Parameter estimation method based on calculation and vision algorithm

Abstract

The invention relates to a parameter estimation method based on a calculation vision algorithm. The parameter estimation method comprises the following steps: S1, acquiring a cascade delay power curve image subjected to low-pass filtering; s2, performing image preprocessing on the cascade delay power curve image to obtain a first image, the image preprocessing including graying; s3, obtaining the curve intensity and the curve angle of each pixel point of the first image by constructing a Hessian matrix of each pixel point of the first image, and further combining with a first line point condition for screening to obtain a first line point set; s4, performing line point connection based on the first line point set to obtain a multipath component trajectory; and S5, mapping the obtained multipath component trajectory to a channel physical parameter coordinate system to obtain parameters of each multipath component. Compared with the prior art, the method can track the birth and death conditions of continuous and blocked multipath components.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a parameter estimation method based on computer vision algorithms. Background Technology

[0002] In time-varying scenarios, the number of Multipath Components (MPCs) and the parameters associated with each MPC typically change over time, leading to the evolution of MPCs and the emergence of time-varying MPC clusters. To ensure the spatial consistency of the channel model, research is underway to capture the complete dynamic behavior of each MPC. However, because sensing parameter estimation is highly nonlinear, most classical linear models widely used in communication channel estimation cannot be directly applied to sensing parameter estimation. Studying the spatiotemporal evolution characteristics of MPCs is helpful for compensating for communication channels and performing object detection, localization, and velocity estimation. Therefore, rapidly and accurately tracking the spatiotemporal evolution characteristics of MPCs is crucial for establishing reasonable and effective time-varying channel models.

[0003] Existing technologies have proposed various algorithms for tracking the spatiotemporal evolution characteristics of MPCs. One proposed Kalman-enhanced super-resolution tracking algorithm can track the evolution of continuous MPCs, but it cannot handle congested scenarios. A proposed MPC tracking method based on geometric spatial consistency can track MPCs that disappear over a long period due to congestion, ensuring the spatial consistency of MPCs; however, this method requires high-precision algorithms for processing, which is very time-consuming. A proposed MPC tracking technique based on Hough transform can effectively track expected MPCs, but its tracking performance is poor for MPCs without defined features. Summary of the Invention

[0004] The purpose of this invention is to provide a parameter estimation method based on computer vision algorithms.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A parameter estimation method based on computer vision algorithms includes:

[0007] Step S1: Obtain the cascaded delay power curve image after low-pass filtering;

[0008] Step S2: After performing image preprocessing on the cascaded time delay power curve image, a first image is obtained, wherein the image preprocessing includes grayscale conversion;

[0009] Step S3: Obtain the curve intensity and curve angle of each pixel in the first image by constructing the Hessian matrix of each pixel in the first image, and further combine the first line point condition to filter and obtain the first line point set;

[0010] Step S4: Connect the line points based on the first set of line points to obtain the multipath component trajectory;

[0011] Step S5: Map the obtained multipath component trajectories to the channel physical parameter coordinate system to obtain the parameters of each multipath component.

[0012] Step S3 includes:

[0013] Step S3-1: Calculate the second derivative of each pixel in an image with respect to x, the second derivative with respect to y, and the second derivative with respect to xy, and construct the Hessian matrix for each pixel in the first image based on this calculation.

[0014]

[0015] Where: H(x,y) is the Hessian matrix of the pixel, with the x-coordinate of the pixel being x and the y-coordinate being y, I xx Let I be the second derivative of the pixel with respect to x. yy Let I be the second derivative of the pixel with respect to y. xy Let x be the second derivative of the pixel with respect to x and y.

[0016] Step S3-2: Obtain the curve intensity and curve angle of each pixel in the image by calculating the eigenvalues ​​and eigenvectors of the Hessian matrix.

[0017] Step S3-3: Based on the eigenvalues ​​and eigenvectors of the Hessian matrix, select line points that meet the first line point condition;

[0018] Step S3-4: Based on the line points obtained from the screening, further select linear structure line points with bright foreground to construct the first line point set.

[0019] The curve intensity and curve angle of each pixel are:

[0020] μ(x,y)=max(λ1,λ2)

[0021]

[0022] Where μ(x,y) is the curve intensity of the pixel. Let λ1 and λ2 be the curve angles of the pixels, λ1 and λ2 be the eigenvalues ​​of the Hessian matrix, and N be the unit eigenvectors of the Hessian matrix. x N represents the x-axis component of the unit eigenvectors of the Hessian matrix. y Let y be the component of the unit eigenvector of the Hessian matrix along the y-axis.

[0023] The first line point condition is:

[0024] C={(x,y)∈I|(p x ,py )∈[-0.5,0.5]×[-0.5,0.5]}

[0025] (p x ,p y )=(tN x ,tN y )

[0026] Where: C is the first line point condition, p x p is the x-coordinate of a sub-pixel. y Let y be the y-coordinate of a subpixel, t be a factor in the second-order Taylor expansion, and I be the first image.

[0027] In steps S3-4, the desired line point matrix E for a bright foreground is:

[0028] E={(x,y)∈C|abs(λ1) <T1∧λ2<T2}

[0029] Where: abs(λ1) is the absolute value of the eigenvalue λ1 of the Hessian matrix, T1 and T2 are the line point structure thresholds, and λ1 and λ2 are the eigenvalues ​​of the Hessian matrix.

[0030] The connection methods in step S4 include eight-neighbor search, distance extension search, and vanishing curve extension search. Eight-neighbor search is performed first. If eight-neighbor search fails, distance extension search is performed. If distance extension search fails, vanishing curve extension search is performed.

[0031] The eight-neighborhood point search includes:

[0032] Using the curve direction information of the line points, the curve angle is divided into eight equal parts. Each angle corresponds to three adjacent pixels in the eight-neighborhood. The next possible connecting line point belongs to these three adjacent pixels.

[0033] After obtaining three adjacent pixels, the best connecting line point is selected by combining the sub-pixel distance difference and angle difference. If none of the three adjacent pixels belong to the line points with bright foreground, the eight-neighborhood point search fails.

[0034] The distance extension point search includes:

[0035] Using the current point position as a base, extend linearly a certain distance to obtain the coordinates of a circle's center;

[0036] Draw a circle at the center of the circle. The set of line points included inside the circle represents the line points that can be found by distance extension point search.

[0037] The desired connecting line point is selected by combining curve intensity and angle conditions. If the possible connecting line points within the circle are an empty set, the distance extension point search fails, and the selected desired connecting line point P is not determined.e for:

[0038] P e ={(x,y)∈E|Δd(E,r) <R∧min([AddCondition SA (E,s)]}

[0039] Where: r is the desired line point matrix, E is the desired line point matrix, and AddCondition is the key. SA Add a condition to the curve intensity-angle, where R is the radius of the extended circle and s is the current line point.

[0040] The vanishing curve extension point search includes:

[0041] Calculate the sum of gray levels of the grayscale image and the defined geometric shape within a certain range, starting from the current line point;

[0042] Select the geometric region corresponding to the maximum sum of gray levels as the region where line points may exist, and perform condition calculations on the line points within this region to obtain the desired connecting line points.

[0043] A parameter estimation device based on a computer vision algorithm includes a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the method described above.

[0044] A storage medium having a program stored thereon, which, when executed, implements the method described above.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] This invention tracks the spatiotemporal evolution characteristics of MPCs with spatial consistency based on the Steger curve tracing algorithm. It improves the connection logic on the Steger algorithm, enabling the method to be widely applied in MPC tracing and parameter estimation in wireless communication. Two connection methods are designed to effectively address MPC congestion, which is difficult to handle with existing techniques, ensuring spatial consistency of MPCs. In implementation, this invention does not require high-precision parameter estimation algorithms for preprocessing, nor does it require setting numerous thresholds or providing MPC trajectory features, thus enabling fast processing and wide applicability. After tracing different MPC trajectories, the delay, power, and time (distance) parameters corresponding to each MPC are estimated, which is significant for studying the complete dynamic behavior of each MPC and improving the accuracy of wireless channel models. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the process of the present invention;

[0048] Figure 2This is the simulation measurement environment for an embodiment of the present invention;

[0049] Figure 3 This is a cascaded delay power curve image according to an embodiment of the present invention;

[0050] Figure 4 The filtered image is an embodiment of the present invention;

[0051] Figure 5 This is a grayscale image from an embodiment of the present invention;

[0052] Figure 6 This is a heatmap of the intensity curve of a grayscale image, representing an embodiment of the present invention.

[0053] Figure 7 This is a vector diagram of the curve intensity superposition curve direction for processing grayscale images according to an embodiment of the present invention;

[0054] Figure 8 This is a line point distribution diagram that satisfies the line point conditions according to an embodiment of the present invention;

[0055] Figure 9 This is a line point distribution diagram that satisfies the line point structure threshold according to an embodiment of the present invention;

[0056] Figure 10 This is the result of eight-neighborhood point finding and connection in an embodiment of the present invention;

[0057] Figure 11 This is the distance extension point finding and connection result of an embodiment of the present invention;

[0058] Figure 12 This is the result of finding and connecting points by extending the vanishing curve in an embodiment of the present invention;

[0059] Figure 13 This is the result of mapping the trajectory to the original coordinate system in an embodiment of the present invention. Detailed Implementation

[0060] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0061] The example provides a simulated measurement environment and simulation data generated using the ray tracing method. Figure 2 To simulate the measurement environment, Figure 3 This is an image of the cascaded time delay power curve generated based on the ray tracing method.

[0062] In the simulation measurement environment, the transmitter is a drone hovering in the middle of the lane, with coordinates (0,0,20), and the transmitter antenna height is 20m. The receiver is a car moving at a constant speed from the leftmost (-100,0,2) to the rightmost (100,0,2) of the lane, with the receiver antenna height at 2m and the car speed at 6m / s. The receiver receives each snapshot for 4.2ms, with a snapshot interval of 0.3367s. Each snapshot consists of 10 frames, with an interval of 0.42ms between each frame. The specific simulation measurement settings are shown in Table 1.

[0063] Table 1

[0064] parameter Value Center frequency 6.0GHz bandwidth 400MHz Frequency points 1001 Snapshot Frames 10 Snapshot duration 4.2ms Snapshot Interval 0.3367s Total number of frames 1000 Receiver moving speed 6m / s Transmitter antenna height 20m Receiver antenna height 2m

[0065] This application proposes a parameter estimation method based on computer vision algorithms, such as... Figure 1 As shown, it includes:

[0066] Step S1: Obtain the cascaded delay power curve image after low-pass filtering;

[0067] Specifically, the process of obtaining the cascaded delay power curve image of the low-pass filter is as follows:

[0068] The procedure involves obtaining channel data and then using filtering techniques to reduce noise and enhance the trajectory of the data, such as... Figure 4 As shown, this is a cascaded time-delay power curve image after low-pass filtering, where sidelobe noise is significantly reduced.

[0069] Step S2: After image preprocessing of the cascaded time delay power curve image, a first image is obtained. The image preprocessing includes grayscale conversion, such as... Figure 5 The converted grayscale image;

[0070] Step S3: By constructing the Hessian matrix of each pixel in the first image, the curve intensity and curve angle of each pixel in the first image are obtained. This is further combined with the first line point condition to obtain the first line point set, including:

[0071] Step S3-1: Calculate the second derivative of each pixel in an image with respect to x, the second derivative with respect to y, and the second derivative with respect to xy, and construct the Hessian matrix for each pixel in the first image based on this calculation.

[0072]

[0073] Where: H(x,y) is the Hessian matrix of the pixel, with the x-coordinate of the pixel being x and the y-coordinate being y, I xx Let I be the second derivative of the pixel with respect to x. yy Let I be the second derivative of the pixel with respect to y. xy Let x be the second derivative of the pixel with respect to x and y.

[0074] Step S3-2: Obtain the curve intensity and curve angle of each pixel in the image by calculating the eigenvalues ​​and eigenvectors of the Hessian matrix.

[0075] The curve intensity and curve angle of each pixel are:

[0076] μ(x,y)=max(λ1,λ2)

[0077]

[0078] Where μ(x,y) is the curve intensity of the pixel. Let λ1 and λ2 be the curve angles of the pixels, λ1 and λ2 be the eigenvalues ​​of the Hessian matrix, and N be the unit eigenvectors of the Hessian matrix. x N represents the x-axis component of the unit eigenvectors of the Hessian matrix. y Let y be the component of the unit eigenvector of the Hessian matrix along the y-axis.

[0079] Figure 6 and Figure 7 This section displays a heatmap of curve intensity for processed grayscale images and a vector diagram of curve intensity overlaid with curve direction for processed grayscale images. In the curve intensity heatmap, the colors in the color bar on the right side of the image represent the strength of the curve intensity. The unit of curve intensity in the image is decibels (dB). Pixels considered as curves in the image have high curve intensity, and the closer to the center point of the curve, the higher the curve intensity. The curve intensity of the noise part is significantly lower than that of the curve part by 5-10dB. In the vector diagram of curve intensity overlaid with curve direction, it can be seen that each pixel has a curve direction, and each pixel is represented by a vector arrow. A portion of the image is enlarged, showing that the curve direction indicates the exact direction of the current curve as the curve changes.

[0080] Step S3-3: Based on the eigenvalues ​​and eigenvectors of the Hessian matrix, select line points that meet the first line point condition;

[0081] The condition for the first line point is:

[0082] C={(x,y)∈I|(p x ,p y )∈[-0.5,0.5]×[-0.5,0.5]}

[0083] (p x ,p y )=(tN x ,tN y )

[0084] Where: C is the first line point condition, p xp is the x-coordinate of a sub-pixel. y Let y be the y-coordinate of a subpixel, t be a factor in the second-order Taylor expansion, and I be the first image.

[0085] Step S3-4: Based on the line points obtained from the screening, further select linear structure line points with bright foreground to construct the first line point set.

[0086] In steps S3-4, the desired line point matrix E for a bright foreground is:

[0087] E={(x,y)∈C|abs(λ1) <T1∧λ2<T2}

[0088] Where: abs(λ1) is the absolute value of the eigenvalue λ1 of the Hessian matrix, T1 and T2 are the line point structure thresholds, and λ1 and λ2 are the eigenvalues ​​of the Hessian matrix.

[0089] Figure 8 and Figure 9 The diagrams show the distribution of line points that meet the line point conditions and the distribution of line points that meet the line point structure thresholds, respectively. Figure 8 The background is a grayscale image, and each blue dot in the foreground represents a line point that meets the line point condition constraints. The line points in the image are mainly divided into three types: the first type is linear structure line points with a bright foreground, which are the line points we are looking for; the second type is linear structure line points with a dark foreground, which, like the first type, have a linear structure and generally exist between two adjacent curves with a bright foreground, and are considered noise; the third type is speckled structure line points with either a bright or dark foreground. These line points are also considered noise in curve detection and need to be eliminated. Combined with the curve intensity heatmap, it can be clearly observed that the curve intensity of these line points is significantly lower than that of linear structure line points. Figure 9 The demonstration shows that by setting the structural threshold appropriately, most noise can be eliminated while retaining the desired line points.

[0090] Step S4: Connect the line points based on the first set of line points to obtain the multipath component trajectory;

[0091] The connection methods include eight-neighbor search, distance extension search, and vanishing curve extension search. Eight-neighbor search is performed first. If eight-neighbor search fails, distance extension search is performed. If distance extension search fails, vanishing curve extension search is performed.

[0092] For a two-dimensional discrete image, each pixel has at most eight neighboring pixels. Eight-neighbor search includes:

[0093] Using the curve direction information of the line points, the curve angle is divided into eight equal parts. Each angle corresponds to three adjacent pixels in the eight-neighborhood. The next possible connecting line point belongs to these three adjacent pixels.

[0094] After obtaining three adjacent pixels, the best connecting line point is selected by combining the sub-pixel distance difference and angle difference. If none of the three adjacent pixels belong to the line points with bright foreground, the eight-neighborhood point search fails.

[0095] Figure 10 The graph shows the connection results for finding points in the eight neighborhoods. The different colored curves in the graph represent the different curves that have been traced. It can be seen that all the points are traversed and connected, but points that are clearly on the same curve will be treated as multiple curves.

[0096] Distance-extended point finding includes:

[0097] Using the current point position as a base, extend linearly a certain distance to obtain the coordinates of a circle's center;

[0098] Draw a circle at the center of the circle. The set of line points included inside the circle represents the line points that can be found by distance extension point search.

[0099] The desired connecting line point is selected by combining curve intensity and angle conditions. If the possible connecting line points within the circle are an empty set, the distance extension point search fails, and the selected desired connecting line point P is not determined. e for:

[0100] P e ={(x,y)∈E|Δd(E,r) <R∧min([AddCondition SA (E,s)]}

[0101] Where: r is the desired line point matrix, E is the desired line point matrix, and AddCondition is the key. SA Add a condition to the curve intensity-angle, where R is the radius of the extended circle and s is the current line point.

[0102] Figure 11 To improve the connection results of distance extension point finding, after introducing the distance extension point finding method, most curves are treated as a single curve, and the LosS path and the two main NLoS paths are completely traced. However, there are still broken connections or incorrect connections in the connection results at this time.

[0103] Vanishing curve extension point finding includes:

[0104] Calculate the sum of gray levels of the grayscale image and the defined geometric shape within a certain range, starting from the current line point;

[0105] Select the geometric region corresponding to the maximum sum of gray levels as the region where line points may exist, and perform condition calculations on the line points within this region to obtain the desired connecting line points.

[0106] Figure 12To extend the connection results of vanishing curves, after adding the vanishing curve extension point finding method, multiple vanishing curves can be tracked again. This point finding method can track curves that have vanished multiple times, as well as curves that have vanished due to intersection.

[0107] Step S5: Map the obtained multipath component trajectories to the channel physical parameter coordinate system to obtain the parameters of each multipath component. In the time delay domain, the parameters include time delay, power, and time (distance).

[0108] Figure 13 The result of mapping the trajectory to the original coordinate system shows that the main MPC trajectories have been tracked and the geometric consistency of the spatiotemporal evolution of MPC has been preserved, which effectively handles the blocking situation that past technologies could not handle. In addition, this method can also handle intersection, fading and other situations, and does not require pre-design of the MPC trajectory form, making it applicable to a wide range of scenarios.

[0109] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A parameter estimation method based on computer vision algorithms, characterized in that, include: Step S1: Obtain the cascaded delay power curve image after low-pass filtering; Step S2: After performing image preprocessing on the cascaded time delay power curve image, a first image is obtained, wherein the image preprocessing includes grayscale conversion; Step S3: Obtain the curve intensity and curve angle of each pixel in the first image by constructing the Hessian matrix of each pixel in the first image, and further combine the first line point condition to filter and obtain the first line point set; Step S4: Connect the line points based on the first set of line points to obtain the multipath component trajectory; Step S5: Map the obtained multipath component trajectories to the channel physical parameter coordinate system to obtain the parameters of each multipath component.

2. The parameter estimation method based on computer vision algorithm according to claim 1, characterized in that, Step S3 includes: Step S3-1: Calculate the relationship between each pixel in an image and its position. x The second derivatives of x, y, and xy are derived, and the Hessian matrix of each pixel in the first image is constructed based on these derivatives. in: Let be the Hessian matrix of the pixels, and let be the x-coordinate of each pixel. x The vertical axis is y , I xx For pixels about x The second derivative, I yy For pixels about y The second derivative, I xy For pixels about xy The second derivative; Step S3-2: Obtain the curve intensity and curve angle of each pixel in the image by calculating the eigenvalues ​​and eigenvectors of the Hessian matrix. Step S3-3: Based on the eigenvalues ​​and eigenvectors of the Hessian matrix, select line points that meet the first line point condition; Step S3-4: Based on the line points obtained from the screening, further select linear structure line points with bright foreground to construct the first line point set.

3. The parameter estimation method based on computer vision algorithm according to claim 2, characterized in that, The curve intensity and curve angle of each pixel are: in: The curve intensity of a pixel. The curve angle of a pixel. These are the eigenvalues ​​of the Hessian matrix. N Let be the unit eigenvector of the Hessian matrix. N x The unit eigenvectors of the Hessian matrix x Components in the axial direction, N y The unit eigenvectors of the Hessian matrix y Components in the axial direction.

4. The parameter estimation method based on computer vision algorithm according to claim 3, characterized in that, The first line point condition is: in: As the condition for the first line point, p x sub-pixel x coordinate, p y sub-pixel y coordinate, t A factor in a second-order Taylor expansion. I This is the first image.

5. The parameter estimation method based on computer vision algorithm according to claim 3, characterized in that, In steps S3-4, the foreground is the desired line-point matrix that is bright. E for: in: Eigenvalues ​​of the Hessian matrix The absolute value, , For line-point structure threshold, denoted as eigenvalues ​​of the Hessian matrix.

6. The parameter estimation method based on computer vision algorithm according to claim 3, characterized in that, The connection methods in step S4 include eight-neighbor search, distance extension search, and vanishing curve extension search. Eight-neighbor search is performed first. If eight-neighbor search fails, distance extension search is performed. If distance extension search fails, vanishing curve extension search is performed.

7. The parameter estimation method based on computer vision algorithm according to claim 6, characterized in that, The eight-neighborhood point search includes: Using the curve direction information of the line points, the curve angle is divided into eight equal parts. Each angle corresponds to three adjacent pixels in the eight-neighborhood. The next possible connecting line point belongs to these three adjacent pixels. After obtaining three adjacent pixels, the best connecting line point is selected by combining the sub-pixel distance difference and angle difference. If none of the three adjacent pixels belong to the line points with bright foreground, the eight-neighborhood point search fails.

8. The parameter estimation method based on computer vision algorithm according to claim 6, characterized in that, The distance extension point search includes: Using the current point position as a base, extend linearly a certain distance to obtain the coordinates of a circle's center; Draw a circle at the center of the circle. The set of line points included inside the circle represents the line points that can be found by distance extension point search. The desired connecting line points are selected by combining curve intensity and angle conditions. If the possible connecting line points within the circle are an empty set, the distance extension point search fails, and the selected desired connecting line points are not considered. P e for: in: r for, E Let be the desired line point matrix. Add conditions to the curve intensity-angle. R To extend the radius of the circle, s This is the current line point.

9. The parameter estimation method based on computer vision algorithm according to claim 6, characterized in that, The vanishing curve extension point search includes: Calculate the sum of gray levels of the grayscale image and the defined geometric shape within a certain range, starting from the current line point; Select the geometric region corresponding to the maximum sum of gray levels as the region where line points may exist, and perform condition calculations on the line points within this region to obtain the desired connecting line points.

10. A parameter estimation device based on a computer vision algorithm, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-9.

11. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method as described in any one of claims 1-9.

Images