High-likelihood background analysis and judgment method for small unmanned aerial vehicle aerial image registration

Abstract

The invention discloses a high-likelihood background analysis and judgment method for small unmanned aerial vehicle aerial image registration. The method comprises the following steps: selecting adjacent images according to collected image numbers and establishing a pixel matrix of the adjacent images; calculating a mean value, a variance and a covariance of adjacent image pixels; the correlation coefficient of the adjacent images # imgabs0 # and # imgabs1 # is calculated; analyzing the availability of the images # imgabs2 # and # imgabs3 # as high-likelihood background samples; and judging whether the adjacent image is a highly-similar background image or not. According to the method, the high-likelihood background image is provided for efficiently and rapidly obtaining the image registration feature points, the high-likelihood background can greatly reduce the search range of the feature points and obviously reduce the search noise of the feature points, the search efficiency of the feature points can be remarkably improved, and the method has very high practical value in three-dimensional live-action modeling of a construction site.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a high-similarity background analysis and determination method for small unmanned aerial vehicle aerial image registration. Background Art

[0002] In the 3D real-scene modeling of construction project construction sites, the use of small drones to collect image data of the construction site is very cost-effective. However, the images of the construction site collected by small drones are small in size and large in number, which has a certain impact on the image registration and feature point extraction in the 3D real-scene modeling of the construction site.

[0003] From the process of finding and extracting image feature points in the 3D real scene modeling of the construction site, the current method of finding image feature points is to input all the construction site images collected by the drone into the 3D modeling software, and then conduct a full range of feature point search and extraction work on all images, such as Figure 1 As shown. Obviously, the main problems of this method of finding feature points for image registration are large range, high noise, slow speed, long time consumption, poor effect and low efficiency. Therefore, if we want to improve the efficiency of finding feature points, reduce the noise of finding feature points, and realize the rapid extraction and determination of feature points, we must give a more valuable and accurate feature point finding range.

[0004] In fact, in image registration and stitching, the images that need to be registered and stitched are all adjacent images, and one common feature of these images is that they all have partially overlapping content. However, due to the influence of various factors such as acquisition time difference, sunshine intensity, wind speed and direction, temperature and humidity in aerial photography, the acquired images have different degrees of differences. Therefore, these partially overlapping images are only images with high similarity, not completely consistent images. Summary of the invention

[0005] In view of the problems and defects described in the above-mentioned prior art, the present invention proposes a high-similarity background analysis and determination method for small UAV aerial image registration, which can analyze adjacent images and select the high-similarity background in the adjacent images.

[0006] A high-similarity background analysis and determination method for small-scale unmanned aerial vehicle aerial image registration, the method comprising the following steps: Selecting adjacent images according to the acquired image numbers and establishing a pixel matrix of the adjacent images; Calculate the mean, variance, and covariance of neighboring image pixels; Calculate neighboring images and The correlation coefficient of Analyze the image and The availability of samples with high similarity to background; Determine whether the adjacent image is highly similar to the background image.

[0007] Furthermore, the method of selecting adjacent images according to the collected image numbers and establishing a pixel matrix of adjacent images specifically includes: assuming that the image and Represent two adjacent images, and Represent the pixels in two adjacent images respectively. and Respectively represent the pixels in the image and The row and column numbers in the image and The pixel matrix is: ; Assume that the BGR value of each pixel in the image is and , the image pixel matrix can be further expressed as: .

[0008] Furthermore, the calculating of the mean, variance and covariance of adjacent image pixels specifically includes: make The mean value of the pixels in the jth column of the image is , the variance is , the covariance of the jkth column is , The mean of the mth column of the image is , the variance is , the covariance of the pm column is ,but, ; ; ; ; ; .

[0009] Furthermore, we calculate the neighboring images and The correlation coefficient includes: The correlation coefficient of the jkth column is ,image The correlation coefficient of the pm column is , then we can get the following formula: ; .

[0010] Further, analyzing the image and The availability of high-like-background samples includes: and The smaller the value, the smaller the correlation between the color vectors of each pixel in the selected image, that is, the image is more suitable as a high-similarity background sample.

[0011] Furthermore, the determination of whether the adjacent image is a highly similar background image specifically includes: performing a preliminary sample analysis, if it is determined that the image and As a high-similarity background image sample, the image and Corresponding image point difference analysis to analyze and judge and Whether it is a high-resolution background image.

[0012] Furthermore, the analysis is specifically as follows: Respectively, images and The B, G, and R differences in the BGR values ​​of each corresponding pixel in the pixel matrix are: Since the maximum value of BGR is 255, and The proportion of the difference value of any pixel in the image in its total BGR value and They are: ; ; in, is the similarity mean of the two images, is the total number of pixels in the image, is the similarity threshold; The larger the value, the lower the similarity, and the smaller the value, the higher the similarity; When it is below 25, it means and Below 10%; and When the similarity mean of is less than the predetermined value, it can be determined and It is a high-resolution background image.

[0013] The present invention has the following beneficial technical effects: (1) The present invention can significantly improve the efficiency of finding feature points. By providing a highly similar background image in advance before searching for feature points in neighboring images, the search for feature points is locked in the highly similar background image, greatly narrowing the search range for image feature points, and providing technical support for the rapid search and extraction of image feature points.

[0014] (2) In the three-dimensional real scene modeling of the construction site, the technology proposed in the present invention can be used to write the technology into a high-similarity background image analysis program, and then the program can be implanted into the current three-dimensional real scene modeling software for the engineering construction site. In this way, the search range of the feature points of the adjacent images that need to be spliced ​​can be greatly reduced, so that the search for feature points is only carried out in the high-similarity background image, and the search area for the adjacent image registration feature points is greatly reduced, which provides domain-bound support for the rapid search and determination of image feature points, and can speed up the aerial calculation and three-dimensional modeling speed of the later modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention has the following accompanying drawings: Figure 1 Methods for finding feature points for existing technologies; Figure 2 It is a schematic diagram of the function of the present invention in three-dimensional modeling work. DETAILED DESCRIPTION

[0016] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0017] The following is combined with Figure 1-2 The present invention is described in further detail.

[0018] The construction site terrain images collected by small drones are all color images, and each pixel in these images can be represented by a BGR (Blue Green Red) value. If the two adjacent images to be registered are named and , the following method can be proposed according to the present invention to carry out analysis and judgment of highly similar background images.

[0019] Step 1: Select adjacent images according to the collected image numbers and establish a pixel matrix of the adjacent images; Set image and Represent two adjacent images, and Represent the pixels in two adjacent images respectively. and Respectively represent the pixels in the image and The row and column numbers in the image and The pixel matrix is: ; Assume that the BGR value of each pixel in the image is and , the image pixel matrix can be further expressed as: ; Step 2: Calculate the mean, variance and covariance of neighboring image pixels; Since any feature point in the image or The BGR value of the image can be obtained by specifying pixel coordinates and assignment languages ​​through various existing image analysis software (for example, when using the image analysis software OpenCV, the image can be read through the cv2.imread() function, and the BGR value of a specific pixel can be directly accessed through the coordinate value index). Then, the image can be obtained through the following analysis. and The mean, variance, and covariance of .

[0020] make The mean value of the pixels in the jth column of the image is , the variance is , the covariance of the jkth column is , The mean of the mth column of the image is , the variance is , the covariance of the pm column is ,but, ; ; ; ; ; ; Step 3: Calculate neighboring images and Correlation coefficient The image is obtained through step 2 and After the mean, variance and covariance of The correlation coefficient of the jkth column is ,image The correlation coefficient of the pm column is , then we can get the following formula: ; ; Step 4: Analyze the image and Availability of high-quality background samples because and It indicates the closeness between the image points in the selected image. Therefore, it can be used to judge whether the selected image is worth selecting as a high-likeness background sample. The analysis method is: and The smaller the value, the smaller the correlation between the color vectors of each pixel in the selected image, which means that the image is more suitable as a high-similarity background sample; vice versa. and There is no strict limit on the selection of the value. The smaller the value, the higher the value of the high-similarity background, and the higher the efficiency of feature point search and extraction. and The value is set below 0.4.

[0021] Step 5: Determine whether the adjacent image is highly similar to the background image If the sample is initially analyzed, the image is determined and can be used as a high-similar background image sample, then the image and Corresponding image point difference analysis to analyze and judge and Whether it is a high-similarity background image. The analysis method is: make Respectively, images and The B, G, and R differences in the BGR values ​​of each corresponding pixel in the pixel matrix are: Since the maximum value of BGR is 255, and The proportion of the difference value of any pixel in the image in its total BGR value and They are: ; ; in, is the similarity mean of the two images, is the total number of pixels in the image, , is the similarity threshold. The larger the value, the lower the similarity, and the smaller the value, the higher the similarity. When it is below 25, it means and Below 10%. and When the similarity mean of is less than the predetermined value, it can be determined and It is a high-resolution background image.

[0022] Figure 2 It is a system diagram describing the role of the present invention in the three-dimensional real scene modeling of the construction site and the related work. The content of the present invention is the blue part in the figure, and the rest expresses the role of the present invention in the three-dimensional real scene modeling of the construction site and the relationship with other work.

[0023] It should be noted that the present invention provides a high-similarity background image for efficiently and quickly acquiring image registration feature points. This high-similarity background can not only greatly narrow the search range of feature points, the noise in the search will also be significantly reduced, but also the search efficiency of feature points will be significantly improved. It has high practical value in three-dimensional real-scene modeling at construction sites.

[0024] The above embodiments describe the technical solutions of the present invention in detail. Obviously, the present invention is not limited to the described embodiments. Based on the embodiments of the present invention, people familiar with the technical field can also make various changes accordingly, but any changes that are equivalent or similar to the present invention belong to the scope of protection of the present invention.

[0025] The contents not described in detail in this specification belong to the prior art and professional common sense known to professional and technical personnel in this field.

Claims

1. A high-similarity background analysis and determination method for small UAV aerial image registration, characterized in that: The method comprises the following steps: Selecting adjacent images according to the acquired image numbers and establishing a pixel matrix of the adjacent images; Calculate the mean, variance, and covariance of neighboring image pixels; Calculate neighboring images and The correlation coefficient of Analyze the image and The availability of samples with high similarity to background; Determine whether the adjacent image is highly similar to the background image.

2. The method according to claim 1, characterized in that The method of selecting adjacent images according to the collected image numbers and establishing a pixel matrix of adjacent images specifically includes: assuming that the image and Represent two adjacent images, and Represent the pixels in two adjacent images respectively. and Respectively represent the pixels in the image and The row and column numbers in the image and The pixel matrix is: ； Assume that the BGR value of each pixel in the image is and , the image pixel matrix can be further expressed as: 。 3. The method according to claim 2, characterized in that The calculating of the mean, variance and covariance of adjacent image pixels specifically includes: make Image The mean value of the pixels in the column is , the variance is , the covariance of the jkth column is , The mean of the mth column of the image is , the variance is , the covariance of the pm column is ,but, ； ； ； ； ； 。 4. The method according to claim 3, characterized in that The calculated neighboring image and The correlation coefficient includes: The correlation coefficient of the jkth column is ,image The correlation coefficient of the pm column is , then we can get the following formula: ； 。 5. The method according to claim 4, characterized in that Analyze the image and The availability of high-like-background samples includes: and The smaller the value, the smaller the correlation between the color vectors of each pixel in the selected image, that is, the image is more suitable as a high-similarity background sample.

6. The method according to claim 5, characterized in that The method of determining whether the adjacent image is high-like Background image, specifically including: through preliminary analysis of the sample, if it is determined that the image and As a high-similarity background image sample, the image and Corresponding image point difference analysis to analyze and judge and Whether it is a high-resolution background image.

7. The method according to claim 6, characterized in that The analysis is specifically as follows: make Respectively, images and The B, G, and R differences in the BGR values ​​of each corresponding pixel in the pixel matrix are: Since the maximum value of BGR is 255, and The proportion of the difference value of any pixel in the image in its total BGR value and They are: ； ； in, is the similarity mean of the two images, is the total number of pixels in the image, is the similarity threshold; The larger the value, the lower the similarity, and the smaller the value, the higher the similarity; When it is below 25, it means and Below 10%; and When the similarity mean of is less than the predetermined value, it can be determined and For high-resolution background images.

8. The method according to claim 7, characterized in that Said and The value is set below 0.4.

Images