided into: grayscale image contour detection, RGB-D image contour detection, and color image contour detection. The common contour detection of grayscale images mostly uses the sudden change between the edge gray value and the background gray value in the image. These sudden changes in gray values are called roof or step changes. In the mathematical model, the first and second order can be used. The first-order derivative is used to realize contour detection, among which the Robert operator, Sobel operator, Prewitt operator, Krisch operator, Canny operator, etc. use the first-order derivative to realize contour detection, and the operator using the second-order derivative has the Laplacian operator and the LOG operator.

Color images are richer in chrominance and luminance information than grayscale images. The edge contour of a color image can be regarded as those pixels where the color changes abruptly. The methods of contour detection mainly include two types of methods: color component output fusion method and vector method. The output fusion method is to process the color channels of the color image according to the method of gray image edge detection, and then output the results obtained by each component to obtain the output edge, but this kind of algorithm ignores the correlation between the components. It is easy to lose the edge, and there is no perfect fusion method at present. The vector method treats each pixel in the color image as a three-dimensional vector, so that the entire image is a two-dimensional three-component vector field, which well preserves the vector characteristics of the color image, but is prone to detected edge discontinuities, Missing inspections, etc.

In recent years, some new edge detection algorithms have been proposed. Om Prakash Verma et al. proposed an optimal fuzzy system for color image edge detection based on bacteria search algorithm. Fangfang Han et al. proposed an algorithm for edge detection of high-speed moving target images in noisy environments. Piotr Dollar et al., proposed a fast edge detection method based on structured forest.

Although the above methods have achieved good results in contour detection, because the casting box is in a noisy environment, and the contour of the casting box has a circular gate contour in addition to a straight line, it is necessary to simply use a certain An edge detection algorithm cannot solve the precise detection and positioning of the casting box, and the precise detection and positioning of the casting box is the premise of the precise operation of the casting robot.

SUMMARY OF THE INVENTION

1. The technical problem to be solved by the invention

The purpose of the present invention is to overcome the inaccurate detection of the contour of the casting box in the prior art, and provide a method for detecting the contour of the casting box in a noise environment. The technical scheme of the present invention can meet the requirements of accurate detection of the casting box. The straight outline also accurately fits the circular gate of the box, and accurately locates the center of the circular gate.

2. Technical solutions

In order to achieve the above object, the technical scheme provided by the invention is:

A method for detecting the contour of a casting box in a noise environment of the present invention comprises the following steps:

Step 1. Input the noise-containing image to be detected;

Step 2. Use bilateral filtering to denoise the input noise image;

Step 3. Build a random structure forest;

Step 4. Use the trained random structure forest to perform preliminary contour detection on the denoised image;

Step 5. Binarize the preliminary contour detection result;

Step 6. Fit the gate of the casting box through the Hough circle transformation;

Step 7, output the final detection result image.

As a further improvement of the present invention, the concrete steps of step 2 are as follows:

2a) Use a two-dimensional Gaussian function to generate a distance template, and use a one-dimensional Gaussian function to generate a range template. The generation formula of the distance template coefficient is:

Among them, (k, l) are the center coordinates of the template window, (i, j) are the coordinates of other coefficients of the template window, and σ _d is the standard deviation of the Gaussian function;

2b) The generation formula of the value domain template coefficient is:

where f(x, y) represents the pixel value of the image at point (x, y), (k, l) is the center coordinate of the template window, (i, j) is the coordinate of other coefficients of the template window, σ _r is the standard deviation of the Gaussian function;

2c) Multiply the above two templates to get the template formula of the bilateral filter:

As a further improvement of the present invention, the concrete steps of step 3 are as follows:

3a) Establish a decision tree: First, sample the input image data. Suppose N represents the number of training samples, M represents the number of features, and sampling with replacement is used for N samples, and then column sampling is performed. Select m sub-features (m<<M) from the features, and then recursively classify the sampled data into the left and right sub-trees for each decision tree, until the leaf node, each node _j of the decision tree ft (x) will be is associated with a binary segmentation function:

h(x, θ _j )∈{0,1}

where x is the input vector, {θ _j } is an independent and identically distributed random variable, j represents the jth node in the tree, if h(x, θ _j )=1, then classify x to the left of node j The node, if the side is not, is classified to the right node, and the input element is predicted by the decision tree and the output y is stored in the leaf node, that is, the output distribution is y∈Y;

3b) Use recursive method to train each decision tree: For a training set S _j ∈ X×Y on a given node j, the goal is to find an optimal θ _j through training to make the data set get good classification results, here An information gain criterion needs to be defined:

选择分割参数θ_j的标准是使得信息增益I_j最大，使用数据集在左右节点进行递归训练，当满足下列条件之一停止训练：a)达到设定的最大深度，b)信息增益或训练集尺度都达到门限值，c)落入节点的样本个数少于设定阈值，in:

The criterion for selecting the segmentation parameter θ _j is to maximize the information gain I _j , use the dataset to perform recursive training at the left and right nodes, and stop the training when one of the following conditions is met: a) reach the set maximum depth, b) information gain or training set The scales all reach the threshold, c) the number of samples falling into the node is less than the set threshold,

The information gain formula is defined as follows:

Where: H _entropy (S)=-∑ _y py log(p _y ) represents Shannon information entropy, and py is the probability that the element labeled _y appears in the set s;

3c) Random forest structured output: map all structured labels y∈Y of leaf nodes to discrete label sets c∈C, where C={1,...,k}, and the mapping relationship is defined as follows:

Π: y∈Y→c∈C{1,2,...,k}

维向量，表示分割掩模y的每对像素编码，且对Z进行m维采样，采样后的映射定义为：

Y→Z，再将给定集合Z映射到离散标签集合C，在从Z空间映射到C空间上之前，先采用主成分分析(PCA)将Z的维数降到5维，PCA提取了样本特征中最具有代表性的特征，对于256维空间中的n个样本

降至5维，最后对n个输出标签y₁，...，y_n∈y进行联合形成一个集合模型。The mapping process is divided into two stages. First, the Y space is mapped to the Z space, that is, Y→Z, where the mapping relationship z=Π(y) is defined as

dimensional vector, representing each pair of pixel codes of the segmentation mask y, and m-dimensional sampling of Z, the sampling mapping is defined as:

Y→Z, and then map the given set Z to the discrete label set C. Before mapping from the Z space to the C space, the principal component analysis (PCA) is used to reduce the dimension of Z to 5 dimensions, and PCA extracts the samples The most representative feature among the features, for n samples in a 256-dimensional space

Down to 5 dimensions, the n output labels y ₁ , ..., y _n ∈ y are finally combined to form an ensemble model.

As a further improvement of the present invention, the concrete steps of step 4 are as follows:

4a) Extract the integral channel of the input image: 3 color channels, 1 gradient map and 4 gradient histograms in different directions, a total of 8 channel features, the feature filters of different directions and scales are different in sensitivity to the edge, so it can be Extract LUV color channel, 1 gradient amplitude channel on 2 scales, and 4 directional gradient histogram channels on the image block, a total of 13 channel information, and then obtain from similar features, the obtained feature is the shape of (16× 16, 13) feature matrix;

4b) Define the mapping function Π: y→z, use (y(j)(1≤j≤256)) to represent the jth pixel of the mask y, so that it can be calculated in the case of (j ₁ ≠j ₂ ) Whether the following y(j ₁ )=y(j ₂ ) holds, it can be obtained that a large binary vector mapping function z=Π(y) is defined, and each pair of feature points of j ₁ ≠j ₂ is paired with y( j ₁ )=y(j ₂ ) code;

4c) Obtain the final casting box outline image through edge mapping y′∈Y′.

As a further improvement of the present invention, the concrete steps of step 6 are as follows:

6a) For the input binarized contour image of the casting box, a point in the coordinate space can be mapped to the corresponding trajectory curve or surface in the parameter space. The equation is:

(xa) ² +(yb) ² =r ²

Among them: (a, b) are the coordinates of the center of the circle, and r is the radius of the circle;

6b) Transform the image space equation (xa) ² +(yb) ² =r ² to obtain the parameter space equation:

(ax) ² +(by) ² =r ² ;

6c) Find the position with the most circle intersections in the parameter space. The circle corresponding to this intersection is the circle that passes through all the points in the image space, so as to realize the detection of the circular gate.

3. Beneficial effects

Adopting the technical scheme provided by the present invention, compared with the prior art, has the following remarkable effects:

The accurate detection of the contour of the casting box is the premise and basis for the precise operation of the casting robot. However, because the casting box is in a noisy environment, and the contour of the casting box contains both straight edges and round gates, it is difficult to detect the contours. come difficult. In view of the above technical problems, the present invention proposes a method for detecting the contour of a casting box in a noise environment, which not only meets the requirements for accurate detection of the linear contour of the casting box, but also accurately fits the circular gate of the box, and accurately locates the The center of the round gate.

Description of drawings

Fig. 1 is the flow chart of the present invention;

Fig. 2 is the effect comparison diagram of the present invention to the casting box contour detection effect and traditional method detection:

(a) The true edge map of the image,

(b) Canny algorithm detects the result image without bilateral filtering,

(c) Canny algorithm detects the result image by bilateral filtering,

(d) The Laplacian algorithm detects the resulting image by bilateral filtering,

(e) Random structure forest detection result image after bilateral filtering,

(f) The image of the detection result after the random structure forest detection by bilateral filtering and binarization,

(g) Fitting diagram through Hough circle transformation;

Fig. 3 is the accuracy curve comparison diagram of the present invention and the traditional algorithm to the casting box contour detection result;

FIG. 4 is a comparison diagram of the recall rate curves of the present invention and the traditional algorithm for the detection results of the contour of the casting box.

Detailed ways

In order to further understand the content of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

Example 1

As shown in FIG. 1 , this embodiment provides a method for detecting the contour of a casting box in a noise environment, including the following steps:

Step 1, input the noise-containing image to be detected;

Using phthon3.5 development software in the computer, read out the noise-containing casting box images stored in the computer space in advance.

Step 2: Use bilateral filtering to perform noise reduction processing on the input casting box image. The specific steps are as follows:

2a) Use a two-dimensional Gaussian function to generate a distance template, and use a one-dimensional Gaussian function to generate a range template. The formula for generating the distance template coefficients is:

Among them, (k, l) are the center coordinates of the template window; (i, j) are the coordinates of other coefficients of the template window; σ _d is the standard deviation of the Gaussian function.

2b) The generation formula of the value domain template coefficient is:

Among them, the function f(x, y) represents the image to be processed, f(x, y) represents the pixel value of the image at point (x, y); (k, l) is the center coordinate of the template window; (i, j) is the coordinates of other coefficients of the template window; σ _r is the standard deviation of the Gaussian function.

2c) Multiply the above two templates to get the template formula of the bilateral filter:

Step 3, build a random structure forest, the specific steps are as follows:

3a) Establish a decision tree: First, sample the input image data. Suppose N represents the number of training samples, M represents the number of features, and sampling with replacement is used for N samples, and then column sampling is performed. Select m sub-features (m<<M) in the feature, and then recursively return the sampled data to the left and right sub-trees for each decision tree until the leaf node. Each node j of the decision tree f _t (x) will be associated with a binary partition function:

h(x, θ _j )∈{0,1}

where x is the input vector, {θ _j } is an independent and identically distributed random variable, j represents the jth node in the tree, if h(x, θj)=1, then classify x to the left node of node j , the no side is classified to the right node until the leaf node, and the process ends. The input element is predicted by the decision tree and the output y is stored in the leaf node, that is, the output distribution is y∈Y. The segmentation function h(x, θ _j ) is very complicated, but the general practice is to compare the input x of a single feature dimension with a threshold, when θ = (k, τ) and h(x, θ) = [ x(k)<τ], [·] represents the indicator function; another common method is θ=(k1, k2, τ) and h(x, θ)=[x(k1)-x(k2)<τ ].

3b) Use recursive method to train each decision tree: For a training set S _j ∈ X×Y on a given node j, the goal is to find an optimal θ _j through training so that the data set can get a good classification result. Here we need to define an information gain criterion:

选择分割参数θ_j的标准是使得信息增益I_j最大，使用数据集在左右节点进行递归训练，当满足下列条件之一停止训练：a)达到设定的最大深度；b)信息增益或训练集尺度都达到门限值；c)落入节点的样本个数少于设定阈值。in:

The criterion for selecting the segmentation parameter θ _j is to maximize the information gain I _j , use the dataset to perform recursive training on the left and right nodes, and stop training when one of the following conditions is met: a) reach the set maximum depth; b) information gain or training set The scales all reach the threshold; c) the number of samples falling into the node is less than the set threshold.

The information gain formula is defined as follows:

Where: _Hentropy (S)=-∑ _y p _y log(p _y ) represents the Shannon information entropy, and p _y is the probability of occurrence in the element set s with the label y.

3c) Random forest structured output: The structured output space is generally high-dimensional and complex, so all structured labels y∈Y of leaf nodes can be mapped to a discrete set of labels c∈C, where C={1,. ..,k}, whose mapping relationship is defined as follows:

∏: y∈Y→c∈C{1,2,...,k}

维向量，表示分割掩模y的每对像素编码，对于每个y计算z的代价仍然很高，为了降低维度，对Z进行m维采样，采样后的映射定义为：

Y→Z。在Z的采样过程中加入了随机性保证了树足够多样性。In this paper, the calculation of information gain depends on the measured similarity on Y. However, for the structured output space, it is difficult to calculate the similarity on Y. Therefore, a temporary space mapping from Y to Z is defined, and the distance of this space Z is easier to measure, Finally, the mapping process is divided into two stages. First, the Y space is mapped to the Z space, that is, Y→Z, where the mapping relationship z=∏(y) is defined as

dimensional vector, representing each pair of pixel codes of the segmentation mask y, the cost of calculating z for each y is still high, in order to reduce the dimension, m-dimensional sampling is performed on Z, and the sampled mapping is defined as:

Y→Z. Adding randomness to the sampling process of Z ensures that the tree is sufficiently diverse.

降至5维。有两种方法实现给定集合Z映射到离散标签集合C：a)使用k-means聚类方法，将Z聚集成k个簇；b)基于log₂k维度的PCA量化Z，根据Z落入的象限分配离散标签c。两种方法运行相似，但是后一种速度更快。文中使用选择k＝2的主成分分析量化方法。Before mapping from Z space to C space, principal component analysis (PCA) is used to reduce the dimension of Z to 5 dimensions. PCA extracts the most representative features of the sample features. For n in the 256-dimensional space sample

down to 5 dimensions. There are two ways to achieve the mapping of a given set Z to a set of discrete labels C: a) use the k-means clustering method to cluster Z into k clusters; b) quantify Z based on PCA of log ₂ k dimensions, according to which Z falls into The quadrants of are assigned discrete labels c. Both methods work similarly, but the latter is faster. In this paper, the quantification method of principal component analysis with k=2 is used.

In order to get a unique output result, n output labels y ₁ , ..., y _n ∈ Y need to be combined to form an ensemble model. The m-dimensional sampling mapping function Π _φ can be used to calculate _zi = ∏ _φ (y _i ) for each label i. Centered on z _k = ∏ _φ (y _k ), select those y _k that minimize the sum of the distances of z _k from all other _zi as output labels. The ensemble model depends on m and the chosen mapping function Π _φ .

Step 4. Use the trained random structure forest to perform preliminary contour detection on the denoised cast box. The specific steps are as follows:

4a) Extract the integral channel of the input casting box image: 3 color channels, 1 gradient map and 4 gradient histograms in different directions, a total of 8 channel features, and the feature filters in different directions and scales are sensitive to edges. , so the LUV color channel, 1 gradient amplitude channel on 2 scales and 4 directional gradient histogram channels can be extracted from the image block, a total of 13 channel information, and then obtained from similar features, the obtained feature is the shape of (16×16, 13) feature matrix.

4b) Define the mapping function ∏: y→z. This paper defines a mapping function, using (y(j)( _1≤j≤256 )) to represent the jth pixel of mask y, so that _y (j ₁ )=y(j ₂ ) is true, from this, we can define a large binary vector mapping function z=∏(y), and each pair of feature points of j ₁ ≠j ₂ is paired with y(j ₁ )= y(j ₂ ) encoding.

4c) By fusing the output results of multiple unrelated decision trees, the output of the random structure forest is made more robust. It is very difficult to effectively fuse multiple segmentation masks y∈Y, so this paper adopts the edge map y′∈Y′ to obtain the final casting box contour image.

Step 5: Binarize the contour image detected by the random structure forest, and find the best threshold through repeated experiments. Set the gray value of the pixel point on the image below this threshold to 0, and set it to 255 if it is higher than this threshold, to obtain a binary image that can reflect the overall and local characteristics of the image, so as to segment the image outline from the background. .

Step 6: Fit the gate of the casting box through Hough circle transformation. The specific steps are as follows:

6a) When Hough transform is used for curve detection, the most important thing is to write out the transformation formula from the image coordinate space to the parameter space. For the input binarized contour image of the casting box, a point in the coordinate space can be mapped to the corresponding trajectory curve or surface in the parameter space. For a known circle equation, the general equation for its rectangular coordinates is:

(xa) ² +(yb) ² =r ²

Among them: (a, b) are the coordinates of the center of the circle, and r is the radius of the circle.

6b) Transform the image space equation (xa) ² +(yb) ² =r ² to obtain the parameter space equation:

(ax) ² +(by) ² =r ²

6c) Find the position with the most circle intersections in the parameter space. The circle corresponding to this intersection is the circle that passes through all the points in the image space, so as to realize the detection of the circular gate.

Step 7, output the final casting box contour detection result image.

As shown in Figure 2, each group of pictures b, c, d, e, f, and g compares the detection results from different angles. It is not difficult to see that there are many disordered lines in the four pictures b, c, d, and e. The more lines, the weaker the anti-interference ability of the algorithm to noise, and the fewer the lines, the stronger the anti-interference ability of the algorithm to noise.

From pictures b and c in Figure 2, it can be found that the canny algorithm is not good for box edge detection whether or not it has been denoised. Figure d shows the Laplacian detection effect. It can be seen that the edge detected by the Laplacian algorithm is clearer than that of the canny algorithm, and it has a certain anti-interference ability to noise. Figure e is the edge detection result of the random structure forest algorithm after bilateral filtering. Compared with the first two algorithms, the detection result shows that the noise on the gate surface of the box is basically removed after the detection by the algorithm in this paper, and the surrounding environmental noise is also obtained to a certain extent. It shows that the algorithm in this paper has stronger anti-interference ability to noise than other algorithms. Figure f is the effect of binarizing Figure e. On the basis of Figure f, Figure g is accurately detected and positioned to the gate of the casting box through Hough circle transformation, and the center point of the gate is marked.

Figure 3 compares the accuracy of the edge detection of the casting box gate under different algorithms. It can be seen that the accuracy of the algorithm in this paper is higher than the other two algorithms, and the accuracy of the algorithm for different images is relatively stable. For different angles of the casting box gate All have good detection results. From the recall rate shown in Figure 4, it can be seen that the algorithm in this paper has a higher recall rate, indicating that the detection result is closer to the real edge map.

The present invention and its embodiments have been described above schematically, and the description is not restrictive, and what is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. Therefore, if those of ordinary skill in the art are inspired by it, without departing from the purpose of the present invention, any structural modes and embodiments similar to this technical solution are designed without creativity, which shall belong to the protection scope of the present invention. .

Claims (5)

1. the detection method of casting box profile under a noise environment, is characterized in that, comprises the following steps: Step 1. Input the noise-containing image to be detected; Step 2. Use bilateral filtering to denoise the input noise image; Step 3. Build a random structure forest; Step 4. Use the trained random structure forest to perform preliminary contour detection on the denoised image; Step 5. Binarize the preliminary contour detection result; Step 6. Fit the gate of the casting box through the Hough circle transformation; Step 7, output the final detection result image. 2. The method for detecting the contour of a casting box under a noise environment according to claim 1, wherein the specific steps of step 2 are as follows: 2a) Use a two-dimensional Gaussian function to generate a distance template, and use a one-dimensional Gaussian function to generate a range template. The generation formula of the distance template coefficient is:

Among them, (k, l) are the center coordinates of the template window, (i, j) are the coordinates of other coefficients of the template window, and σ _d is the standard deviation of the Gaussian function; 2b) The generation formula of the value domain template coefficient is:

where f(x, y) represents the pixel value of the image at point (x, y), (k, l) is the center coordinate of the template window, (i, j) is the coordinate of other coefficients of the template window, σ _r is the standard deviation of the Gaussian function; 2c) Multiply the above two templates to get the template formula of the bilateral filter:

3. The method for detecting the contour of a casting box under a noise environment according to claim 1, wherein the specific steps of step 3 are as follows: 3a) Establish a decision tree: First, sample the input image data. Suppose N represents the number of training samples, M represents the number of features, and sampling with replacement is used for N samples, and then column sampling is performed. Select m sub-features (m<<M) from the features, and then recursively classify the sampled data into the left and right sub-trees for each decision tree, until the leaf node, each node _j of the decision tree ft (x) will be is associated with a binary segmentation function: h(x,θ _j )∈{0,1} where x is the input vector, {θ _j } is an independent and identically distributed random variable, j represents the jth node in the tree, if h(x,θ _j )=1, then classify x to the left of node j The node, if the side is not, is classified to the right node, and the input element is predicted by the decision tree and the output y is stored in the leaf node, that is, the output distribution is y∈Y; 3b) Use recursive method to train each decision tree: For a training set S _j ∈ X×Y on a given node j, the goal is to find an optimal θ _j through training to make the data set get good classification results, here An information gain criterion needs to be defined:

in:

The criterion for selecting the segmentation parameter θ _j is to maximize the information gain I _j , use the dataset to perform recursive training at the left and right nodes, and stop the training when one of the following conditions is met: a) reach the set maximum depth, b) information gain or training set The scales all reach the threshold value, c) the number of samples falling into the node is less than the set threshold value, the information gain formula is defined as follows:

Where: H _entropy (S)=-∑ _y p _y log(p _y ) represents the Shannon information entropy, and p _y is the probability that the element labeled y appears in the set s; 3c) Random forest structured output: map all structured labels y∈Y of leaf nodes to discrete label sets c∈C, where C={1,...,k}, and the mapping relationship is defined as follows: Π: y∈Y→c∈C{1,2,...,k} The mapping process is divided into two stages. First, the Y space is mapped to the Z space, that is, Y→Z, where the mapping relationship z=Π(y) is defined as

dimensional vector, representing each pair of pixel codes of the segmentation mask y, and m-dimensional sampling of Z, the sampling mapping is defined as:

Then map the given set Z to the discrete label set C. Before mapping from the Z space to the C space, the principal component analysis (PCA) is used to reduce the dimension of Z to 5 dimensions. PCA extracts the most characteristic features of the sample features. Representative features, for n samples in 256-dimensional space

Down to 5 dimensions, the n output labels y ₁ , . . . , y _n ∈ Y are finally combined to form an ensemble model. 4. The method for detecting the contour of the casting box under a noise environment according to claim 3, wherein the specific steps of step 4 are as follows: 4a) Extract the integral channel of the input image: 3 color channels, 1 gradient map and 4 gradient histograms in different directions, a total of 8 channel features, the feature filters of different directions and scales are different in sensitivity to the edge, so it can be Extract LUV color channel, 1 gradient amplitude channel on 2 scales, and 4 directional gradient histogram channels on the image block, a total of 13 channel information, and then obtain from similar features, the obtained feature is the shape of (16× 16, 13) feature matrix; 4b) Define the mapping function Π: y→z, use (y(j)(1≤j≤256)) to represent the jth pixel of the mask y, so that it can be calculated in the case of (j ₁ ≠j ₂ ) Whether the following y(j ₁ )=y(j ₂ ) holds, it can be obtained that a large binary vector mapping function z=Π(y) is defined, and each pair of feature points of j ₁ ≠j ₂ is paired with y( j ₁ )=y(j ₂ ) code; 4c) Obtain the final casting box outline image through edge mapping y′∈Y′. 5. The method for detecting the outline of a casting box under a noise environment according to claim 1, wherein the specific steps of step 6 are as follows: 6a) For the input binarized contour image of the casting box, a point in the coordinate space can be mapped to the corresponding trajectory curve or surface in the parameter space. The equation is: (xa) ² +(yb) ² =r ² Among them: (a, b) are the coordinates of the center of the circle, and r is the radius of the circle; 6b) Transform the image space equation (xa) ² +(yb) ² =r ² to obtain the parameter space equation: (ax) ² +(by) ² =r ² ; 6c) Find the position with the most circle intersections in the parameter space. The circle corresponding to this intersection is the circle that passes through all the points in the image space, so as to realize the detection of the circular gate.

Images