A method and system for tracking individual emitters based on optical flow

Abstract

The application belongs to the technical field of communication signal detection and recognition, and discloses a radiation source individual tracking method and system based on an optical flow method, which comprises the following steps: filling a one-dimensional time domain waveform of a radiation source signal into a two-dimensional matrix in a spiral mode to obtain multiple signal images; constructing an optical flow equation between pixel points of a first signal image and a second signal image across time periods; based on the optical flow equation, constructing an over-determined equation according to a local smoothness constraint, and solving an optical flow vector by using a least square method; generating a Gaussian pyramid of the signal image in a Gaussian scale space, and sequentially calculating and superimposing the optical flow vector from the top of the Gaussian pyramid; and calculating a displacement vector of a pixel in the image in time sequence according to the optical flow vector, and matching the signal images across time periods. The application combines the optical flow method and the Gaussian pyramid, mines the internal correlation of multiple signal images in different time scales, matches the signal images across time periods, and improves the accuracy and efficiency of radiation source recognition and tracking.

Description

Technical Field

[0001] This invention relates to the field of communication signal detection and identification technology, and in particular to a method and system for tracking individual radiation sources based on optical flow. Background Technology

[0002] Research indicates that device defects or variations result in unique properties in different transmitters, making the details of their radiated signals (i.e., signal "fingerprints") unique. Researchers characterize the nonlinear behavior of these signals by modeling the transmitters and use the model parameters as identifying features of individual radiation sources. Related research is divided into two levels: modeling based on transmitter devices and modeling based on the overall transmitter system.

[0003] At the transmitter component level: Shortwave communication transmitters contain various components, such as digital-to-analog converters (DACs), power amplifiers (PAs), modulators, filters, and antennas, all of which exhibit certain nonlinear characteristics during operation. In 2011, ACPolak et al. used a Brownian bridge model to analyze the nonlinear errors introduced by the DAC; in 2014, T. Lee et al., and in 2019, Liu Cheng et al., respectively studied the nonlinear distortion of amplifiers from different dimensions. In 2015, ACPolak et al. characterized component properties by the in-band signal distortion and spectral regeneration caused by the nonlinear components of the device's power amplifier. In 2019, S. Balakrishnan et al. demonstrated that manufacturing tolerances of antenna arrays in millimeter-wave devices result in different devices having different beam patterns. The aforementioned literature all utilizes the characterization of key components at the signal level to complete the individual identification of the radiation source, but does not cover the overall characteristics of the transmitter.

[0004] At the transmitter system level: In 2009, T. Carroll et al. used phase space to achieve joint reconstruction of the nonlinear properties of various devices. In 2019, Sun Zhi et al. derived the mathematical expression for the physical process from modulation to power amplification, demonstrating the uniqueness of the hardware and signal processing in the radiation source transmitter. In 2020, Shi et al. used memory polynomials to model the behavior of sonar transmitters and used the power spectrum estimate of the output signal as a signal fingerprint to identify the transmitter.

[0005] Currently, transmitter modeling has been achieved at the device and system levels, but the changes in the transmitter have not been considered from a time perspective. The temporal variation pattern of the signal fingerprint is still unclear, posing a challenge to the tracking of individual radiation sources across time periods. Summary of the Invention

[0006] The transmitter of a radiation source often changes due to internal and external factors, causing variations in its signal fingerprint. When the time interval between two communications from a target is long, excessive changes can lead to the same individual being misidentified as different individuals. This invention addresses this problem by proposing a radiation source individual tracking method and system based on optical flow. By understanding the time-varying characteristics of the transmitter's radiation signal and overcoming the influence of internal and external factors on subtle signal features, accurate tracking of individual radiation sources is ensured.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention proposes a method for tracking individual radiation sources based on optical flow, comprising:

[0009] Step 1: Fill the one-dimensional time-domain waveform of the radiation source signal into a two-dimensional matrix in a spiral manner to obtain multiple frames of signal images;

[0010] Step 2: For two signal images spanning two time periods, the first signal image and the second signal image, the grayscale value of the same target in the first signal image and the second signal image is constant, but the position on the image is different; construct the optical flow equation between the pixels of the first signal image and the second signal image;

[0011] Step 3: Based on the optical flow equation, construct an overdetermined equation according to the local smoothness constraint, and solve for the optical flow vector using the least squares method;

[0012] Step 4: Generate a Gaussian pyramid of the signal image in Gaussian scale space, and calculate and superimpose the optical flow vector sequentially from the top of the Gaussian pyramid downwards;

[0013] Step 5: Calculate the displacement vector of pixels in the time-series image based on the optical flow vector, and match the signal image across time periods.

[0014] Furthermore, the optical flow equation between the pixels of the first and second signal images is constructed as follows:

[0015] Since the grayscale value of the same target in the first signal image and the second signal image is constant, the following relationship is obtained:

[0016] I(x,y,l)=I(x+δx,y+δy,r)

[0017] Expanding the above equation using Taylor and ignoring higher-order terms, we obtain the optical flow equation:

[0018] I x δx+I y δy+I △ =0

[0019] Where I(x,y,l) represents the gray value of pixel (x,y) in the first signal image, l refers to the first signal image, and I(x+δx,y+δy,r) represents the gray value of pixel (x+δx,y+δy) in the second signal image that matches pixel (x,y), r refers to the second signal image. I △ δx and δy represent the grayscale difference between the matching pixels in the first and second signal images, respectively, and the offsets of the same target pixel in the horizontal and vertical directions of the first and second signal images.

[0020] Furthermore, the overdetermined equation is:

[0021] I x (X i )·δx+I y (X i )·δy+I △ (X i )=0,X i ∈Ω(i=0,1,…,n-1)

[0022] Where X i Let Ω be the pixel position coordinates, and let Ω represent a local region containing (x, y) with n pixels.

[0023] Furthermore, in step 3, the least squares method is used to solve for the optical flow vector [δx, δy]. T .

[0024] Furthermore, the Gaussian pyramid for generating the signal image in Gaussian scale space includes:

[0025] The original image of the signal image is coupled into the first layer of the Gaussian pyramid. The Gaussian blur is removed by deconvolution and interpolation to generate the 0th layer image. The 2nd layer image is generated by downsampling and Gaussian blurring of the 1st layer image. The 3rd layer image is generated by downsampling and Gaussian blurring of the 2nd layer image.

[0026] This invention also proposes a radiation source individual tracking system based on optical flow, comprising:

[0027] The signal conversion module is used to fill the one-dimensional time-domain waveform of the radiation source signal into a two-dimensional matrix in a spiral manner to obtain multiple frames of signal images;

[0028] The optical flow equation construction module is used to construct optical flow equations between pixels in two signal images spanning two time periods, namely a first signal image and a second signal image, where the grayscale value of the same target is constant in the first signal image and the second signal image but the position on the image is different.

[0029] The overdetermined equation construction module is used to construct overdetermined equations based on the optical flow equations and local smoothness constraints, and to solve the optical flow vectors using the least squares method.

[0030] The Gaussian pyramid generation module is used to generate a Gaussian pyramid of a signal image in Gaussian scale space. It calculates and superimposes optical flow vectors sequentially from the top of the Gaussian pyramid downwards.

[0031] The matching relationship confirmation module is used to calculate the displacement vector of pixels in the time-series image based on the optical flow vector and match the signal image across time periods.

[0032] Furthermore, the optical flow equation between the pixels of the first and second signal images is constructed as follows:

[0033] Since the grayscale value of the same target in the first signal image and the second signal image is constant, the following relationship is obtained:

[0034] I(x,y,l)=I(x+δx,y+δy,r)

[0035] Expanding the above equation using Taylor and ignoring higher-order terms, we obtain the optical flow equation:

[0036] I x δx+I y δy+I △ =0

[0037] Where I(x,y,l) represents the gray value of pixel (x,y) in the first signal image, and I(x+δx,y+δy,r) represents the gray value of pixel (x+δx,y+δy) in the second signal image that matches pixel (x,y). I △ δx and δy represent the grayscale difference between the matching pixels in the first and second signal images, respectively, and the offsets of the same target pixel in the horizontal and vertical directions of the first and second signal images.

[0038] Furthermore, the overdetermined equation is:

[0039] I x (X i )·δx+I y (X i )·δy+I △ (X i )=0,X i ∈Ω(i=0,1,...,n-1)

[0040] Where X i Let Ω be the pixel position coordinates, and let Ω represent a local region containing (x, y) with n pixels.

[0041] Furthermore, in the overdetermined equation construction module, the least squares method is used to solve for the optical flow vector [δx, δy]. T .

[0042] Furthermore, the Gaussian pyramid for generating the signal image in Gaussian scale space includes:

[0043] The original image of the signal image is coupled into the first layer of the Gaussian pyramid. The Gaussian blur is removed by deconvolution and interpolation to generate the 0th layer image. The 2nd layer image is generated by downsampling and Gaussian blurring of the 1st layer image. The 3rd layer image is generated by downsampling and Gaussian blurring of the 2nd layer image.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] This invention proposes a radiation source individual tracking method and system based on optical flow. By combining optical flow and Gaussian pyramid, it can mine the inherent correlation of multi-frame signals at different time scales, match signals across time periods, alleviate the constraints of transmitter attribute changes of radiation sources on individual identification algorithms, and help achieve optimal matching relationships in the frequency and spatial domains of signals, thereby improving the accuracy and efficiency of radiation source identification and tracking. Attached Figure Description

[0046] Figure 1 This is a basic flowchart of a radiation source individual tracking method based on optical flow, according to an embodiment of the present invention.

[0047] Figure 2 A schematic diagram of signal image matching under the Gaussian pyramid LK optical flow method provided in an embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of the architecture of an individual radiation source tracking system based on optical flow method according to an embodiment of the present invention. Detailed Implementation

[0049] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments:

[0050] To accurately track individual radiation sources, this invention utilizes the basic idea of ​​the Lucas-Kanade (LK) optical flow method to describe the temporal changes in the transmitter signal of the radiation source. Optical flow refers to the instantaneous velocity of pixels moving on the observation imaging plane of a spatially moving object, primarily used for the correlation of moving objects in video sequence processing. For example... Figure 1 As shown, a method for tracking individual radiation sources based on optical flow is proposed, which specifically includes the following steps:

[0051] S101: Fill the one-dimensional time-domain waveform of the radiation source signal into a two-dimensional matrix in a spiral manner to obtain multiple frames of signal images;

[0052] S102: For two signal images spanning two time periods, the first signal image and the second signal image, the gray value of the same target in the first signal image and the second signal image is constant, but the position on the image is different; construct the optical flow equation between the pixels of the first signal image and the second signal image;

[0053] S103: Based on the optical flow equation, an overdetermined equation is constructed according to the local smoothness constraint, and the optical flow vector is solved by the least squares method.

[0054] S104: Generate a Gaussian pyramid for the signal image in Gaussian scale space, and calculate and superimpose optical flow vectors sequentially from the top of the Gaussian pyramid downwards;

[0055] S105: Calculate the displacement vector of pixels in the time-series image based on the optical flow vector, and match the signal image across time periods.

[0056] As one possible implementation, in S101, the present invention fills the one-dimensional time-domain waveform of the radiation source signal into a two-dimensional matrix in a spiral manner. The number of sampling points of a single frame signal is m*n, which can be converted into a signal image of size m*n, where m and n are the number of rows and columns of the image, respectively.

[0057] Furthermore, S102 includes:

[0058] For two signal images L (the first image) and R (the second image), the grayscale value of pixel (x,y) in image L is I(x,y,l), and the grayscale value of the matching pixel (x+δx,y+δy) in image R is I(x+δx,y+δy,r). l and r represent the images in which the pixel resides; l corresponds to signal image L, and r corresponds to image R. Based on the assumption of constant image grayscale in optical flow, we have:

[0059] I(x,y,l)=I(x+δx,y+δy,r) (1)

[0060] Expanding the above equation using Taylor series and ignoring higher-order terms, we get:

[0061] I x δx+I y δy+I △ =0 (2)

[0062] The fundamental equation of optical flow is obtained, where I △Let δx and δy be the gray value difference of the matched pixels on images L and R, respectively, and let δx and δy be the offsets of the same target pixel in the horizontal and vertical directions on the first and second signal images, respectively.

[0063] Furthermore, S103 includes:

[0064] An overdetermined equation is constructed based on local smoothness constraints, and the optical flow vector [δx, δy] is solved using the least squares method. T Assume that the optical flow of the image remains constant within a small local region Ω containing (x, y) (Ω contains n pixels, X...). i (where is the pixel position coordinate), then the overdetermined equation is:

[0065] I x (X i )·δx+I y (X i )·δy+I △ (X i )=0,X i ∈Ω(i=0,1,...,n-1) (3)

[0066] It can be simply remembered as:

[0067]

[0068] Where the coefficient matrix A and the constant term b are respectively:

[0069]

[0070] Furthermore, S104 includes:

[0071] To achieve feature extraction and prediction at different signal scales, a feature pyramid is generated in the Gaussian scale space, such as... Figure 2 As shown, the LK optical flow method is performed sequentially from the top of the pyramid downwards, and the optical flow vectors are superimposed. In this process, the original image of the signal image is coupled to the first layer image of the Gaussian pyramid. The Gaussian blur is removed by deconvolution and interpolation to generate the 0th layer image. The 2nd layer image is generated by downsampling and Gaussian blurring of the 1st layer image. The 3rd layer image is generated by downsampling and Gaussian blurring of the 2nd layer image.

[0072] Furthermore, S105 includes:

[0073] The signal matching relationship is confirmed by using the optical flow vector of the signal image features obtained from the feature pyramid, and then calculating the displacement vector of the pixels in the temporal image based on the optical flow vector. Finally, the signal image is transformed from two dimensions into a one-dimensional vector to complete the signal matching.

[0074] This invention utilizes the basic ideas of pyramids and LK optical flow to describe the temporal changes of the transmitter signal of a radiation source, which can accurately track individual radiation sources.

[0075] Based on the above embodiments, such as Figure 3 As shown, this invention also proposes a radiation source individual tracking system based on optical flow, comprising:

[0076] The signal conversion module is used to fill the one-dimensional time-domain waveform of the radiation source signal into a two-dimensional matrix in a spiral manner to obtain multiple frames of signal images;

[0077] The optical flow equation construction module is used to construct optical flow equations between pixels in two signal images spanning two time periods, namely a first signal image and a second signal image, where the grayscale value of the same target is constant in the first signal image and the second signal image but the position on the image is different.

[0078] The overdetermined equation construction module is used to construct overdetermined equations based on the optical flow equations and local smoothness constraints, and to solve the optical flow vectors using the least squares method.

[0079] The Gaussian pyramid generation module is used to generate a Gaussian pyramid of a signal image in Gaussian scale space. It calculates and superimposes optical flow vectors sequentially from the top of the Gaussian pyramid downwards.

[0080] The matching relationship confirmation module is used to calculate the displacement vector of pixels in the time-series image based on the optical flow vector and match the signal image across time periods.

[0081] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for tracking individual radiation sources based on optical flow, characterized in that, include: Step 1: Fill the one-dimensional time-domain waveform of the radiation source signal into a two-dimensional matrix in a spiral manner to obtain multiple frames of signal images; Step 2: For two signal images spanning different time periods, the first signal image and the second signal image, the grayscale value of the same target in the first signal image and the second signal image is constant, but the position on the image is different; construct the optical flow equation between the pixels of the first signal image and the second signal image; Step 3: Based on the optical flow equation, construct an overdetermined equation according to the local smoothness constraint, and solve for the optical flow vector using the least squares method; Step 4: Generate a Gaussian pyramid of the signal image in Gaussian scale space, and calculate and superimpose the optical flow vector sequentially from the top of the Gaussian pyramid downwards; Step 5: Calculate the displacement vector of pixels in the time-series image based on the optical flow vector, and match the signal image across time periods.

2. The method for tracking individual radiation sources based on optical flow according to claim 1, characterized in that, The optical flow equation between pixels in the first and second signal images is constructed as follows: Since the grayscale value of the same target in the first signal image and the second signal image is constant, the following relationship is obtained: I(x,y,l)=I(x+δx,y+δy,r) Expanding the above equation using Taylor and ignoring higher-order terms, we obtain the optical flow equation: I x δx+I y δy+I △ =0 Where I(x,y,l) represents the gray value of pixel (x,y) in the first signal image, l refers to the first signal image, and I(x+δx,y+δy,r) represents the gray value of pixel (x+δx,y+δy) in the second signal image that matches pixel (x,y), r refers to the second signal image. I △ δx and δy represent the grayscale difference between the matching pixels in the first and second signal images, respectively, and the offsets of the same target pixel in the horizontal and vertical directions of the first and second signal images.

3. The method for tracking individual radiation sources based on optical flow according to claim 2, characterized in that, The overdetermined equation is: I x (X i )·δx+I y (X i )·δy+I △ (X i )=0,X i ∈Ω(i=0,1,…,n-1) Where X i Let Ω be the pixel position coordinates, and let Ω represent a local region containing (x, y) with n pixels.

4. The method for tracking individual radiation sources based on optical flow according to claim 3, characterized in that, In step 3, the least squares method is used to solve for the optical flow vector [δx, δy]. T .

5. The method for tracking individual radiation sources based on optical flow according to claim 1, characterized in that, The Gaussian pyramid for generating signal images in Gaussian scale space includes: The original image of the signal image is coupled into the first layer of the Gaussian pyramid. The Gaussian blur is removed by deconvolution and interpolation to generate the 0th layer image. The 2nd layer image is generated by downsampling and Gaussian blurring of the 1st layer image. The 3rd layer image is generated by downsampling and Gaussian blurring of the 2nd layer image.

6. A radiation source individual tracking system based on optical flow method, characterized in that, include: The signal conversion module is used to fill the one-dimensional time-domain waveform of the radiation source signal into a two-dimensional matrix in a spiral manner to obtain multiple frames of signal images; The optical flow equation construction module is used to construct optical flow equations between pixels in two signal images spanning two time periods, namely a first signal image and a second signal image, where the grayscale value of the same target is constant in the first signal image and the second signal image but the position on the image is different. The overdetermined equation construction module is used to construct overdetermined equations based on the optical flow equations and local smoothness constraints, and to solve the optical flow vectors using the least squares method. The Gaussian pyramid generation module is used to generate a Gaussian pyramid of a signal image in Gaussian scale space. It calculates and superimposes optical flow vectors sequentially from the top of the Gaussian pyramid downwards. The matching relationship confirmation module is used to calculate the displacement vector of pixels in the time-series image based on the optical flow vector and match the signal image across time periods.

7. A radiation source individual tracking system based on optical flow method according to claim 6, characterized in that, The optical flow equation between pixels in the first and second signal images is constructed as follows: Since the grayscale value of the same target in the first signal image and the second signal image is constant, the following relationship is obtained: I(x,y,l)=I(x+δx,y+δy,r) Expanding the above equation using Taylor and ignoring higher-order terms, we obtain the optical flow equation: I x δx+I y δy+I △ =0 Where I(x,y,l) represents the gray value of pixel (x,y) in the first signal image, and I(x+δx,y+δy,r) represents the gray value of pixel (x+δx,y+δy) in the second signal image that matches pixel (x,y). I △ δx and δy represent the grayscale difference between the matching pixels in the first and second signal images, respectively, and the offsets of the same target pixel in the horizontal and vertical directions of the first and second signal images.

8. A radiation source individual tracking system based on optical flow method according to claim 7, characterized in that, The overdetermined equation is: I x (X i )·δx+I y (X i )·δy+I △ (X i )=0,X i ∈Ω(i=0,1,…,n-1) Where X i Let Ω be the pixel position coordinates, and let Ω represent a local region containing (x, y) with n pixels.

9. A radiation source individual tracking system based on optical flow method according to claim 8, characterized in that, In the overdetermined equation construction module, the least squares method is used to solve for the optical flow vector [δx, δy]. T .

10. A radiation source individual tracking system based on optical flow method according to claim 6, characterized in that, The Gaussian pyramid for generating signal images in Gaussian scale space includes: The original image of the signal image is coupled into the first layer of the Gaussian pyramid. The Gaussian blur is removed by deconvolution and interpolation to generate the 0th layer image. The 2nd layer image is generated by downsampling and Gaussian blurring of the 1st layer image. The 3rd layer image is generated by downsampling and Gaussian blurring of the 2nd layer image.

Images