Method for preparing discrete point image with noise, centroid extraction method and experimental platform

Abstract

The present invention relates to a method for preparing an image of noisy discrete points, a method for extracting the centroid, and an experimental platform. The experimental platform mainly includes three parts: local projection coordinate transformation, preparation of a noisy image, and centroid extraction after noise removal. Based on the discrete point data on the global sphere, the experimental platform uses an imaging plane to intercept a part of the sphere, projects the local discrete points onto the imaging plane, and adds background brightness, image noise, and discrete point brightness to the local discrete point data to complete the preparation of the local discrete point image. The experimental platform realizes the interception, local projection, and coordinate transformation of the discrete points on the global sphere, adds various types of noise at the same time to prepare a local discrete point image with noise, realizes a high similarity simulation with the feature point map in the real environment, and obtains the discrete point data in the local discrete point image with noise through denoising and centroid extraction, which is used as the data basis of point information for subsequent image matching and recognition.

Description

Technical Field

[0001] The present invention relates to image preparation and image processing technologies, and in particular to a method for preparing a discrete point image with noise based on local projection, a method for extracting the centroid, and an experimental platform. Background Art

[0002] In recent years, image matching has become an important field in computer vision, and object recognition and matching based on feature points have also developed into an important application direction of image matching. Discrete point matching based on feature point data has wide demands and applications in fields such as scene matching and starlight navigation. The position matching and recognition of local discrete point images in global discrete points rely on the preparation and processing of image data. In order to prepare a local discrete point map with noise from global spherical discrete points and extract the centroid coordinate data of discrete points from this map, an experimental platform capable of preparing a local discrete point map with noise with specified local projection and noise parameters, as well as high-precision centroid coordinate extraction, is required.

[0003] Currently, the methods for recognizing and extracting the centroid of target discrete points from local images with noise mainly include two methods: filtering and simple threshold segmentation by frame-by-frame accumulation. Both methods have achieved good recognition accuracy, meet the subsequent requirements for object recognition and matching, and have been widely used in engineering. However, since the filtering algorithm cannot effectively and reliably remove noises similar to discrete points themselves, such as pseudo-discrete points and noise points, under noisy conditions, and is prone to losing real discrete points with low brightness and blurred with the background, this method has strong noise requirements for images, and its application scenarios are limited. The method based on frame-by-frame accumulation threshold segmentation can achieve strong anti-interference ability and stability, and the accuracy of centroid extraction can be further improved with the accumulation of the number of frames. However, this algorithm requires multiple images at the same space and the same time, or video data collected at a high frame rate. Therefore, there are certain restrictions on the imaging device and imaging conditions.

[0004] Traditional discrete point image preparation is mostly based on real environment sampling and feature point extraction. Object recognition and matching algorithms usually have a strong dependence on data. Currently, there is no experimental platform that can provide a large number of feature point maps in an environment approximately similar to the real environment. The applicability of the discrete point centroid extraction algorithm is currently limited. There is currently no centroid extraction algorithm with strong anti-disturbance ability and high precision for a single discrete point image under complex noise conditions. Summary of the Invention

[0005] In view of the various deficiencies in the prior art, a method for preparing a discrete point image with noise based on local projection, a method for extracting the centroid, and an experimental platform are proposed. The experimental platform removes the background and determines real discrete points for the prepared local discrete point image with noise with set denoising parameters, realizes the extraction of the centroid coordinates of discrete points, and meets the high-precision requirement of a recognition accuracy higher than 98% under noisy conditions.

[0006] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0007] In a first aspect, the present invention provides a method for preparing a noisy discrete point image based on local projection. The process of this noisy discrete point image preparation method is as follows:

[0008] Obtain two-dimensional discrete point data of local projection to obtain a local discrete point map;

[0009] Set noise parameters, including the mean background brightness, background brightness variance, mean discrete point brightness, discrete point brightness variance, two-direction intensity of the diffusion effect, number of pseudo-discrete points, brightness of pseudo-discrete points, and size of pseudo-discrete points:

[0010] The mean background brightness is a number input manually within 20 or input in the form of a random function between 0 and 20. Preferably, the mean background brightness is 10 - 20;

[0011] The background brightness variance is a number input manually between 1 and 5 or input in the form of a random function between 1 and 5. When the mean background brightness takes a larger value within the range, the background brightness variance correspondingly takes a smaller value within the range;

[0012] The mean discrete point brightness is a number set manually between 100 and 255 or input in the form of a random function between 100 and 255;

[0013] The discrete point brightness variance has a brightness change around the mean discrete point brightness and is greater than the background brightness variance;

[0014] The two-direction intensity of the diffusion effect includes diffusion intensity x and diffusion intensity y. Both diffusion intensity x and diffusion intensity y are between 3 and 5. When there is a difference between x and y, noise representing linear change can be formed;

[0015] The number of pseudo-discrete points is a number input manually between 3 and 5 or input in the form of a random function between 3 and 5; the brightness of pseudo-discrete points is greater than the mean discrete point brightness, and the size of pseudo-discrete points is set between 3*3 and 5*5;

[0016] Introduce the input mean background brightness and mean discrete point brightness into the local discrete point map as background brightness noise and discrete point brightness noise respectively;

[0017] Introduce the Gaussian white noise formed by the input mean background brightness and background brightness variance as the specified mean and intensity into the local discrete point map as background Gaussian white noise;

[0018] According to This formula simulates the noise caused by the linear change and blurring effect of discrete points due to movement during imaging, and introduces it into the local discrete point map;

[0019] Among them, A is the average brightness of discrete points that meets the input requirements, and Q(x q ’, y q ’) is the brightness of discrete points generated randomly according to the Gaussian distribution of the variance parameter of discrete point brightness, σ x’ and σ y’ are the intensity of the dispersion effect of discrete point Q in the x’-axis and y’-axis directions. x’ and y’ are the positions in the local discrete point map, and f(x’, y’) represents the brightness value at any position;

[0020] Using the input pseudo-discrete point brightness as the mean and the discrete point brightness variance as the variance to generate a Gaussian function. The positions of the pseudo-discrete points are completely randomly determined in the entire local discrete point map according to the set size of the pseudo-discrete points. Substituting the randomly determined coordinate values of the pseudo-discrete points into the Gaussian function to obtain the brightness values of the pseudo-discrete points at the corresponding positions, and then determining the image with points. This image with points is used as the noise of the imaging high-brightness noise and pseudo-discrete points and input into the local discrete point map;

[0021] So far, a noisy local discrete point map with background brightness noise, image noise, and discrete point brightness noise introduced in the local discrete point map is obtained;

[0022] The image noise includes the noise of imaging high-brightness noise and pseudo-discrete points, the noise caused by the linear change and blurring effect of discrete points due to movement during imaging, and background Gaussian white noise.

[0023] In the second aspect, a centroid extraction method. The centroid extraction method uses the noisy local discrete point map obtained by the method for preparing a noisy discrete point image based on local projection described above as a reference for centroid extraction,

[0024] Set denoising coefficients, including the threshold window size, the standard deviation multiple value, and the mean median ratio threshold; the threshold window size is 1 / 15 - 1 / 30 of the size of the entire noisy local discrete point map, and the mean median ratio threshold is 1.2 - 1.5;

[0025] Remove the background: Divide the entire noisy local discrete point map into multiple square regions with side lengths equal to the set threshold window size, and calculate the average value μ k and variance δ k :

[0026]

[0027]

[0028] where h is the threshold window size, G(i, j) is the pixel gray value at point (i, j), and μ kis the average gray value of the k-th square region, where i and j are the pixel values of the i-th row and j-th column in the image;

[0029] Then, according to T k = μ k + βδ k calculate the local threshold T within the k-th square region k , where β is the set standard deviation multiple;

[0030] Obtain the local thresholds T of all square regions k . Take the minimum value among all the obtained local thresholds as the minimum local threshold, select the minimum local threshold as the approximation of the background brightness, and retain the image parts greater than this minimum local threshold as discrete points, pseudo-discrete points, and noise regions to form multiple retained regions, thereby achieving background removal;

[0031] Noise removal and determination of true discrete points: For each remaining retained region after background removal, calculate the mean-to-median ratio P(Q′) within the range of each retained region according to . μ is the average of the brightness of all points within the retained region, and med is the median of the brightness of all points within the retained region;

[0032] The ratio of the mean to the median of true discrete points is much smaller than that of pseudo-discrete points and noise. If the mean-to-median ratio within the range of the retained region is greater than the set mean-to-median ratio threshold, it is considered a non-true discrete point and is excluded. Calculate the centroid coordinates (x w ’, y w ’) for the gray values within the remaining retained regions:

[0033]

[0034]

[0035] where p(x’, y’) is the gray value of each coordinate point in the retained region after background removal.

[0036] In a third aspect, the present invention provides an experimental platform for preparing and extracting the centroid of a discrete point image with noise based on local projection, characterized in that: the platform includes a local projection coordinate transformation module, a noise-added image preparation module, and a noise-removing centroid extraction module;

[0037] The local projection coordinate transformation module includes intercepting a part of the global spherical discrete points on a specified imaging plane, projecting the intercepted discrete points onto the imaging plane in a local projection manner, constructing a discrete point image containing only point data, and obtaining a local discrete point map;

[0038] The noise-added image preparation module is used to generate an image approximating the real environment based on local discrete point data, and add background brightness, image noise, and discrete point brightness to the discrete point image containing only point data to prepare a complete feature point map with noise in the approximate real environment, that is, a discrete point image with noise;

[0039] The noise-removed centroid extraction module is used to generate the discrete point coordinate data required for subsequent image matching and recognition, and implement discrete point centroid extraction by using the centroid extraction method described above.

[0040] The software interface of this experimental platform includes:

[0041] The input area for local discrete point imaging parameters, including the number of global discrete points, the field of view range, and the local discrete point imaging size;

[0042] The input area for noise parameters, including the mean background brightness, the variance of background brightness, the mean discrete point brightness, the variance of discrete point brightness, the intensities in two directions of the diffusion effect, the number of pseudo-discrete points, the brightness of pseudo-discrete points, and the size of pseudo-discrete points;

[0043] The input area for denoising parameters, including the threshold window size, the standard deviation multiple, and the mean-median ratio;

[0044] The result graph window, including the imaging image containing only local discrete point data intercepted from the imaging plane, the local discrete point graph with noise after adding noise, and the discrete point centroid graph with the extracted discrete point centroid coordinates marked on this graph.

[0045] Fourthly, the present invention provides a computer-readable storage medium, on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the method for preparing a discrete point image with noise based on local projection or the centroid extraction method can be implemented.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] (1) Through the setting of three imaging parameters, namely the number of global discrete points, the field of view range, and the local discrete point imaging size, the experimental platform of the present invention calculates and obtains the radius and focal length of the global spherical surface, and generates discrete points meeting the set number on this spherical surface. The set imaging parameters are introduced into the imaging plane, and a part of the discrete points on the spherical surface is intercepted based on the imaging plane. After adding three types of noise, namely background brightness, image noise, and discrete point brightness, generated with the set noise parameters in the imaging plane, the preparation of the local discrete point image with noise is completed.

[0048] (2) The experimental platform of this application prepares local discrete point data at any angle, with any noise, and specified imaging parameters, and extracts the centroids of discrete points under noisy conditions. During the denoising process of extracting the centroids of discrete points, it changes the traditional filtering method or the requirement for multiple consecutive frames of images. Based on a single image, it calculates its local threshold to segment the background and the point region, and captures the difference in image continuity between discrete points and pseudo-discrete points by calculating the mean-median ratio of the gray values in the retained region, thereby removing noise. By using the local projection method to intercept local discrete point maps from the global spherical discrete points, local discrete point images can be prepared flexibly and diversely, ensuring the diversity and sufficiency of the data.

[0049] (3) The method for preparing noisy discrete point images of this invention adopts various noise addition methods, approximating the feature point maps collected and calculated in the real environment, and providing rich and approximately real data for subsequent extraction of discrete point centroids or image recognition and matching.

[0050] (4) The centroid extraction method of this invention performs background removal based on the noisy discrete point map obtained by the method for preparing noisy discrete point images, and obtains real discrete points. Then, based on the image gray values and the centroid method, the centroid coordinates of the discrete points are extracted, solving the problems of low accuracy, poor anti-disturbance ability, and high complexity, and achieving accurate recognition and extraction of more than 98% of the discrete point centroids under noisy conditions. Brief Description of the Drawings

[0051] Figure 1 is a schematic diagram of the structure for preparing noisy local discrete point maps and extracting centroids of this invention;

[0052] Figure 2 is a schematic diagram of local projection and coordinate transformation of this invention;

[0053] Figure 3 is a schematic diagram of image denoising and centroid extraction of this invention;

[0054] Figure 4 is a schematic diagram of the experimental plane interface of this invention;

[0055] In the figure:

[0056] F’ - the center point of one side of the imaging plane; H’ - the center point of the side opposite to F’ on the imaging plane, The connecting line of the center points of the opposite sides on the imaging plane. Detailed Embodiments

[0057] The technical solutions of this invention will be clearly and completely described and explained below in combination with the embodiments and the content of the drawings, but this is not used as a limitation of the protection scope of this application.

[0058] Noise similar to the discrete points themselves refers to the pseudo points caused by the direct or deflected light of external noise such as light during the imaging process, as well as the noise points formed during the imaging process. Some pseudo points and noise points have high brightness and are very similar to the discrete points. These noises have a certain degree of randomness and are not easy to remove.

[0059] Multiple images refer to the multiple images at the same position and at the same time required by the method of frame-by-frame cumulative threshold segmentation. However, this is an ideal situation. Therefore, images within a video with a high frame rate are usually used. These images are generated at similar positions within an extremely short period of time. Since the noise has a certain degree of randomness while the characteristic discrete points are fixed, multiple images taken at the same position and at the same time can be superimposed on each other to enhance the regional characteristics of the discrete points with fixity and weaken the noise characteristics with randomness, thus achieving a strong anti-interference ability.

[0060] Image recognition and matching have a strong dependence on data. The quantity, type, and noise inclusion situation of the data directly affect the accuracy and effect of subsequent matching and recognition. Therefore, it is necessary to prepare a large amount of data from multiple angles and containing various noises. However, due to the limited number of data images collected in the real environment, restricted by conditions such as the shooting environment and light, the data volume is often much smaller than the requirement, and the data types and situations are few. Therefore, an experimental platform that can provide a large amount of feature point data similar to the real environment is needed.

[0061] The present invention proposes an experimental platform for preparing and centroid extracting noisy discrete point images based on local projection, including a local projection coordinate transformation module, a noise-adding image preparation module, and a noise-removing centroid extraction module. As Figure 1 shown, the three parts are carried out in sequence. After the noise addition, a local discrete point map with noise can be prepared directly for image recognition and matching. After the centroid coordinate data of the discrete points in the local discrete point map with noise is extracted, it can be directly used for discrete point matching based on data.

[0062] The local projection coordinate transformation module needs to import imaging parameters into the experimental platform, generate discrete points approximately uniformly distributed on the global spherical surface, and introduce the imaging plane to intercept the discrete points on the spherical surface with a specified focal length. The intercepted discrete points on the spherical surface are projected onto the imaging plane by means of local projection, and the corresponding coordinate transformation is performed to convert the three-dimensional spherical coordinate system into a two-dimensional imaging coordinate system.

[0063] The noise-adding image preparation module needs to import the two-dimensional discrete point data that has completed local projection and coordinate transformation into the experimental platform, and add background brightness, image noise, and discrete point brightness by setting noise parameters. The image noise includes background Gaussian white noise, noise formed by the linear change and blur effect of discrete points due to movement during imaging, imaging high-brightness noise points, and pseudo-discrete point noise. After the noise addition is completed, the experimental platform automatically exports the prepared local discrete point map with noise.

[0064] Noise removal centroid extraction module: Before starting, import the prepared local discrete point map with noise in the experimental platform, set the denoising parameters according to the prepared image conditions, calculate the regional threshold in the form of area scanning, perform threshold segmentation on the image based on this threshold, further calculate the mean-median ratio of the extracted discrete point region to remove the noise points therein, and finally calculate the centroid of the real discrete point region based on the gray value to complete the extraction of the discrete point centroid coordinate data. The denoising parameters depend on the intensity of the added image noise itself; the mean-median ratio, the mean is the average of the gray values in the point region, the median is the middle value of the gray values in the point region, and the average divided by the middle value is the mean-median ratio. The specific method for extracting the discrete point centroid after removing the background to determine the real discrete points is the prior art.

[0065] Both local projection and coordinate transformation are prior methods. Local projection is the process of projecting the discrete points on the spherical surface onto the imaging plane, calculating the coordinates of the intersection point of the line connecting the discrete points on the spherical surface and the center of the sphere with the imaging plane to obtain the position of the projection points. Coordinate transformation converts the three-dimensional spherical coordinates of the projection points into two-dimensional imaging coordinates for subsequent image processing. In this application, since the local discrete point image is not obtained by real shooting, a local projection coordinate transformation module needs to be introduced.

[0066] In this application, the selection of the noise type needs to be combined with the actual image noise and can reproduce the noise contained in the image taken in the real environment. Therefore, there are certain requirements for the selection of the noise type and the addition method. According to the camera imaging principle, the noise sources of the camera are mainly three types: non-uniform response noise, random noise, and fixed noise. Among them, the non-uniform response noise is brought by the manufacture of the electronic circuit, which is manifested as different and non-uniform responses of different pixels in the camera under the condition of the same brightness, specifically manifested as the non-uniformity of the background noise. Random noise is due to the fact that the light entering the camera and being received by the sensor is not continuous but intermittent, so it will show a certain randomness during the photosensitive imaging. Fixed noise is manifested as the inherent bias of each pixel. Therefore, Gaussian white noise is added to the entire image, and a blur effect is added to the point region to simulate the noise of the camera. Under high-sensitivity conditions, the random shot noise of the camera will increase and the fixed noise will decrease, which is manifested as some small-area but high-brightness noise points generated during the imaging process. Therefore, high-brightness pseudo points are added to the image to simulate the noise points. During the long exposure process of the camera, slight jitter will bring a certain linear change to the imaging. Therefore, a certain linear bias is added to the point imaging region to simulate this part of the noise.

[0067] The present invention uses three-dimensional spherical coordinates to define the global spherical discrete points. With the origin O of this coordinate system as the center of the sphere, a sphere with a radius of r is generated, and a set number of points represented by three-dimensional spherical coordinates (r, Discrete points, where the azimuth angle θ ∈ (0°, 360°) follows a uniform distribution, and the polar angle follows a normal distribution. After inputting the three imaging parameters of the global discrete point number n, the field of view α, and the local discrete point imaging size in the experimental platform, a square imaging plane with a side length equal to the imaging size is generated in the three-dimensional spherical coordinate system, and the line connecting the center point of the imaging plane and the origin is perpendicular to the imaging plane. As Figure 2 shown, the center of the sphere, i.e., the origin, is point O, the distance between the center point O' of the imaging plane and the origin O is the focal length f, and the calculation of the focal length f and the spherical radius r is as follows:

[0068]

[0069]

[0070] Among them, the field of view α is the angle formed by the lines connecting the center points F' and H' of the two opposite sides of the imaging plane and the origin O. Convert the coordinates of all discrete points on the sphere from three-dimensional spherical coordinates to three-dimensional rectangular coordinates (x, y, z):

[0071]

[0072]

[0073]

[0074] Among them, the point (x, y, z) is the coordinate of the point in the three-dimensional rectangular coordinate system. The line connecting the discrete point on the sphere and the origin O intersects the imaging plane, and the intersection point is the local projection point of the discrete point on the imaging plane. If there is no intersection point, it is considered that the discrete point on the sphere has no projection on the imaging plane and is not included in the local discrete point diagram. Let a discrete point M(x m , y m , z m ) on the sphere, and its line connecting with the origin O This line The intersection point M'(x m’ , y m’ , z m’ ) with the imaging plane is:

[0075]

[0076] If x m x m' + y m y m' + z m z m' ≥ f 2, then point M is outside the imaging plane, and it is considered that the local projection of the discrete point M on the spherical surface exists on the imaging plane. If x m x m' +y m y m' +z m z m' ≤f 2 , then point M is inside the imaging plane, and it is considered that its local projection on the imaging plane does not exist. Here, the inside and outside are relative to the relative position between the imaging plane and the spherical surface. The position between the imaging plane and the spherical surface is the inside, and the position far from the imaging plane is the outside. After calculating all the local projection coordinates, taking the center point O' of the imaging plane as the origin of the imaging plane, and taking the connection line between the center point O' and the local projection point with the farthest distance as the x'-axis to construct a two-dimensional imaging coordinate system. Intercept the abscissa and the ordinate of the local projection. x' and y' are the coordinates of the x'-axis and y'-axis of the two-dimensional imaging coordinate system respectively, and the local projection coordinate transformation of the local discrete point image is completed.

[0077] Set the noise parameters in the experimental platform, including the background brightness mean, background brightness variance, discrete point brightness mean, discrete point brightness variance, two-direction intensity of the diffusion effect, number of pseudo-discrete points, brightness of pseudo-discrete points, and size of pseudo-discrete points. Denoising coefficient: threshold window size, standard deviation multiple, mean-median ratio threshold (1.2 - 1.5). Imaging parameters: field of view, number of discrete points.

[0078] This experimental platform generates a total of three types of noise: background brightness, image noise, and discrete point brightness. Among them, the image noise includes background Gaussian white noise, noise caused by the linear change and blur effect of discrete points due to movement during imaging, imaging high-brightness noise points (randomly generate a specified number of noise points), and pseudo-discrete points. As Figure 3 shown, by inputting two noise parameters, the background brightness mean and background brightness variance, in the platform, Gaussian white noise with a specified intensity and mean can be added, that is, the background Gaussian white noise and background brightness can be determined. The mean is the set background brightness mean, the intensity is the set background brightness variance, and the change caused by the variance is Gaussian white noise.

[0079] In the actual imaging process, affected by the sensor exposure time and shaking, point imaging usually has a diffusion function blur effect and corresponding linear changes, which are manifested as a diffusion effect in which the gray level gradually decreases from the center of the point imaging to the surrounding area. Considering the linear changes brought by movement, the diffusion has directionality. This phenomenon can be simulated by the following formula:

[0080]

[0081] Among them, A is the average brightness of discrete points that meet the input requirements, and Q(x q ’, y q ’) is the brightness of discrete points randomly generated according to the Gaussian distribution of the variance parameter of discrete point brightness. σ x’ and σ y’ are the diffusion effect intensities of discrete point Q in the x’-axis and y’-axis directions. x’, y’ are the positions in the local discrete point map, and f(x’, y’) represents the brightness value at any position, indicating the influence of the diffusion and linear change of a discrete point on the entire image. When the intensity of the y-axis is large and the intensity of the x-axis is small, linear change noise can be shown.

[0082] According to the imaging characteristics of imaging high-brightness noise points and pseudo-discrete points, namely high brightness, high degree of separation from the background, and small pixel area, a point image with a high mean value and a low standard deviation is generated locally using the Gaussian function. Imaging high-brightness noise points and pseudo-discrete points are determined by these factors: the number of pseudo-discrete points (randomly set to 3 - 5 to match the actual situation), the brightness of pseudo-discrete points (the brightness is set higher than the average brightness of discrete points), and the size of pseudo-discrete points (the size is small, set between 3*3 - 5*5). Taking the brightness of pseudo-discrete points as the mean value and the variance of discrete point brightness as the variance, a Gaussian function is generated. The positions of pseudo-discrete points are completely randomly determined in the entire local discrete point map according to the set size of pseudo-discrete points. The coordinate values of randomly determined pseudo-discrete points are substituted into the Gaussian function to obtain the brightness values of pseudo-discrete points at the corresponding positions, and then the point-containing image is determined. This point-containing image is used as the noise of imaging high-brightness noise points and pseudo-discrete points and input into the local discrete point map.

[0083] So far, a local discrete point map with noise is obtained.

[0084] After completing the preparation of the local discrete point map with noise, the denoising coefficients (threshold window size, standard deviation multiple value, mean median ratio threshold) are set in the experimental platform. Since there is no prior background noise information and the discrete points are sparse, the brightness of the brightness region without points (true discrete points and pseudo-discrete points) is used as an approximation of the background brightness value. The entire local image is divided into multiple square regions with side lengths equal to the set threshold window size, and the average value μ k and variance δ k of the pixel gray values in the k-th square region are calculated as follows:

[0085]

[0086]

[0087] where h is the threshold window size, G(i, j) is the pixel gray value of this point, μ k is the average gray value of the k-th square region, and i, j are the pixel values of the i-th row and j-th column in the image. The local threshold T k in the k-th square region is:

[0088] T k = μ k + βδ k

[0089] where β is the set standard deviation multiple coefficient. Obtain the local threshold T of all square regions k , and the minimum value among all local thresholds is the minimum local threshold. Select the minimum local threshold as the approximation of the background brightness, and retain the image part greater than the minimum local threshold as the discrete point, pseudo-discrete point, and noise region, obtaining multiple retained regions.

[0090] For each retained region, calculate the ratio P(Q′) of its range mean to median as follows:

[0091]

[0092] where μ is the average of the brightness of all points in the retained region, and med is the median of the brightness of the points in the retained region.

[0093] Since the brightness distribution of real discrete points has obvious dispersion characteristics and continuity, the ratio of its mean to median is usually much smaller than that of pseudo-discrete points and noise. If the ratio of the mean to median in the retained range is greater than the set mean-median ratio threshold, it is considered as the region where non-real discrete points are located, and this region is excluded. Determine the discrete points through the mean-median ratio, and calculate the centroid coordinates ((x w ’, y w ’) of the remaining regions according to the gray values within their ranges as follows:

[0094]

[0095]

[0096] where p(x’, y’) is the gray value of each coordinate point in the retained region after removing the background.

[0097] The centroid results of the discrete points extracted from the local discrete point map with noise are as Figure 3 shown, and mark the centroid of each discrete point with *.

[0098] Figure 4This is an experimental platform for preparing noisy discrete point images and extracting centroid based on local projection of the present invention, which has a processing unit and a display unit. The above modules are integrated in the processing unit and the methods for preparing noisy discrete point images and extracting centroid based on local projection are executed. The software interface of the display unit includes a parameter input window, a result graph window, a button for generating local graph, a button for adding noise, and a button for centroid extraction. Imaging parameters, noise parameters, and denoising parameters are all input in the parameter input window. After setting the parameters, click the button for generating local graph, and a random imaging plane can be generated, and the processes of local projection and coordinate transformation can be completed. The intercepted local discrete point graph is displayed in the coordinate area below. After generating the local discrete point graph, click the button for adding noise to complete the noise addition process, and the prepared noisy local discrete point graph is displayed in the coordinate area below this button. Finally, click the button for centroid extraction to perform denoising processing on the noisy local discrete point graph to complete the extraction of the centroid of discrete points. The noisy local discrete point graph and the result of centroid extraction will be displayed below.

[0099] Test Figure 4 The experimental platform in the test generates 10,280 random discrete points on the spherical surface with a field of view of 20° and an image size of 1024*1024px, and constructs 1000 noisy local discrete point graphs. The centroid extraction is performed in sequence. Finally, the success rate of discrete point centroid extraction under noisy conditions exceeds 98%. It is proved that the experimental platform can complete the extraction of local discrete point graphs based on local projection and coordinate transformation, project the spherical surface in the three-dimensional spherical coordinate system to the two-dimensional imaging coordinate system, and realize the preparation of noisy local discrete point graphs of approximate real feature point graphs through the addition of noise. Under this condition, high-precision extraction of discrete point centroid coordinate data based on threshold segmentation and centroid extraction algorithms is achieved. In the generation and centroid extraction of local images of more than ten thousand random discrete points, an extraction accuracy of more than 98% is achieved.

[0100] The present invention uses the local projection method to realize the coordinate transformation from the three-dimensional spherical coordinate system to the two-dimensional imaging coordinate system, and completes the preparation of noisy local discrete point graphs of feature point graphs in an approximate real environment by adding background brightness, image noise, and discrete point brightness to the image. In the process of extracting discrete point centroid coordinates, algorithms such as background removal, noise point removal, and weighted centroid extraction are used to obtain high-precision discrete point centroid coordinate data under noisy conditions for subsequent image matching and recognition.

[0101] Embodiment

[0102] The method for preparing noisy discrete point images based on local projection in this embodiment is used in scene matching and starlight navigation, and is used to prepare any noisy discrete point images that conform to the global spherical images in scene matching and starlight navigation, and then is used in subsequent image matching and recognition. The specific process is as follows:

[0103] Obtain the two-dimensional discrete point data of the local projection to get the local discrete point map;

[0104] Obtain the global spherical coordinate data in scene matching and starlight navigation, and intercept it through local projection to obtain the two-dimensional discrete point data of the local projection, and then obtain the local discrete point map;

[0105] Set the noise parameters, including the background brightness mean, background brightness variance, discrete point brightness mean, discrete point brightness variance, two-direction intensity of the diffusion effect, number of pseudo-discrete points, brightness of pseudo-discrete points, size of pseudo-discrete points:

[0106] The background brightness mean is a number input manually within 20 or input by a random function between 0 and 20. Preferably, the background brightness mean is 10 - 20;

[0107] The background brightness variance is a number input manually between 1 and 5 or input by a random function between 1 and 5. When the background brightness mean takes a larger value within the range, the background brightness variance takes a smaller value within the range accordingly;

[0108] The discrete point brightness mean is a number set manually between 100 and 255 or input in the form of a random function between 100 and 255;

[0109] The discrete point brightness variance has a brightness change around the discrete point brightness mean and is greater than the background brightness variance;

[0110] The two-direction intensity of the diffusion effect includes diffusion intensity x and diffusion intensity y. Both diffusion intensity x and diffusion intensity y are between 3 and 5. When there is a difference between x and y, noise representing linear change can be formed;

[0111] The number of pseudo-discrete points is a number input manually between 3 and 5 or input by a random function between 3 and 5; the brightness of the pseudo-discrete points is greater than the discrete point brightness mean, and the size of the pseudo-discrete points is set between 3*3 and 5*5;

[0112] Introduce the input background brightness mean and discrete point brightness mean into the local discrete point map as background brightness noise and discrete point brightness noise respectively;

[0113] Introduce the Gaussian white noise formed by the input background brightness mean and background brightness variance as the specified mean and intensity into the local discrete point map as background Gaussian white noise;

[0114] According to This formula simulates the noise caused by the linear change and blur effect of discrete points due to movement during imaging, and introduces it into the local discrete point map;

[0115] Among them, A is the average brightness of discrete points that meets the input requirements, and Q(x q ’, y q ’) is the brightness of discrete points generated randomly according to the Gaussian distribution of the variance parameter of discrete point brightness. σ x’ and σ y’ are the intensity of the diffusion effect of discrete point Q in the x’-axis and y’-axis directions. x’ and y’ are the positions in the local discrete point map, and f(x’, y’) represents the brightness value at any position;

[0116] Generate a Gaussian function with the input pseudo-discrete point brightness as the mean and the discrete point brightness variance as the variance. The positions of the pseudo-discrete points are randomly determined completely in the entire local discrete point map according to the set size of the pseudo-discrete points. Substitute the randomly determined coordinate values of the pseudo-discrete points into the Gaussian function to obtain the brightness values of the pseudo-discrete points at the corresponding positions, and then determine the image with points. This image with points is used as the noise of the imaging high-brightness noise and pseudo-discrete points and input into the local discrete point map;

[0117] So far, a noisy local discrete point map with background brightness noise, image noise, and discrete point brightness noise introduced in the local discrete point map has been obtained, realizing the preparation of a noisy discrete point image in scene matching or starlight navigation for subsequent image matching.

[0118] Matters not described in the present invention are applicable to the prior art.

Claims

1. A method for preparing a noisy discrete point image based on local projection, characterized in that, the process of this noisy discrete point image preparation method is: Obtain two-dimensional discrete point data of local projection to get a local discrete point map; Set noise parameters, including background brightness mean, background brightness variance, discrete point brightness mean, discrete point brightness variance, two-direction intensity of diffusion effect, number of pseudo-discrete points, brightness of pseudo-discrete points, size of pseudo-discrete points: The background brightness mean is a number input manually within 20 or input by a random function between 0 - 20, and preferably the background brightness mean is 10 - 20; The background brightness variance is a number input manually between 1 - 5 or input by a random function between 1 - 5. When the background brightness mean takes a larger value within the range, the background brightness variance correspondingly takes a smaller value within the range; The discrete point brightness mean is a number set manually between 100 - 255 or input in the form of a random function between 100 - 255; The discrete point brightness variance has a brightness change around the discrete point brightness mean and is greater than the background brightness variance; The two-direction intensity of the diffusion effect includes diffusion intensity x and diffusion intensity y. Both diffusion intensity x and diffusion intensity y are between 3 - 5. When there is a difference between x and y, noise characterizing linear change can be formed; The number of pseudo-discrete points is a number input manually between 3 - 5 or input by a random function between 3 - 5; the brightness of the pseudo-discrete points is greater than the discrete point brightness mean, and the size of the pseudo-discrete points is set between 3*3 - 5*5; Introduce the input background brightness mean and discrete point brightness mean as background brightness noise and discrete point brightness noise into the local discrete point map respectively; Introduce the Gaussian white noise formed by the input background brightness mean and background brightness variance as the specified mean and intensity as background Gaussian white noise into the local discrete point map; According to This formula simulates the noise caused by the linear change and blurring effect of discrete points due to motion during imaging, and introduces it into the local discrete point map; Among them, A is the average brightness of discrete points that meets the input requirements, and Q(x q ’, y q ’) is the brightness of discrete points generated randomly according to the Gaussian distribution of the variance parameter of discrete point brightness. σ x’ and σ y’ are the intensity of the dispersion effect of discrete point Q in the x'-axis and y'-axis directions. x', y' are the positions in the local discrete point map, and f(x', y') represents the brightness value at any position; Generate a Gaussian function with the input brightness of the pseudo-discrete points as the mean and the discrete point brightness variance as the variance. The positions of the pseudo-discrete points are completely randomly determined in the entire local discrete point map according to the set size of the pseudo-discrete points. Substitute the randomly determined coordinate values of the pseudo-discrete points into the Gaussian function to obtain the brightness values of the pseudo-discrete points at the corresponding positions, and then determine the point-containing image. This point-containing image is used as the noise of the imaging high-brightness noise points and pseudo-discrete points and input into the local discrete point map; Thus, a noisy local discrete point map with background brightness noise, image noise, and discrete point brightness noise introduced into the local discrete point map is obtained; The image noise includes the noise of imaging high-brightness noise points and pseudo-discrete points, the noise caused by the linear change and blurring effect of discrete points due to movement during imaging, and background Gaussian white noise.

2. A centroid extraction method, characterized in that, The centroid extraction method uses the noisy local discrete point map obtained by the method for preparing a noisy discrete point image based on local projection described in claim 1 as a reference for centroid extraction, Set denoising coefficients, including threshold window size, standard deviation multiple value, and mean median ratio threshold; the threshold window size is 1 / 15 - 1 / 30 of the size of the entire noisy local discrete point map, and the mean median ratio threshold is 1.2 - 1.5; Background removal: The entire noisy local discrete point map is divided into multiple square regions with a side length equal to the set threshold window size, and the average value μ of the pixel gray values within the k-th square region is calculated. k and variance δ k : where h is the threshold window size, G(i, j) is the pixel gray value at the point (i, j), and μ k is the average gray value of the k-th square region, and i, j are the pixel values of the i-th row and j-th column in the image; Then, according to T k = μ k + βδ k Calculate the local threshold T within the k-th square region k , where β is the set standard deviation multiple value; Obtain the local threshold T for all square regions k , take the minimum value among all the obtained local thresholds as the minimum local threshold, select the minimum local threshold as the approximation of the background brightness, retain the image parts greater than the minimum local threshold as discrete points, pseudo-discrete points and noise regions, form multiple retained regions, and achieve background removal; Noise removal, determining true discrete points: For each remaining retained area after background removal, according to Calculate the mean-to-median ratio P(Q′) within each retained area. μ is the average of the brightness values of all points within the retained area, and med is the median of the brightness values of all points within the retained area; The ratio of the mean to the median of the true discrete points is much smaller than that of the pseudo-discrete points and noise points. If the ratio of the mean to the median within the retention region is greater than the set mean-median ratio threshold, it is considered a non-true discrete point and is excluded. Calculate the centroid coordinates (x w ’, y w ’) for the gray values within the remaining retention region: where p(x’, y’) is the gray value of each coordinate point in the remaining area after background removal.

3. An experimental platform for preparing a noisy discrete point image based on local projection and extracting the centroid, characterized in that: The platform includes a local projection coordinate transformation module, a noise-added image preparation module, and a noise-removed centroid extraction module; The local projection coordinate transformation module includes intercepting a part of the global spherical discrete points with a specified imaging plane, projecting the intercepted discrete points onto the imaging plane in a local projection manner, constructing a discrete point image containing only point data, and obtaining a local discrete point map; The noise-added image preparation module is used to generate an image approximating the real environment based on the local discrete point data, and add background brightness, image noise, and discrete point brightness to the discrete point image containing only point data to prepare a complete feature point map of the approximate real environment with noise, that is, a noisy discrete point image; The noise-removed centroid extraction module is used to generate the discrete point coordinate data required for subsequent image matching and recognition, and implement the discrete point centroid extraction by using the centroid extraction method described in claim 2.

4. The experimental platform for preparing a noisy discrete point image based on local projection and extracting the centroid according to claim 3, characterized in that The software interface of the experimental platform includes: A local discrete point imaging parameter input area, including the number of global discrete points, the field of view range, and the local discrete point imaging size; A noise parameter input area, including the mean background brightness, the background brightness variance, the mean discrete point brightness, the discrete point brightness variance, the intensities in two directions of the diffusion effect, the number of pseudo-discrete points, the pseudo-discrete point brightness, and the pseudo-discrete point size; A denoising parameter input area, including the threshold window size, the standard deviation multiple, and the mean-median ratio; A result graph window, including an imaging image containing only local discrete point data intercepted on the imaging plane, a noisy local discrete point graph after adding noise, and a discrete point centroid graph with the extracted discrete point centroid coordinates marked on the graph.

5. A computer-readable storage medium, on which a computer program is stored, characterized in that When the program is executed by a processor, the steps of the method described in claim 1 or 2 can be implemented.

Images