A mobile target extraction image processing method and system based on hough transform

Abstract

The application discloses a kind of mobile target extraction image processing method and system based on Hough transform, mainly related to astronomical application field.It includes the following steps: collecting image data and carrying out image preprocessing, star image extraction and astronomical positioning;Weak mobile target in the star image extracted by the above steps is detected by three-dimensional Hough transform.The beneficial effects of the present application are that it can greatly improve the processing speed and capacity of monitoring data, and is more conducive to improving the monitoring capacity in the field of space debris, and can promote the industrial development of space debris image processing and application.

Description

Technical Field

[0001] This invention relates to the field of astronomical applications, specifically to an image processing method and system for extracting moving targets based on Hough transform. Background Technology

[0002] In astronomical optical image processing, it is crucial to clearly and accurately distinguish whether the source of a signal in an image is noise or a target source. Current optical image processing generally involves two steps: image preprocessing and target source extraction. Image preprocessing primarily corrects biases in the observed image and sometimes performs background processing, such as basic image transformations (median filtering, mean filtering); Bayesian methods for background estimation; matched filtering to highlight objects and reduce background fluctuations; and multi-scale parallel methods. Target source extraction methods include thresholding (in computer vision, thresholding is a simple method for target source extraction, converting grayscale images into binary images), local maximum search (the main principle of local peak search is to search for pixels considered to be peaks, or in other words, local maxima in the neighborhood), and neural networks (transforming the target detection problem into a classification problem, classifying pixels, and using a neural network for filtering, similar to machine learning). However, these methods are highly susceptible to the influence of the image background when extracting faint targets, cannot handle situations where background stars overlap with the target, and cannot distinguish between noise and the target at low signal-to-noise ratios.

[0003] Target detection in astronomical observation images involves detecting specific targets from star images extracted through image processing. Common target detection methods can be broadly categorized into two types: Detect before track (DBT) and Track before detect (TBD). DBT-based target detection is a relatively intuitive approach. Its implementation involves sampling within a single frame of the image to identify all possible candidate targets. Then, various restrictive conditions are applied to eliminate false targets, ultimately obtaining the true target, after which tracking begins. TBD-based target detection algorithms, however, do not know which target is the true target during tracking. Therefore, they need to continuously track all visible targets within the field of view for a period of time. After tracking concludes, all trajectories are analyzed, and the posterior probability of each trajectory belonging to the true target is calculated. The most probable trajectory is then selected as the target's motion trajectory, thus achieving the true target detection function.

[0004] Currently, there are many methods for detecting targets in astronomical images. In 2005, Han Jiantao et al. improved the symmetric difference operation of images and developed a new method for detecting small moving targets in astronomical images. The method assumes that at a certain time interval (t... k, t k+1 , ..., t k+m-1 A sequence of m images acquired by an internal CCD, after preprocessing, is represented as follows:

[0005] {f(x i y j , t k ), f(x) i y j , t k+1 ), …, f(x) i y j , t k-m+1 )} i, j = 1, ..., N

[0006] The algorithm steps are as follows:

[0007] (1) Consider a sequence of three consecutive frames of images f(x) i y j , t k ), f(x) i y j , t k+1 ), f(x) i y j , t k+2 ). Calculate the absolute difference in grayscale value g(x) between two adjacent frames. i y j , t k ), g(x i y j , t k+1 ):

[0008] g(x i y j , t k )=|f(x i y j , t k+2 )-f(x i y j , t k )|

[0009] g(x i y j , t k+1 )=|f(x i y j , t k+2 )-f(x i y j , t k+1 )|

[0010] (2) Background estimation using the first three frames:

[0011]

[0012] (3) Convert the grayscale image g(x) i y j , t k ), g(x i y j , t k+1 g is obtained by multiplying the corresponding pixel values. 2 (x i y j , t k );

[0013] (4) Suppress f(x) i y j , t k+2 The background of ) is:

[0014] f -Bg (x i y j , t k )=f(x i y j , t k+2 )-Bg(x i y j , t k )

[0015] (5) Obtain the final enhanced image:

[0016] f En (x i y j , t k )=f -Bg (x i y j , t k )*g 2 (x i y j , t k )

[0017] (6) For grayscale image f En (x i y j , t k Automatic threshold segmentation is performed to obtain a binary image B(x). i y j , t k ); for B(x) i y j , t k Post-processing is performed, using sequential filtering to remove isolated noise points segmented from the background and filling background gaps in the target region, resulting in the original image f(x). i y j , t kThe moving target segmentation image F(x) i y j , t k ).

[0018] (7) Perform the above operation on the next three consecutive frames to obtain the moving target segmentation image of the m-2 frame sequence:

[0019] {F(x i y j , t k ), F(x) i y j , t k+1 ), …, F(x) i y j , t k+m-3 )} i, j = 1, ..., N

[0020] (8) Apply the target point set matching method based on simple prediction to identify the moving target image of the original image m-2 frame sequence image and remove false targets in the m-2 frame segmented image.

[0021] However, this method is only suitable for stationary telescopes or situations where the background of the corresponding sky area changes very little within the frame interval. It also requires a small field of view and slow processing speed.

[0022] Therefore, current astronomical optical image processing suffers from problems such as low extraction rates of faint targets and inability to distinguish overlapping star images; target detection cannot effectively and promptly extract moving targets such as space debris from large field-of-view survey images. Therefore, there is a need to develop a method for rapid detection of faint moving targets in large field-of-view survey images, capable of accurately extracting and rapidly detecting moving target star images, especially faint moving target star images. Summary of the Invention

[0023] The purpose of this invention is to provide a moving target extraction image processing method and system based on Hough transform, which can not only greatly improve the processing speed and capability of monitoring data, but also improve the monitoring capability in the field of space debris, and promote the industrialization of space debris image processing and application.

[0024] To achieve the above objectives, the present invention employs the following technical solution:

[0025] The process includes the following steps: acquiring image data and performing image preprocessing, star image extraction and astronomical positioning; and detecting faint moving targets in the star images extracted in the above steps using three-dimensional Hough transform.

[0026] Preferably, the image preprocessing includes background correction, dark flow correction, and flat field correction.

[0027] Preferably, the star image extraction includes time extraction, position extraction, flow rate extraction, and signal-to-noise ratio extraction of non-extended star images.

[0028] Preferably, the astronomical positioning specifically involves: using the coordinates of stars in a known reference star catalog as a standard, determining the transformation relationship between the pixel coordinate system and the celestial coordinate system, and then calculating the actual spatial location of the target.

[0029] Preferably, the detection of faint moving targets by three-dimensional Hough transform mainly includes the following steps: constructing a point cloud; performing a three-dimensional Hough transform to obtain the linear equation of the target's motion trajectory in the point cloud; and evaluating the linear equation of the trajectory obtained in the above steps.

[0030] Preferably, the point cloud construction specifically involves: using the right ascension and declination coordinates of all extracted targets as the horizontal and vertical coordinates of the point cloud, and using time as the Z-axis to construct the point cloud.

[0031] Preferably, the process of obtaining the linear equation of the target's trajectory in the point cloud through the three-dimensional Hough transform specifically involves defining a straight line in the three-dimensional space XYZ, described as follows: in Define a plane passing through the origin and perpendicular to the aforementioned line, and define x. ′ and y ′ The coordinates of the intersection point of a line and the plane within the plane's own 2D coordinate framework are obtained by rotating the original XY plane around the origin to the X... ′ -Y ′ The resulting rotation matrix is ​​based on the linear direction vector in the plane. for:

[0032]

[0033] From any point on the straight line Let's begin and transform:

[0034]

[0035] get:

[0036]

[0037]

[0038]

[0039] and direction vector

[0040] at this time This can be represented by two parameters x′ and y′. By setting the step size dx, we can discretize X′ and Y′ to obtain... The X′-Y′ plane is discretized into individual face cells. Where i∈N2, j∈N3, a parameter space is constructed using B, X′, and Y′. The point cloud is input into the parameter space, and the coordinates (x0, y0, z0) of any input point are transformed into (x′0, y′0, z′0) using a transformation matrix, which represents in the parameter space. These cells, where k∈N1, i0∈N2, j0∈N3 and Since the point (x0, y0, z0) has no definite direction, That is, the point (x0, y0, z0) votes once for each of these cells in the parameter space, and each cell in the parameter space... Where k∈N1, i∈N2, j∈N3, represents a straight line. Find the cell with the highest number of votes, determine the corresponding line parameters, and then return to the point cloud to find all points whose distance to the line is less than the cell width. Use the least squares method to fit the best straight line that passes through all points, delete these points from the parameter space, and repeat the above operation until the line contains fewer points than the set parameters, then stop the iteration.

[0041] Preferably, the evaluation of the trajectory straight line equation specifically involves: using the straight line equation to find the target point on the straight line, interpolating the target point over time, eliminating interpolation points outside the field of view, and searching for target points near the interpolation points; making a decision on the straight line based on the condition of the target point to determine whether the straight line is true or false; and outputting the target point on the reliable straight line.

[0042] Preferably, the direction vector Specifically, the acquisition is as follows: on the straight line middle, The direction vectors are straight lines and can be discretized using Pareto tessellation. An icosahedron is chosen, and polygonal triangulation is used to divide each triangle on the icosahedron surface into four new triangles. A new vertex is inserted between each pair of adjacent vertices of a triangle, and its length is normalized. Performing this operation on all polygonal triangles generates a new vertex for each edge and three new edges for each new vertex. This process is repeated for the desired refinement level, stopping after four subdivisions of the icosahedron, resulting in 1281 distinct direction vectors.

[0043] A moving target extraction image processing system based on Hough transform includes a data acquisition module, a data processing module, and a data capture module. The data acquisition module is used to acquire specific data of the star image where the moving target is located. The data processing module is used to calculate and process the acquired data. The data capture module is used to capture data information related to the line where the moving target is located from the processed data.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] It can be applied to space debris monitoring systems to enhance the system's image processing capabilities. It can be used for the rapid extraction of space targets by large-aperture, wide-field telescopes and arrays, and can also be used for the search for asteroids in large-scale sky survey data.

[0046] In recent years, with the large-scale construction of telescopes, the amount of astronomical observation data will inevitably increase. Space debris is related to the space environment and aerospace safety, and the observation data of space debris will increase significantly. Therefore, a technology that can process space debris on a large scale and quickly is needed.

[0047] It can not only greatly improve the processing speed and capacity of monitoring data, but also enhance the monitoring capabilities in the field of space debris, and promote the industrialization of space debris image processing and application. Attached Figure Description

[0048] Figure 1 It is a flowchart of the method.

[0049] Figure 2 It is the trajectory of a moving target across multiple images.

[0050] Figure 3 It is the original point cloud construction.

[0051] Figure 4 It is a point cloud after noise reduction.

[0052] Figure 5 This is a diagram illustrating coordinate rotation.

[0053] Figure 6 This is a schematic diagram of a Plato mosaic.

[0054] Figure 7 This is a schematic diagram of the three-dimensional Hough transform result. Detailed Implementation

[0055] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0056] Example: An image processing method and system for moving target extraction based on Hough transform

[0057] The specific methods and procedures are as follows: Figure 1 As shown.

[0058] For faint targets, extracting target information from images often leads to misidentification. However, in the field of space debris, the targets to be observed are mostly moving targets. If multiple images are taken consecutively, aligning and overlaying them will highlight the trajectory of moving, faint targets, such as... Figure 2 As mentioned above, identifying such trajectories can yield information about faint targets, thereby enabling accurate extraction of faint moving targets in images. In actual observations, target trajectories are usually straight lines observed over short periods; therefore, we propose using the Hough transform method for identification.

[0059] The Hough transform is a feature detection method widely used in image analysis, computer vision, and digital image processing. The traditional two-dimensional Hough transform can transform the motion trajectories of spatial fragments in multiple images to a single point in parameter space, enhancing faint star images and thus identifying the motion trajectories of faint targets. However, the two-dimensional Hough transform has significant limitations in practical image processing. Therefore, after extensive research, we propose using the three-dimensional Hough transform. While also a parameter space transformation, the three-dimensional Hough transform's approach and solution are more efficient and accurate than the two-dimensional Hough transform for our specific problem. The specific process of this method is as follows:

[0060] (1) Image preprocessing, star image extraction, and astronomical positioning

[0061] Before target extraction, the obtained image needs to be preprocessed to correct the influence of the CCD itself on the image. Image preprocessing includes background correction, dark flow correction, and flat field correction.

[0062] Background refers to the minimum amount of photocharge recorded by each pixel on a CCD when it is not exposed to light. Due to minor fluctuations in grid voltage, environmental noise, and instrument variations, the background changes during observation. Therefore, it is necessary to measure the background in multiple images before, during, or after observation, and take the average value as the usable value. Background correction involves subtracting the background from all images.

[0063] Dark flow refers to the hot electrons generated in the potential well of a CCD when it is powered on. Dark flow has a cumulative effect, its magnitude is related to the integration time, and it is very sensitive to temperature. If some pixels are abnormal, hot electrons will accumulate rapidly, forming noise. The method to measure dark flow is to close the shutter, power the CCD for a relatively long time, and read out the CCD image, which is the dark flow image. Dark field correction is to darken all images.

[0064] Fiat is a measure of the sensitivity of each pixel on a CCD. It can arise from factors such as inconsistent spectral response of pixels, uneven filter thickness, and misalignment between the CCD plane and the focal plane. The method for measuring fiat involves exposing the CCD to a uniform diffuse light source multiple times and taking the average or median value as the fiat image. To eliminate the influence of fiat, the target image is divided by the merged fiat image, thus completing fiat correction.

[0065] Astronomical positioning is a technique that uses the coordinates of stars in a known reference star catalog as a standard to determine the transformation relationship between the pixel coordinate system and the celestial coordinate system, and then calculates the actual spatial position of the target. For large field-of-view survey observation data, it can be assumed that the background stars and moving target star images are not significantly elongated, and can be approximated as non-elongated star images. Star image extraction mainly includes the extraction of time, position, flux, and signal-to-noise ratio of non-elongated star images such as background stars and space debris, all of which are performed using traditional methods for target extraction.

[0066] (2) Detection of weak moving targets

[0067] Low-light moving target detection consists of three steps. The first is point cloud construction: using the right ascension and declination coordinates of all extracted sources as the horizontal and vertical coordinates of the point cloud, and time as the Z-axis, the point cloud is constructed, as follows: Figure 3 As shown. Since the magnitudes of right ascension and declination are not the same as those of time, we divide right ascension and declination by the pixel scale to enlarge them to the scale of image pixels; we convert time to seconds and then enlarge it by a factor of 10, which facilitates parameter settings in subsequent processes. At this point, the point cloud data volume is huge and there is a lot of interference from stars, so we remove duplicates of points with the same right ascension and declination coordinates at different times and remove stars to achieve noise reduction. After the above processing, the motion trajectory of the moving target in the point cloud becomes prominent, as shown. Figure 4 As shown.

[0068] Secondly, there's the Hough transform. Essentially, the Hough transform is a cumulative effect, enabling the detection of weak target signals. To achieve this cumulative effect, a three-dimensional Hough transform is required. In XY space, points are represented by coordinates (X, Y); lines are described by the parametric equation (B, K): Y = KX + B. Conversely, a KB space can be constructed, where points are represented by coordinates (B, K); lines are described by the parametric equation (X, Y): B = -KX + Y. There's a transformation relationship between XY and KB spaces; XY space is called the original space, and KB space is called the parameter space. This is the spatial transformation of the two-dimensional Hough transform. In the three-dimensional Hough transform, the original space is XYZ space, and six parameters are needed to describe a straight line. in The parameter space is the px-py-pz-bx-by-bz space. Space transformations are quite complex at this point. Depending on the specific situation, we can simplify the parameters and discretize the parameter space to simplify the solution.

[0069] When the position of a line is represented by an arbitrary anchor point, this will result in a position vector. Of the three parameters, one is redundant. Roberts' optimal line representation is a method to eliminate this redundancy. This method first defines a plane passing through the origin and perpendicular to the line; then, it sets the two parameters x... ′ and y ′ Defined as the coordinates of the intersection point of a line and the plane within the plane's own 2D coordinate framework, such as... Figure 5 As shown, this is equivalent to rotating the original XY plane around the origin to the X... ′ -Y ′ Plane, rotation matrix A based on the direction vector of the line for:

[0070]

[0071] From any point on the straight line Let's begin and transform:

[0072]

[0073] get:

[0074]

[0075]

[0076]

[0077] at this time Two parameters x can be used ′ and y ′express.

[0078] In a straight line middle, It is a direction vector of a straight line, which can be discretized using Pareto tessellation. An icosahedron is chosen, such as... Figure 6 As shown in (a), polygonal triangulation is used to divide each triangle on the surface of the icosahedron into four new triangles. This is done by inserting a new vertex between each pair of adjacent vertices of the triangle and normalizing its length, as follows: Figure 6 As shown in (b). Performing this operation on all polygonal triangles generates a new vertex for each edge and three new edges for each new vertex. This operation can be repeated as needed for the desired level of refinement. We stop after subdividing the icosahedron four times, resulting in 1281 different direction vectors.

[0079] Linear position vector Let x′ and y′ represent the values. Before performing the three-dimensional Hough transform, a step size dx needs to be set to discretize x′ and y′, resulting in... The X′-Y′ plane is discretized into individual face cells. Where i∈N2, j∈N3. Construct a parameter space L using B, X′, and Y′. Input the point cloud D into L. The coordinates (x0, y0, z0) of any input point are transformed into (x′0, y′0, z′0) using the transformation matrix A, representing [the coordinates] in the parameter space. These cells, where k∈N1, i0∈N2, j0∈N3 and Since the point (x0, y0, z0) has no definite direction, That is, the point (x0, y0, z0) votes once for each of these cells in the parameter space. Each cell in the parameter space... Where k∈N1, i∈N2, j∈N3, represents a straight line. Find the cell with the highest number of votes, determine the corresponding line parameters, and then return to the point cloud to find all points whose distance to the line is less than the cell width. Fit the best straight line passing through all points using the least squares method, and remove these points from the parameter space. Repeat the above operation until the line contains fewer points than the set parameters, then stop the iteration. Finally, obtain the linear equation of the target's motion trajectory in the point cloud. Points on these lines are potential moving targets, such as... Figure 7 As shown.

[0080] Before outputting the target, the line needs to be evaluated. The target point on the line is found using the line equation. The target point is then interpolated over time, interpolated points outside the field of view are removed, and target points near the interpolated points are searched. Then, the line is evaluated based on the condition of the target point to determine whether the line is true or false. Finally, the target point on the reliable line is output, realizing fast and reliable detection of faint moving targets.

[0081] Example 2: An Image Processing System for Moving Target Extraction Based on Hough Transform

[0082] It includes a data acquisition module, a data processing module, and a data capture module. The data acquisition module is used to acquire specific data of the star image where the moving target is located. The data processing module is used to calculate and process the acquired data. The data capture module is used to capture data information related to the line where the moving target is located from the processed data.

Claims

1. An image processing method for moving target extraction based on Hough transform, characterized in that, Includes the following steps: S1 acquires image data and performs image preprocessing, star image extraction, and astronomical positioning; S2 uses a three-dimensional Hough transform to detect faint moving targets in the star image extracted in step S1. The detection of faint moving targets using three-dimensional Hough transform includes the following steps: S21 Point Cloud Construction: The point cloud is constructed using the right ascension and declination coordinates of all extracted targets as the horizontal and vertical coordinates of the point cloud, and time as the Z-axis. S22 performs a three-dimensional Hough transform to obtain the linear equation of the target's trajectory in the point cloud; S23 evaluates the trajectory line equation obtained in step S22, specifically: using the line equation to find the target point on the line, interpolating the target point in time, eliminating interpolation points outside the field of view, and searching for target points near the interpolation points; making a decision on the line based on the condition of the target point to determine whether the line is true or false; and outputting the target point on the reliable line.

2. The image processing method for moving target extraction based on Hough transform according to claim 1, characterized in that, The image preprocessing includes background correction, dark flow correction, and flat field correction.

3. The image processing method for moving target extraction based on Hough transform according to claim 1, characterized in that, The star image extraction includes time extraction, position extraction, flow extraction, and signal-to-noise ratio extraction of non-extended star images.

4. The image processing method for moving target extraction based on Hough transform according to claim 1, characterized in that, The astronomical positioning specifically involves using the coordinates of stars in a known reference star catalog as a standard to determine the transformation relationship between the pixel coordinate system and the celestial coordinate system, and then calculating the actual spatial location of the target.

5. The image processing method for moving target extraction based on Hough transform according to claim 1, characterized in that, The specific method for obtaining the linear equation of the target's trajectory in the point cloud through the three-dimensional Hough transform is as follows: A straight line is defined in the three-dimensional space XYZ, and the straight line is described as follows: ,in , Define a plane passing through the origin and perpendicular to the aforementioned line. and The coordinates of the intersection point of a line and the plane within the plane's own 2D coordinate framework, that is, the coordinates of the original XY plane rotated around the origin to... - The resulting rotation matrix is ​​based on the linear direction vector in the plane. for: From any point on the straight line Let's begin and transform: get: and direction vector B = ; at this time Use two parameters and This means that setting the step size dx will... and Discretize, and obtain = , = , - The plane is discretized into individual face cells. Using B , Construct a parameter space by inputting the point cloud into the parameter space, where the coordinates of any input point are... Transform using the transformation matrix In parameter space, it represents Due to the point There is no definite direction, so = , i.e., point Each of these cells in the parameter space is voted on once. Representing a straight line, find the cell with the highest number of votes, determine the corresponding line parameters, and then return to the point cloud to find all points whose distance to the line is less than the cell width. Use the least squares method to fit the best line passing through all points, delete these points from the parameter space, and repeat the above operation until the line contains fewer points than the set parameters, at which point the iteration stops.

6. The image processing method for moving target extraction based on Hough transform according to claim 5, characterized in that, The direction vector B = Specifically, the acquisition is as follows: on the straight line middle, The direction vector is a straight line. Discretization is performed using a Pareto tessellation. An icosahedron is chosen, and polygonal triangulation is used to divide each triangle on the icosahedron surface into four new triangles. A new vertex is inserted between each pair of adjacent vertices of a triangle, and its length is normalized. This process is repeated for all polygonal triangles, generating a new vertex for each edge and three new edges for each new vertex. This process is repeated for the desired refinement level, stopping after four subdivisions of the icosahedron, resulting in 1281 distinct direction vectors. The direction vector B = .

7. A moving target extraction image processing system based on Hough transform, used to implement the moving target extraction image processing method based on Hough transform as described in any one of claims 1-6, characterized in that, It includes a data acquisition module, a data processing module, and a data capture module. The data acquisition module is used to acquire specific data of the star image where the moving target is located. The data processing module is used to calculate and process the acquired data. The data capture module is used to capture data information related to the line where the moving target is located from the processed data.

Images