CN114926391B - Perspective transformation method based on improved RANSAC algorithm - Google Patents

Abstract

The perspective transformation method based on the improved RANSAC algorithm utilizes the SIFT algorithm to obtain the coordinates of the characteristic matching points of the reference graph and the auxiliary graph, so that a transformation matrix can be obtained by randomly selecting 4 pairs of characteristic matching points, and the perspective transformation can be carried out on the auxiliary graph by utilizing the transformation matrix. Only 4 pairs of matching points are needed in calculating the transformation matrix, and an improved RANSAC algorithm is provided. The invention can help better assist the intelligent inspection robot to identify the running state of the pressing plate in the image and improve the anti-interference capability of the intelligent inspection robot.

Description

Perspective transformation method based on improved RANSAC algorithm

Technical Field

The present invention relates to the technical field of intelligent power grid inspection, and in particular to a perspective transformation method based on an improved RANSAC algorithm.

Background Art

With the continuous development of communication technology and artificial intelligence technology, the rapid construction of smart grids has been driven. Intelligent inspection has gradually become an important auxiliary operation and maintenance means in the current unmanned mode of substations. As an important protection device in the power operation system, the relay secondary equipment protection pressure plate has a large number of relays, and the traditional manual inspection workload is large and complicated. False detection or missed detection often occurs. Therefore, the use of intelligent robots for inspection has also become a research hotspot. At present, the application of image processing related technologies in power systems is mainly concentrated in primary equipment. In actual production, the operation status switching of secondary equipment is not easy to find, and it needs to be paid attention to and monitored in real time.

By observing the collected pressure plate images, we can see that since most of the protective cabinet doors use glass cabinet doors, there is local highlight interference in the pressure plate image under uneven lighting conditions. In severe cases, the highlight interference will cause the texture information to be seriously missing and unrecognizable, resulting in false detection and missed detection of the pressure plate status.

There are three main reasons for the reflection of glass cabinet doors in substations:

(1): The reflection formed by natural light shining through the window onto the glass cabinet door;

(2) Reflections from interior lighting on glass cabinet doors;

(3): The reflection of the camera flash on the cabinet door when the light is dim.

Highlight area detection based on two-dimensional OTSU algorithm:

Among various algorithms for image threshold segmentation, the maximum between-class variance method (OTSU) is widely used because of its simple calculation and stable performance. The traditional OTSU algorithm uses a bimodal histogram to count the average grayscale value of the pixels in the image and divide the image into two categories: the foreground area and the background area. The threshold is obtained when the variance of the two categories is the largest. Since the pixels in the same area of an image have strong consistency and correlation in position and grayscale, the traditional OTSU algorithm only considers the grayscale information provided by the histogram and ignores the spatial position information of the image.

Summary of the invention

The present invention proposes a protection pressure plate operating status identification method based on image fusion, which is used in the intelligent inspection of secondary relay protection pressure plates in substations. The method adopts a threshold segmentation method to detect the highlight area, and performs image restoration on this basis, so as to effectively remove the light and shadow interference in the relay protection pressure plate image, so as to better assist the intelligent inspection robot to identify the operating status of the pressure plate in the image and improve its anti-interference ability.

The technical solution adopted by the present invention is:

The method for identifying the operating state of a protective pressure plate based on image fusion comprises the following steps:

Step 1: Protect the platen image highlight area detection:

Step 1.1: The intelligent robot inspects the relay protection pressure plate and takes pictures of the protection pressure plate;

Step 1.2: Since the entire area of the protective pressure plate must be captured during the shooting, the captured protective pressure plate images are input into the computer image processing system, and the protective pressure plate images are screened, and the protective pressure plate images that do not capture the entire area are deleted to avoid affecting the subsequent test results. The screened protective pressure plate images are compressed to allow the computer to process the images faster.

The computer image processing system is implemented by configuring OpenCV, python and other plug-ins in the Microsoft Visual Studio development platform.

Step 1.3: grayscale the protective plate image to enlarge the difference between the highlight area and the background area;

Step 1.4: Use the two-dimensional OTSU algorithm to find the optimal threshold, segment the grayscale image of the protective pressure plate according to the optimal threshold, and quickly detect the highlight area in the image.

Step 2: Highlight area removal based on image fusion:

Step 2.1, feature point detection:

The SIFT algorithm is used to detect feature points of the reference image and the auxiliary image, generate feature description vectors, and the nearest neighbor method is used to match the features of the reference image and the auxiliary image.

Step 2.2: Perspective transformation based on improved RANSAC algorithm:

The perspective transformation matrix is obtained through the obtained feature matching points, and the perspective transformation is used to adjust the viewing angle and size of the auxiliary image to make it consistent with the reference image.

Step 2.3, Image restoration:

According to the position of the highlight area detected in the reference image, the corresponding area position is filled and repaired with the auxiliary image after perspective transformation, thereby removing the highlight in the reference image.

Step 3: Identify the protection plate status:

Step 3.1: Connected region extraction:

The reference image after highlight removal is subjected to color region screening, binarization, and morphological processing, and the connected regions are extracted using the 8-connectivity method.

Step 2: Effective platen area screening:

The area, size and shape are analyzed according to the morphological characteristics, and the effective pressure plate area is accurately extracted from the connected area.

Step 3: Identification of the platen insertion and retraction status:

The screened effective pressure plate area is identified. After the effective pressure plate insertion and retraction status is recognized, the effective pressure plates are sorted from left to right and from top to bottom using the center of gravity coordinates, and finally a state sequence containing only 0 and 1 is obtained.

Step 2.2 The specific steps include the following steps:

Step 1: Use the nearest neighbor method to perform initial matching on the effective feature points extracted by the SIFT algorithm, and the initially selected Euclidean distance threshold is 0.6. Divide the image into 4 equal areas, and determine whether the number of feature matching points in the current 4 areas is greater than 4. If so, proceed to the next step, otherwise add 0.1 to the Euclidean distance threshold and rematch.

Step 2: Select 4 pairs of matching points with the smallest Euclidean distance from each of the 4 regions, for a total of 16 pairs. Combine these 16 pairs of matching points into groups of 4, and sort them from small to large according to the sum of the Euclidean distances of the 4 pairs of matching points after combination, 1, 2, ..., N, and select the top 50 groups.

Step 3: First, take 4 pairs of matching points with serial number 1 in order to calculate the transformation matrix H, and use the matrix H to check all the matching point pairs in the image to determine whether the proportion of the points in the local area to the total number of matching point pairs is greater than 50%. If it is greater than 50%, the currently calculated matrix H is the optimal transformation matrix, otherwise select the next group of 4 pairs of matching points in order to calculate the transformation matrix H.

The present invention provides a method for identifying the operating status of a protective pressure plate based on image fusion, and the technical effects are as follows:

1) It can be widely used in the intelligent inspection of relay protection pressure plates in substations, better assist the inspection robot in checking the status of the pressure plates, improve the accuracy of identifying the status of the relay protection pressure plates, reduce the labor intensity of inspection personnel, reduce misoperation in power grid operations, avoid economic losses, and ensure the safe and stable operation of the power grid.

2) The main function of the present invention is to better assist the intelligent inspection robot in checking the status of the pressure plate, improve the accuracy of identifying the status of the relay protection pressure plate, reduce the labor intensity of inspection personnel, reduce misoperation in power grid operation, avoid economic losses, and ensure the safe and stable operation of the power grid.

3) The computer image processing technology of the present invention detects the highlight area of the reference image mainly including: graying the collected reference image, using the improved OTSU two-dimensional threshold segmentation method to quickly detect the highlight interference in the image, reducing the influence of noise, and making the detection of the highlight area in the reference image more accurate.

4) The present invention uses the SIFT algorithm to detect feature points of the reference image and the auxiliary image and uses the nearest neighbor method for matching. The improved RANSAC algorithm is introduced to remove erroneous matching points and obtain the best perspective transformation matrix. The auxiliary image is transformed into the main image to repair the highlight area. On the basis of image restoration, the inclination angle of the edge of the pressure plate is detected to judge its operating status. The present invention can help better assist the intelligent inspection robot to identify the operating status of the pressure plate in the image and improve its anti-interference ability.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a two-dimensional histogram.

FIG. 2 is a diagram of the perspective transformation process.

FIG3 is a flowchart of the detection of highlight areas of the protection platen image.

FIG4 is a flowchart of highlight removal based on image fusion.

FIG5 is a diagram showing the perspective transformation steps based on the improved RANSAC algorithm.

FIG. 6 is a flow chart of the protection pressure plate state identification.

FIG. 7 is a diagram showing the overall structure of the method of the present invention.

DETAILED DESCRIPTION

The protection platen operating state identification method based on image fusion includes:

1. Highlight area detection based on OTSU optimization algorithm:

1. Platen image feature analysis:

The present invention adopts a threshold segmentation method to detect the highlight area, and performs image restoration on this basis, laying a foundation for identifying the operating status of the pressing plate.

2. Highlight area detection based on two-dimensional OTSU algorithm:

In order to better segment the foreground and background and improve the algorithm's anti-noise ability, the present invention increases the dimension of the traditional one-dimensional OTSU algorithm to two dimensions. The specific steps are as follows:

Step 1: Suppose there is an image I, and the grayscale of image I(x,y) is L, then the average grayscale of the field of image I is also L.

Step 2: Let f(x,y) be the gray value of the pixel (x,y), and g(x,y) be the average gray value in the K×K area of the center pixel (x,y). Let f(x,y)=i, g(x,y)=j, and then a tuple (i,j) is formed.

Step 3: Let the number of occurrences of the binary group (i, j) be _fij , and find the probability density _Pij corresponding to the binary group, _Pij = _fij /N,i,j = 1,2,…,L, where N is the total number of image pixels.

Step 4: Randomly select a threshold vector (s, t) to divide the two-dimensional histogram of the image into four regions. Regions B and C represent the foreground and background of the image, and regions A and D represent noise points, as shown in Figure 1.

Step 5: Assume that the probabilities of background and foreground appearing are ω ₁ and ω ₂ respectively, and the corresponding mean vectors are μ ₁ and μ ₂ . The mean vector corresponding to the entire image is μ, and the formula is as follows:

Where: ω ₁ is the probability of background occurrence, _{and Pij} is the probability density of the occurrence of the binary group (i, j).

Where: ω ₂ is the probability of foreground appearance, _Pij is the probability density of the occurrence of the binary group (i, j).

Where: μ ₁ is the mean vector corresponding to the background.

Where: μ ₂ is the mean vector corresponding to the foreground.

Where: μ is the mean vector corresponding to the entire image.

Step 6: Use the discrete measure matrix S _{(s, t)} to obtain the discrete measure tr(S _{(s, t)} ) of the image. The formula is as follows:

S _(s,t) =ω ₁ (μ ₁ -μ)(μ ₁ -μ) ^T +ω ₂ (μ ₂ -μ)(μ ₂ -μ) ^T (6)

Where: S _{(s, t)} is the discrete measurement matrix of the image.

tr(S _(s,t) )=ω ₁ [(μ _1i -μ _i ) ² +(μ _1j -μ _j ) ² ]+ω ₂ [(μ _2i -μ _i ) ² +(μ _2j -μ _j ) ² ] (7)

Where: tr(S _(s,t) ) is the discrete measure of the image.

Step 7: The larger the discrete measure, the larger the inter-class variance. The maximum discrete measure corresponds to the optimal threshold (s ^* , t ^* ).

(s ^* ,t ^* )＝argmax{tr(S _(s,t) )} (8)

Where: (s ^* , t ^* ) is the optimal threshold of the image.

After obtaining the optimal threshold through the above steps, the threshold is used to perform binarization processing on the grayscale image with brightness levels of 0 to 255 to separate the foreground area and the background area. At this time, the foreground area is the highlight area.

2. Highlight area removal based on image fusion:

Image fusion is the process of combining two or more images of the same target into a new image, effectively improving the utilization of image information so that the fused image has a more comprehensive and clear description of the target.

The present invention aims at the problem that after the feature description vector is generated by the SIFT algorithm, the nearest neighbor method is used for feature matching, which is overly dependent on the preset threshold and has wrong matching point pairs. The RANSAC algorithm is introduced to further eliminate the wrong matching point pairs, complete the multi-view image feature matching and obtain the best perspective transformation matrix. At the same time, in order to avoid the drawbacks of the traditional RANSAC algorithm of random selection and reduce unnecessary iteration times and time, the present invention improves the RANSAC algorithm, so that the effect of image perspective transformation and highlight removal is better.

1. Feature point detection:

At present, the most classic algorithm for feature point description is SIFT (Scale-invariant feature transform), which is widely used for feature point detection and generating feature description vectors because of its ability to remain unchanged under rotation, scale scaling and brightness changes. The specific steps are as follows:

Step 1: Take the input image I(x,y) and continuously downsample it to obtain a series of images of different sizes. These images are sorted from large to small and from bottom to top to form a pyramid model. Then, for each layer of image, a two-dimensional Gaussian function G(x,y,σ) with continuously changing scale is convolved with the image I(x,y) to obtain the scale space L(x,y,σ) of the image.

L(x,y,σ)＝G(x,y,σ)*I(x,y) (9)

Where: * represents the convolution operation, σ is the scale.

Step 2: Let multiple images of each layer in the scale space be called a group, and subtract two adjacent layers of images in the same group to obtain a Gaussian difference image. Compare each pixel of the Gaussian difference image of each layer except the top and bottom layers in the same group with 8 pixels in the same layer and 9×2 pixels in the upper and lower adjacent layers, a total of 26 pixels. When the pixel value of the point is the maximum or minimum, the pixel is an extreme point. The Gaussian difference function formula is as follows:

D(x,y,σ)=L(x,y,kσ)-L(x,y,σ) (11)

Among them, k is a fixed coefficient.

Step 3: Since the detected extreme points are the extreme points of the discrete space, not the real feature points of the continuous space, it is necessary to perform curve fitting on the scale space Gaussian difference function to recalculate the coordinates of the extreme points, that is, to expand the scale space Gaussian difference function using the Taylor formula:

Where D(X) is the Gaussian difference function, X = (x, y, σ) ^T .

Taking the derivative and setting the equation equal to zero gives the offset of the extreme point:

The corresponding extreme point equation value is:

in, is the value of the equation corresponding to the extreme point of the offset.

The new coordinates of the extreme points are generated by adding the offset to the original extreme point coordinates. The pixel values of the generated new coordinates are compared with the set contrast threshold, and the extreme points with low contrast are eliminated. At this time, the remaining extreme points are the feature points.

Step 4: To make the descriptor rotationally invariant, it is necessary to assign a direction to each feature point. To this end, the gradient method is used to obtain the gradient modulus and direction of the pixel points in the neighborhood of the feature point in the image. The formulas for the gradient modulus m(x, y) and direction θ(x, y) are as follows:

θ(x,y)＝tan ^-1 ((L(x,y+1)-L(x,y-1))/(L(x+1,y)-L(x-1,y)) ) (16)

Among them, the scale used by L is the scale of each feature point.

Then, a two-dimensional histogram is used for statistics, and the direction with the highest amplitude in the histogram is taken as the main direction of the feature point. To enhance the robustness of the matching, the direction with an amplitude greater than 80% of the main direction amplitude is retained as the auxiliary direction of the feature point.

Step 5: Take a 16×16 pixel window with the feature point as the center, and divide the window into 4×4 subdomains. Use the gradient direction histogram to count the accumulated gradient amplitudes in 8 directions in each subdomain. At this time, each subdomain can be represented by an 8-dimensional feature description vector. Finally, each feature point has a 4×4×8=128-dimensional feature vector to describe it. In order to make it illumination invariant, the 128-dimensional feature vector is normalized and a threshold is taken to limit the gradient amplitude, which can effectively reduce the impact of uneven illumination on the matching results.

After the SIFT feature vectors of the reference image and the auxiliary image are generated, they are matched using the nearest neighbor method, that is, a ratio threshold is set. If the closest Euclidean distance between the description vectors of two feature points divided by the second closest Euclidean distance is less than this ratio threshold, the two feature points are considered to be matched correctly.

2. Perspective transformation based on improved RANSAC algorithm:

When the highlight removal image is fused, the perspective and size of the auxiliary image need to be kept consistent with the perspective and size of the reference image.

To do this, the auxiliary image needs to be adjusted using perspective transformation. The perspective transformation formula is as follows:

Among them, (x', y') is the coordinate value of the feature matching point of the reference image, (u, v) is the coordinate value of the corresponding feature matching point of the auxiliary image, S is the transformation coefficient between images, and H is a 3×3 transformation matrix, that is:

in, The image can be rotated, scaled, and distorted. T ₂ = [a ₁₃ a ₂₃ ] ^T can translate the image. T ₃ = [a ₃₁ a ₃₂ ] can generate image perspective transformation, as shown in FIG2 .

Since the coordinates of the feature matching points of the reference image and the auxiliary image have been obtained using the SIFT algorithm, we only need to randomly select 4 pairs of feature matching points to obtain the transformation matrix H, and use the transformation matrix to perform perspective transformation on the auxiliary image. The formula for the pixel coordinates (x, y) of the transformed auxiliary image is as follows:

Among them, (x, y) is the pixel coordinate of the auxiliary image after transformation.

Since the nearest neighbor method is overly dependent on the currently preset threshold when matching two feature points, the threshold value cannot be accurately determined. When the threshold is set to a large value, more wrong matching point pairs will appear; when the threshold is set to a small value, although the number of wrong matching point pairs will be reduced, the number of matching point pairs will be significantly reduced, which seriously affects the optimal selection of the transformation matrix H. Therefore, it is necessary to introduce the RANSAC algorithm to further remove the wrong matching point pairs and obtain the optimal transformation matrix.

The idea of the RANSAC algorithm is to fit the estimated model by randomly selecting a set of random subsets from the data, and then test other data through the estimated model. If a certain data is suitable for the estimated model, it is classified as an inlier point. If enough points are classified as hypothetical inliers, the estimated model is considered to be reasonable enough. Then all hypothetical inliers are used to re-estimate the model, and the model is evaluated by estimating the error rate between the inliers and the model. This process is repeated a fixed number of times, and the model generated each time is either discarded because there are too few inliers, or selected because it is better than the existing model.

Since only 4 pairs of matching points are needed when calculating the transformation matrix H, the present invention proposes an improved RANSAC algorithm, and the steps are as follows:

Step 1: First, the effective feature points extracted by the SIFT algorithm are initially matched using the nearest neighbor method, and the initially selected Euclidean distance threshold is 0.6.

Step 2: Divide the image into 4 equal areas, and determine whether the number of feature matching points in the 4 areas is greater than 4. If so, proceed to the next step; otherwise, add 0.1 to the Euclidean distance threshold and return to the previous step to rematch.

Step 3: Select 4 pairs of matching points with the smallest Euclidean distance from each of the 4 regions, for a total of 16 pairs.

Step 4: Combine the 16 pairs of matching points into groups of 4, and sort them from small to large by 1, 2, ..., N according to the Euclidean distance of the 4 pairs of matching points after combination, and select the first 50 groups.

Step 5: First, take the 4 pairs of matching points numbered 1 in order to calculate the transformation matrix H.

Step 6: Use the matrix H to check all matching point pairs in the image. When the number of points in the local area accounts for more than 50% of the total number of matching point pairs, the currently calculated matrix H is considered to be the best transformation matrix. Otherwise, return to the previous step and select the next group of 4 matching point pairs in sequence to calculate the transformation matrix H.

The auxiliary image is adjusted using the optimal transformation matrix, and its corresponding area is masked in the highlight area of the reference image to obtain the reference image with highlights removed.

3. Identification of the status of the pressure plate:

In order to accurately identify the state of the pressure plate in the reference image after highlight removal, the present invention adopts a relay protection pressure plate state recognition method based on image processing and morphological feature analysis. First, in order to improve the accuracy of the overall feature extraction of the pressure plate image, the reference image after highlight removal is subjected to color region screening, binarization, and morphological processing, and the connected area is extracted using an 8-connected method; then, the area, size, and shape analysis are performed based on the morphological features to accurately extract the effective pressure plate area from all areas; finally, the state of the effective area is identified based on the direction angle of the pressure plate when it is in the retracted state, and the effective pressure plate is sorted using the center of gravity coordinates to obtain a sequence of all effective pressure plate states.

1. Extraction of connected regions of pressure plate images:

In order to better reflect the overall and local feature information of the image and accurately extract the connected area of the platen image, the following steps are taken:

Step 1: Since the reference image after highlight removal is a color image, the effective platen in the image is red and yellow, the spare platen is camel and red, and the background area is white, so the red and yellow areas can be screened out by setting a certain RGB threshold. Considering that other interference factors such as other components and logos that may exist in the image will lead to incorrect screening, through a large number of experiments, it is known that the difference between the maximum and minimum values of the red and yellow pixels in the three channels of R, G, and B is not less than 40. Therefore, the pixels in the image that are not less than 40 are retained, and then the remaining areas are set to R, G, B, etc., that is, they are turned black.

Step 2: In order to improve the operation speed, the reference image after the red and yellow areas are filtered is grayed out, and the binarization threshold is obtained using the OTSU algorithm, and the grayscale image is binarized using the threshold.

Step 3: After the image is binarized, some uneven edges will be generated, and there will be holes at the connection points of the pressure plate, which will seriously affect the effect of subsequent feature extraction. Therefore, it is necessary to perform morphological processing on the binary image, fill the holes with dilation and erosion operations, and extract the connected areas in an 8-connected manner (if a pixel and its neighboring pixels are connected at the top, bottom, left, right, upper left corner, lower left corner, upper right corner or lower right corner, they are considered to be connected), and the N connected areas extracted are numbered 1, 2, ..., N.

2. Effective platen area screening:

In order to accurately screen out the effective pressing plate area, the morphological characteristics are analyzed from three aspects: area, size and shape. The specific steps are as follows:

Step 1: Area analysis. Considering that there may be invalid spare plates, logos, etc. in the image after the connected area extraction, it can be seen from the observation that the area of the connected area where this part is located is small. Therefore, an area threshold V _area-thre is set to remove the interference area. The formula is as follows:

The threshold V _area-thre is obtained by multiplying the average area of the top five pixel areas in the binary image by 0.3, and V _area (i) is the area of the i-th area arranged from large to small.

According to formula (20), the area whose pixel area of the connected region is greater than the threshold V _area-thre is determined as a candidate valid pressure plate area, otherwise it is an interference area.

Step 2: Size analysis. Considering that the effective pressure plate in the image after the connected area extraction has a certain size, there is a certain ratio between the effective pressure plate area boundary length and the image pixel size in the X and Y directions. Therefore, the pixel size thresholds X _width-thre and Y _width-thre are set using the image pixels P _X and P _Y in the X and Y directions. The formula is as follows:

According to formula (21), the area whose border lengths of the connected areas in the X and Y directions are greater than the corresponding thresholds is determined to be a candidate effective pressure plate area, otherwise it is an interference area.

Step 3: Shape analysis. Considering that the effective pressure plate in the image after the connected area extraction has a certain shape, the effective pressure plate area has a certain equivalent aspect ratio in the image. At the same time, in order to eliminate other interference information with similar shapes in the image, the equivalent aspect ratio threshold S _ratio-thre is set to 2<S _ratio-thre <5. If the equivalent aspect ratio of the connected area is within the threshold, the area is determined to be a candidate effective pressure plate area, otherwise it is an interference area.

Step 4: Search all connected areas in the image and repeat steps 1-3 until the Nth area is searched and the area determined as a candidate valid pressure plate area is the final valid pressure plate area.

3. Identification of the platen insertion and retraction status:

In order to accurately identify the platen insertion and retraction state in the protective platen highlight removal image after the effective platen area is screened, the present invention uses the direction angle of the platen insertion and retraction state, that is, the direction angle of the platen when it is inserted is ±90°, and the direction angle when it is retracted is ±45°, and a margin of ±10° is set. The judgment formula is as follows:

Among them, the input state is marked as 1, and the exit state is marked as 0. After identifying the effective pressure plate input and exit states, the effective pressure plates are sorted from left to right and from top to bottom using the center of gravity coordinates, and finally a state sequence containing only 0 and 1 is obtained.

From the above, it can be seen that the present invention is based on the protection platen image taken by the intelligent robot inspection, uses the threshold segmentation method to detect the highlight area in the image, and on this basis uses the image fusion method to repair the image, eliminating the situation where the light source at the substation site interferes with the platen status identification, and finally judges its operating status by the inclination angle detected on the edge of the platen based on the image repair. It better assists the inspection robot in checking the platen throw-in and throw-out status, improves the accuracy of the relay protection platen throw-in and throw-out status identification, reduces the labor intensity of the inspection personnel, reduces the misoperation in the power grid operation, avoids economic losses, and ensures the safe and stable operation of the power grid.

Claims (1)

1. The perspective transformation method based on the improved RANSAC algorithm is characterized by comprising the following steps of:

The perspective transformation formula is as follows:

Wherein (x ', y') is the coordinate value of the feature matching point of the reference image, (u, v) is the coordinate value of the feature matching point corresponding to the auxiliary image, S is the transformation coefficient between images, and H is the transformation matrix of 3×3, namely:

wherein, The image can be subjected to rotation, scaling and distortion transformation, T ₂＝[a₁₃ a₂₃]^T can be subjected to translation transformation, and T ₃＝[a₃₁ a₃₂ can generate image perspective transformation;

Because coordinates of the feature matching points of the reference graph and the auxiliary graph are obtained by using the SIFT algorithm, a transformation matrix H can be obtained by randomly selecting 4 pairs of feature matching points, and perspective transformation can be carried out on the auxiliary graph by using the transformation matrix; the transformed auxiliary map pixel coordinates (x, y) are formulated as follows:

Only 4 pairs of matching points are needed when calculating the transformation matrix H, and an improved RANSAC algorithm is provided, which comprises the following steps:

Step ①: firstly, carrying out initial matching on effective feature points extracted by a SIFT algorithm by utilizing a nearest neighbor method, wherein an initial selected Euclidean distance threshold value is 0.6;

Step ②: equally dividing the image into 4 areas, judging whether the number of pairs of characteristic matching points of the 4 areas which are currently divided is larger than 4, if so, carrying out the next step, otherwise, adding 0.1 to the Euclidean distance threshold value, and returning to the previous step for re-matching;

Step ③: screening 4 pairs of matching points with the smallest Euclidean distance from the 4 areas, wherein the total number of the matching points is 16;

step ④: combining the 16 pairs of matching points 4 pairs, sorting 1, 2, … and N according to the Euclidean distance sum of the 4 pairs of matching points after combination from small to large, and selecting the first 50 groups of serial numbers;

Step ⑤: firstly, 4 pairs of matching points with the sequence number of 1 are selected according to the sequence number order to calculate a transformation matrix H;

Step ⑥: and checking all matching point pairs in the image by using the matrix H, when the proportion of the number of the matching point pairs in the office is greater than 50%, considering the currently calculated matrix H as the optimal transformation matrix, otherwise, returning to the previous step, and selecting the next 4 pairs of matching points according to the sequence number to calculate the transformation matrix H.