An imaging method for tracking weak signal small targets in complex systems

Abstract

The application provides an imaging method for tracking a weak signal small target in a complex system, comprising: collecting an image sequence of a target region by using an imaging device; reconstructing motion contrast information based on a non-stationary signal time-frequency transform according to the image sequence to obtain time distribution information of a motion position of the target in the image sequence; and determining actual time distribution information of the motion position of the target based on known imaging parameters and the time distribution information of the motion position of the target in the image sequence. The method of the application introduces motion contrast imaging into the tracking imaging of a weak signal target in a complex background, and realizes high-sensitivity tracking imaging of the target. According to the time-varying characteristics of illumination and background, a motion contrast imaging method based on a non-stationary signal time-frequency transform is developed, the influence of illumination intensity variation and background random variation is effectively eliminated, and the method can be applied to the fields of visible light or infrared.

Description

Technical Field

[0001] This invention relates to an imaging method for tracking small targets with weak signals in complex systems, particularly for imaging targets moving at long distances, in low visibility, and on free trajectories. Background Technology

[0002] Nature contains a variety of complex systems, often comprising intricate components and relative motions. Typically, the motion of a target within a complex system overlaps with the motion of the background, inevitably introducing motion artifacts into direct imaging. For moving target detection, a simple and effective solution is to distinguish the target of interest from dynamic background noise by detecting pixel information within adjacent frames, thereby eliminating background interference with target tracking. Typical examples of this approach include spatiotemporal multi-channel clustering, inter-frame difference methods, optical flow methods, and combinations of the latter two. These algorithms have achieved good results in tracking large targets with strong signals. However, when the target is as small as a few pixels and the signal is weak, complex backgrounds and detector noise reduce image contrast, rendering these methods ineffective.

[0003] In recent years, two main categories of low-visibility target detection algorithms have been developed to address this problem: single-frame detection and image sequence detection.

[0004] In a typical single-frame detection algorithm, small targets are detected by judging whether the local gray-level envelope surface of the image has the features of a two-dimensional Gaussian function surface. However, for weak signals, a single frame image contains too little signal energy, so the disadvantage of the single-frame detection algorithm is its poor noise resistance.

[0005] In image sequence detection algorithms, Gaussian difference filters and clustering methods are used to separate targets from complex backgrounds; a method combining high-pass filtering and adaptive threshold segmentation is used to detect fast-moving targets with low signal-to-noise ratios; background suppression based on two-dimensional least mean square filters is combined with target enhancement based on neighborhood gray-level difference to improve the detection capability of airborne imaging equipment for low-visibility targets in complex backgrounds; wavelet transform denoising and a point moving target detection algorithm based on kernel theory are combined to enhance the detection capability of low-visibility targets. Based on the multi-frame detection concept, a series of Track Before Detect (TBD) algorithms, including Hough transform, dynamic programming, multi-level hypothesis testing, and particle filtering, have been developed.

[0006] For the aforementioned algorithms such as single-frame detection, the combination of Gaussian filter and clustering, the combination of Gaussian filter and adaptive threshold segmentation, the combination of background suppression processing of two-dimensional minimum mean square filter and target enhancement based on neighborhood gray-level difference, the combination of wavelet transform denoising and point moving target detection algorithm based on kernel theory, and the pre-detection tracking based on Hough transform, dynamic programming, multi-level hypothesis testing and particle filtering, these algorithms all work based on gray-level images in the spatiotemporal domain, have weak noise resistance, and are not ideal when detecting highly maneuvering targets.

[0007] The pre-detection tracking algorithms mainly include five types: Hough transform, dynamic programming, multi-level hypothesis testing, particle filtering, and three-dimensional matched filtering. Hough transform, dynamic programming, multi-level hypothesis testing, and particle filtering operate in the spatial domain, while three-dimensional matched filtering operates in the frequency domain. Pre-detection tracking algorithms based on three-dimensional matched filters have stronger noise resistance because they operate in the Fourier frequency domain, but they can only track uniform linear targets with known velocities, and the velocity window is very narrow. Once velocity mismatch occurs, it is difficult to obtain effective imaging results.

[0008] In summary, the existing methods described above are insufficient for tracking and imaging low-visibility targets whose direction and speed of motion are not subject to regular changes.

[0009] X-ray motion contrast imaging can achieve high-sensitivity imaging of low-visibility moving targets in the X-ray band (see references [1] to [6]). The principle of X-ray motion contrast imaging is as follows: the gray-scale change of a certain point in the image reflects the motion of the corresponding point in the imaging area. When an object passes through a certain point, it will cause a sharp rise or fall in the gray-scale of the corresponding position. The gray-scale fluctuation caused by weak signal small targets is extremely small. From the perspective of the spatiotemporal domain, this is no different from the intensity fluctuation of the background. Therefore, conventional methods are difficult to track such signals. The frequency domain has extremely high sensitivity to the gray-scale changes caused by the target motion. By converting the image gray-scale to the frequency domain, the spectral information of the corresponding point can be restored, thereby separating the moving target, high-frequency noise and complex background, and achieving high-sensitivity imaging of weak signal small targets.

[0010] The specific process of X-ray motion contrast imaging includes: acquiring image sequences at a certain frame rate; performing time-frequency transformation on the image sequences to convert the time-domain signal to the frequency-domain space; using a bandpass filter to filter out the target frequency components from the spectrum; and converting the filtered target components into real space to complete the imaging.

[0011] However, X-ray imaging differs from outdoor visible light imaging in two ways: First, X-ray imaging is typically performed indoors with a stable light source, resulting in image sequences with similar brightness. Second, the background in X-ray imaging is usually relatively monotonous and uncomplicated. However, when tracking small targets in complex backgrounds (such as forests) under natural light illumination, the overall brightness of the image changes with the lighting conditions. Furthermore, the background is constantly and irregularly changing (e.g., leaves sway randomly in the wind, and differences in angle cause random changes in their brightness). These two factors exacerbate the aliasing of the target and background spectra, leading to severe artifacts in traditional X-ray motion contrast imaging under these conditions, making it unsuitable for direct outdoor target tracking.

[0012] References:

[0013] [1] Wang Feixiang, Research on first-order photon correlation imaging in X-ray spacetime [D]. University of Chinese Academy of Sciences (Shanghai Institute of Applied Physics, Chinese Academy of Sciences), 2019.

[0014] [2]Wang F, Zhou P, Li K, et al. Sensitive imaging of intact microvessels in vivo with synchrotron radiation[J]. IUCrJ, 2020, 7(5):793-802.

[0015] [3] Li Ke, Research on X-ray Imaging and Its Application in Complex Systems of Low-Z Materials [D]. University of Chinese Academy of Sciences, Shanghai Institute of Applied Physics, Chinese Academy of Sciences, 2021. DOI:10.27585 / d.cnki.gkshs.2021.000055.

[0016] [4] Xiao Tiqiao, Wang Feixiang, Li Ke, Xu Mingwei, Ju Xiaolu, Motion Contrast X-ray Imaging and Its Applications [J]. Acta Optica Sinica, 2022, 42(11):11-27.

[0017] [5] Ju Xiaolu, Li Ke, Yu Fucheng, Xu Mingwei, Deng Biao, Li Bin, Xiao Tiqiao. Motion contrast X-ray imaging of electrochemical reaction process in electrolytic cell [J]. Acta Physica Sinica, 2022, 71(14):98-107.

[0018] [6] Ju Xiaolu. Study on efficient and sensitive X-ray imaging of distribution and migration of metal elements [D]. University of Chinese Academy of Sciences (Shanghai Institute of Applied Physics, Chinese Academy of Sciences), 2022. DOI:10.27585 / d.cnki.gkshs.2022.000052. Summary of the Invention

[0019] This invention provides an imaging method for tracking small targets with weak signals in complex systems, thereby improving the sensitivity of target tracking imaging in the visible or infrared fields.

[0020] To achieve the above objectives, the present invention provides an imaging method for tracking small targets with weak signals in complex systems, comprising:

[0021] S1: Use an imaging device to capture images of the target area to obtain an image sequence;

[0022] S2: Based on the image sequence, reconstruct the motion contrast information based on the time-frequency transformation of non-stationary signals to obtain the motion position time distribution information of the target in the image sequence. The motion position time distribution information refers to the position of the target in the image of the image sequence at each time point.

[0023] S3: Based on the known imaging parameters, determine the actual motion position time distribution information of the target according to the motion position time distribution information of the target in the image sequence.

[0024] Step S1 specifically includes:

[0025] S11: Select a target area that can contain the target's motion trajectory, and install an imaging device at a location aligned with the target area for data acquisition;

[0026] S12: Set the focal length and aperture of the imaging device so that the imaging device can capture clear images, and set the frame rate of the imaging device as needed;

[0027] S13: Start the imaging equipment, aim at the target area to start shooting, and continuously record the image sequence.

[0028] Step S2 specifically includes:

[0029] Step S21: Perform time-frequency transformation on the grayscale signal of the image in the image sequence to obtain the spectrum at different positions, and further filter the spectrum to obtain the phase map of motion contrast imaging and the motion trajectory of the target.

[0030] Step S22: The phase map of motion contrast imaging obtained in step S21 and the motion trajectory of the target are fused to obtain the target motion trajectory containing time information, which serves as the time distribution information of the target's motion position in the image sequence.

[0031] Step S21 specifically includes:

[0032] S211: Perform time-frequency transformation on the grayscale signals at all locations in the image to obtain the spectrum at different locations in the image;

[0033] S212: The spectrum is filtered by a first bandpass filter, the passband of which is the spectral range with the highest contrast in the spectrum, to obtain the amplitude map of motion contrast imaging; and the spectrum is filtered by a second bandpass filter, the passband of which is a single specific frequency, to obtain the phase map of motion contrast imaging.

[0034] S213: Determine the target position based on the linear trajectory on the amplitude map of the motion contrast imaging obtained in step S212. Obtain its spectrum by performing non-stationary signal time-frequency transformation on the gray signal at the target position, and then determine the passband of the third bandpass filter. The passband of the third bandpass filter covers the frequency range with the highest target frequency contrast.

[0035] S214: Select the third bandpass filter to filter the spectrum obtained in step S211 to obtain the target's motion trajectory.

[0036] In step S211, the time-frequency transformation methods used for the time-frequency transformation include Fourier transform, short-time Fourier transform, Gabor transform, synchronous compression transform, Laplace transform, Z-transform, Wigner-Ville distribution, Hilbert-Huang transform, and wavelet transform.

[0037] GPU parallel computing is used to perform time-frequency transformation on image sequences, thereby accelerating the reconstruction of motion contrast in image sequences.

[0038] The target's motion position time distribution information in the image sequence includes the length of the target's motion trajectory in the image and the corresponding time interval, and the target's actual motion position time distribution information includes the target's actual motion direction and the target's actual average speed.

[0039] The known imaging parameters include the lens focal length of the imaging device, the shooting frame rate of the imaging device, and the pixel size of the imaging device; the known imaging parameters also include the imaging distance of the imaging device; or, step S3 specifically includes: determining the imaging distance of the imaging device based on the initial value of the actual average speed of the target, and then determining the actual motion position time distribution information of the target based on the imaging distance of the imaging device and the known imaging parameters.

[0040] The actual average rate of the target for:

[0041]

[0042] in, The actual average speed of the target is represented by ; n represents the length of the target's trajectory in the image, in pixels; a represents the pixel size of the imaging device; L represents the imaging distance of the imaging device; and f represents the focal length of the lens of the imaging device.

[0043] The imaging distance L of the imaging device is:

[0044]

[0045] in, The actual average speed of the target is represented by ; n represents the length of the target's trajectory in the image, in pixels; a represents the pixel size of the imaging device; L represents the imaging distance of the imaging device; and f represents the focal length of the lens of the imaging device.

[0046] This invention introduces motion contrast imaging into tracking imaging of small targets with weak signals against complex backgrounds. Based on the characteristics of illumination and background changes over time under natural conditions, a motion contrast imaging method based on the time-frequency transformation of non-stationary signals is developed. This effectively eliminates the influence of changes in illumination intensity and random background variations, improving the sensitivity of target tracking imaging. It can be applied to the visible light or infrared fields, achieving high-sensitivity tracking imaging of small, freely moving targets with low visibility where effective contrast cannot be obtained through direct imaging. Attached Figure Description

[0047] Figure 1 This is a flowchart of an imaging method for tracking small targets with weak signals in complex systems, according to an embodiment of the present invention.

[0048] Figure 2 This is a schematic diagram of the imaging device and the target area. Detailed Implementation

[0049] The present invention will be further described below with reference to specific embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0050] According to the principle of the motion contrast tracing method of the present invention, only tens to hundreds of consecutive time-series images need to be acquired. The data acquisition device is an imaging device with time-series frame-frequency acquisition capability. The specific steps of the imaging method for tracking small targets with weak signals in complex systems are as follows:

[0051] Step S1: Use an imaging device to capture images of the target area to obtain an image sequence;

[0052] Step S1 specifically includes:

[0053] Step S11: Select a target area that can contain the target's motion trajectory, and install an imaging device at the location aligned with the target area for data acquisition;

[0054] Step S12: Set the focal length and aperture of the imaging device to enable the imaging device to capture clear images, and set the frame rate of the imaging device as needed;

[0055] There are no special requirements for shooting parameters such as focal length and aperture during data acquisition, as long as the image is relatively clear. The higher the frame rate, the better the tracking effect.

[0056] Step S13: Start the imaging device, aim at the target area and start taking pictures, continuously recording the image sequence.

[0057] In this embodiment, the number of images in the sequence is more than 20, and there are no special requirements for the shooting time.

[0058] Step S2: Based on the image sequence, reconstruct the motion contrast information based on the time-frequency transformation of the non-stationary signal to obtain the time distribution information of the target's motion position in the image sequence;

[0059] Among them, the motion position time distribution information refers to the position of the target in the image sequence at each time point, which includes the target's motion trajectory and time information.

[0060] Step S2 specifically includes:

[0061] Step S21: Perform time-frequency transformation on the image sequence to obtain the spectrum at different locations, and further filter the spectrum to obtain the phase map of motion contrast imaging and the motion trajectory of the target;

[0062] Step S21 specifically includes:

[0063] Step S211: Perform time-frequency transformation on the grayscale signals of all locations in the image sequence to obtain the spectrum of different locations in the image;

[0064] Time-frequency transformation is the core of motion contrast imaging. Commonly used time-frequency transformation methods include Fourier transform, short-time Fourier transform, Gabor transform, synchronous compression transform, Laplace transform, Z-transform, Wigner-Ville distribution, Hilbert-Huang transform, and wavelet transform (continuous wavelet transform, discrete wavelet transform, synchronous compression wavelet transform), etc. These time-frequency transformation methods have complementary advantages in terms of computational complexity and time-frequency resolution, and different schemes can be adopted according to the actual application requirements.

[0065] To accelerate time-frequency transformation, this invention utilizes GPU parallel computing to perform time-frequency transformation on image sequences and, as detailed below, time-frequency transformation of non-stationary signals. Motion-contrast imaging requires independent time-frequency transformation at different locations in the image, making it the most time-consuming step in the reconstruction process. Leveraging the parallel computing capabilities of the GPU, these independent time-frequency transformation steps are distributed to multiple CUDA cores, performing time-frequency transformation at different locations simultaneously, thereby achieving rapid reconstruction of motion-contrast imaging. The computational load for extracting spectral phase shifts and deriving target distance or velocity information, as described below, is very small and does not require GPU acceleration.

[0066] Time-frequency transformation requires significant computation, and motion-contrast imaging necessitates frequent time-frequency transformations. CPUs typically have 10-20 cores; while these cores are fast, they generally perform calculations sequentially, suitable for overall system scheduling and control tasks. CPUs can also perform parallel computing, but this requires a large amount of memory, and due to the limited number of cores, parallel computing efficiency is low, generally only improving computational speed by less than 10 times. Time-frequency transformation, on the other hand, is largely a parallel computing task. GPUs contain thousands or even tens of thousands of CUDA cores. Although the processing speed of these CUDA cores is not as fast as CPU cores, their advantage lies in the sheer number of cores. GPUs can simultaneously distribute tasks across multiple cores, executing computations in parallel, resulting in a significant increase in computational speed. Against this backdrop, this invention applies GPU parallel computing technology to motion-contrast reconstruction tasks, improving reconstruction speed by 1-2 orders of magnitude compared to CPU serial computation, greatly enhancing the practicality of the target tracking imaging proposed in this invention.

[0067] Based on the principle of motion contrast, the motion of a target introduces a phase shift factor into the frequency spectrum. By observing this phase shift, the temporal distribution of the target's motion can be determined, thus revealing the target's position at each point in time. Based on this information, we can ascertain the target's direction and speed of motion over a given period.

[0068] Specifically, the temporal variation of image grayscale values ​​reflects the motion information of the corresponding sample location. First, the temporal grayscale sequence g(x,y,t) is transformed into the frequency domain using Fourier transform to obtain the spectrum of the grayscale sequence.

[0069] The spectrum of a time-domain grayscale sequence is represented as follows:

[0070]

[0071] Where g(x,y,t) is the temporal grayscale sequence at coordinate (x,y) in the image sequence as time t changes, G(x,y,k) is the spectral grayscale sequence at coordinate (x,y) in the image sequence as k changes, N is the sequence length of the temporal grayscale sequence, and k is the discretized sampling of the frequency domain space of the time-series signal, representing the components of different frequencies in the spectrum. Numerically, f = k·f s / N, where f is the actual frequency of the signal. s Where N is the frame rate and N is the sequence length of the temporal grayscale sequence.

[0072] The phase of the spectrum is:

[0073]

[0074] Where G(x,y,k) is the spectral grayscale sequence at coordinates (x,y) in the image sequence as k changes, Im() is the imaginary part, Re() is the real part, and k is the discretized sampling of the time-series signal in the frequency domain, representing the components of different frequencies in the spectrum. Numerically, f = k·f s / N, where f is the actual frequency of the signal. s Where N is the frame rate and N is the sequence length of the temporal grayscale sequence.

[0075] Suppose that within a time interval Δt, the target moves from a certain spatiotemporal position (x0, y0, t0) to a nearby spatiotemporal position (x1, y1, t0 + Δt), and assumes that the pixel grayscale values ​​remain approximately unchanged during this process. Then, the spectral grayscale sequence after the target's motion becomes:

[0076]

[0077] Wherein, G(x0,y0,k) is the spectral grayscale sequence of the target at coordinates (x0,y0) as k changes before the target moves in the image sequence, and G(x1,y1,k) is the spectral grayscale sequence of the target at coordinates (x1,y1) as k changes after the target moves in the image sequence. This is the phase shift factor.

[0078] The above equation means that the motion of the target will introduce a phase shift factor into the frequency spectrum. Temporal information can be characterized by the phase shift factor of the Fourier spectrum.

[0079] Step S212: The spectrum is filtered by a first bandpass filter, the passband of which is the spectral range with the highest contrast in the spectrum, to obtain the amplitude map of motion contrast imaging; and the spectrum is filtered by a second bandpass filter, the passband of which is a single specific frequency, to obtain the phase map of motion contrast imaging.

[0080] To obtain the amplitude map of motion contrast imaging, the passband of the first bandpass filter should be set in the spectral region with the highest contrast [k]. LP ,k HP To obtain the phase map of motion-contrast imaging, the passband of the second bandpass filter must be a single specific frequency, meaning a bandpass filter with a single specific frequency is required.

[0081] In the obtained amplitude map, the horizontal and vertical axes represent the spatial location of the target. The amplitude map only contains the horizontal and vertical axes and the gray values ​​of their corresponding points.

[0082] The resulting phase map contains only the horizontal and vertical coordinates and their corresponding grayscale values. The horizontal and vertical coordinates represent the spatial location of the target. The normalized phase shift factor is converted into grayscale values ​​within a certain range (the range of grayscale values ​​depends on the image bit depth), and the resulting image is the phase map. Therefore, the phase map reflects temporal information through its grayscale values; points with larger grayscale values ​​represent points that the target passed earlier in time, and points with smaller grayscale values ​​represent points that the target passed later in time.

[0083] Specifically, the amplitude map of motion contrast imaging represents the target's motion trajectory, and the phase map represents time information. Combining the two can obtain the temporal distribution information of the target's motion position.

[0084] Step S213: Determine the target position based on the linear trajectory on the amplitude map of the motion contrast imaging obtained in step S212. Obtain its spectrum by performing time-frequency transformation on the grayscale signal at the target position, and then determine the passband of the third bandpass filter. The passband of the third bandpass filter covers the frequency range with the highest target frequency contrast.

[0085] In this embodiment, the time-frequency transformation at step S213 is a non-stationary signal time-frequency transformation, and the Hilbert-Huang transform is used for the non-stationary signal time-frequency transformation.

[0086] In the amplitude map of motion contrast imaging, the background is cluttered, while the target's trajectory should be linear. Therefore, the target lies on this linear trajectory in the motion contrast amplitude map, and any point on this trajectory can be considered the target point. Extracting the (x, y) coordinates of any point on the linear trajectory yields the target's location.

[0087] The specific object of time-frequency transformation is the sequence of grayscale signals at any target location in the image.

[0088] The purpose of time-frequency transformation of non-stationary signals is to determine the characteristic spectral range of the target, eliminate spectral aliasing caused by complex backgrounds and changes in illumination intensity, and thus use the contrast of the target in the frequency domain for imaging.

[0089] Step S214: Select the third bandpass filter to filter the spectrum obtained in step S211 to obtain the target's motion trajectory;

[0090] The second bandpass filter is selected by analyzing the characteristics of the target location's spectral signal. It needs to meet the condition that the passband range covers the frequency range with the highest contrast in the target's frequency domain.

[0091] The direct result of filtering the spectrum at the target location is obtaining the spectral distribution within the passband. The method for determining the target's trajectory is to sum (i.e., integrate) the amplitudes of the spectrum within the passband; the specific calculation formula is... The passband interval is [k] LP ,k HP A(x,y,k) is the spectral amplitude at point (x,y), and the meaning of k has been introduced above.

[0092] The resulting filtering can more clearly display the target's trajectory, remove clutter, and improve the sensitivity of target tracking imaging.

[0093] Step S22: Perform image fusion between the phase map of motion contrast imaging obtained in step S21 and the target motion trajectory obtained in step S214 to obtain the target motion trajectory containing time information, which serves as the time distribution information of the target's motion position in the image sequence.

[0094] First, the grayscale distribution of the phase map of the Fourier transform motion contrast provides information on the temporal sequence. Second, the target's motion trajectory is obtained by binarizing and thresholding the amplitude map of the time-frequency transform motion contrast of the non-stationary signal. Finally, image fusion of the phase map and the target's motion trajectory yields the target's motion trajectory that includes time.

[0095] Step S3: Based on the known imaging parameters, determine the actual motion position time distribution information of the target according to the motion position time distribution information of the target in the image sequence.

[0096] Known imaging parameters of an imaging device include its lens focal length, frame rate, pixel size, and imaging distance. In some embodiments, the imaging distance may not be known and can be derived from the actual average rate of the target.

[0097] The target's motion position time distribution information in the image sequence includes the length of the target's motion trajectory in the image and the corresponding time interval, while the target's actual motion position time distribution information includes the target's actual motion direction and actual average speed.

[0098] Figure 2 For the model of the imaging device and the target area, such as Figure 2 As shown, since the imaging device and the target area satisfy a similar triangle proportional relationship, and the imaging system also satisfies a similar triangle proportional relationship, the actual movement distance s of the target can be obtained based on this proportional relationship:

[0099]

[0100] Where s represents the actual movement distance of the target, n represents the length of the target's trajectory in the image (unit: pixels), a represents the pixel size of the imaging device, L represents the imaging distance of the imaging device, and f represents the focal length of the lens of the imaging device.

[0101] Then, within the time interval Δt, the target's actual average speed for:

[0102]

[0103] in, The actual average speed of the target is represented by n, the length of the target's trajectory in the image (in pixels), a, the pixel size of the imaging device, L, the imaging distance of the imaging device, and f, the focal length of the lens of the imaging device.

[0104] The known imaging parameters include the lens focal length of the imaging device, the shooting frame rate of the imaging device, and the pixel size of the imaging device. In addition, the known imaging parameters also include the imaging distance of the imaging device. Alternatively, step S3 specifically includes: determining the imaging distance of the imaging device based on an initial value of the target's actual average velocity, and then determining the target's actual motion position time distribution information based on the imaging distance of the imaging device and the known imaging parameters. In other words, given the focal length of the imaging device, the pixel size of the imaging device, the shooting frame rate, and the target trajectory length information in the image, the other can be derived from either the imaging distance of the imaging device or the initial value of the target's actual average velocity.

[0105] The formula for estimating the actual average velocity of a target based on the imaging distance of the imaging device is:

[0106] The formula for estimating the imaging distance of the imaging device based on the actual average velocity of the target is:

[0107]

[0108] in, The actual average speed of the target (which can be the initial value of the actual average speed of the target); n represents the length of the target's trajectory in the image, in pixels; a represents the pixel size of the imaging device; L represents the imaging distance of the imaging device; f represents the focal length of the lens of the imaging device.

[0109] An initial value for either the imaging distance of the imaging device or the actual average velocity of the target can be estimated based on existing knowledge or a fixed reference point.

[0110] The data acquisition method of this invention is easy to implement. Based on various time-frequency transformation methods, it utilizes the gray-scale variation characteristics of moving targets to perform imaging in the frequency domain. This method has the advantages of high imaging contrast and good actual tracking performance, significantly improving the sensitivity of motion tracking of small targets with weak signals in complex backgrounds. In addition, the tracking method based on motion contrast imaging can also obtain information such as the target's motion direction, motion speed, time distribution, and target distance.

[0111] This invention proposes to introduce motion contrast imaging into tracking imaging of small targets with weak signals in complex backgrounds. Based on the time-varying characteristics of illumination intensity and complex backgrounds, motion contrast imaging based on the time-frequency transformation of non-stationary signals is developed, which effectively eliminates the interference of complex backgrounds and high-frequency noise, thereby achieving high-sensitivity tracking imaging of targets.

[0112] From an imaging principle perspective, this invention, based on the time-varying characteristics of illumination and background, develops a motion contrast imaging method based on the time-frequency transformation of non-stationary signals. This effectively eliminates the influence of changes in illumination intensity and random background variations, and can be applied to the visible light or infrared fields. It improves the sensitivity of target tracking imaging, enabling high-sensitivity tracking of small, low-visibility targets where effective contrast cannot be obtained through direct imaging. Based on the principle of motion contrast imaging, this invention achieves high-sensitivity tracking imaging of freely moving targets. By combining phase and amplitude information from motion contrast image fusion, it achieves tracking imaging in complex backgrounds that simultaneously contain target motion direction, velocity / distance information.

[0113] Experimental results on low-visibility target tracking imaging demonstrate that motion-contrast imaging can image small targets with weak signals with a sensitivity two orders of magnitude higher than time subtraction, while also obtaining information on the temporal evolution of target motion. Based on the essence of motion-contrast imaging's time-frequency analysis of pixel grayscale changes, imaging methods using electromagnetic waves in different bands such as microwaves, terahertz, infrared, visible light, and X-rays, as well as radar methods utilizing the time-frequency domain characteristics of grayscale values ​​from multiple frames, all fall under the category of motion-contrast imaging. This method provides a highly efficient and sensitive target tracking imaging approach for multiple fields, including scientific research and industry.

[0114] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention. Various variations can be made to the above embodiments of the present invention. All simple and equivalent changes and modifications made in accordance with the claims and description of this application fall within the protection scope of the claims of this patent. All aspects not described in detail in this invention are conventional technical content.

Claims

1. An imaging method for tracking small targets with weak signals in complex systems, characterized in that, include: Step S1: Use an imaging device to capture images of the target area to obtain an image sequence; Step S2: Based on the image sequence, reconstruct the motion contrast information based on the time-frequency transformation of the non-stationary signal to obtain the motion position time distribution information of the target in the image sequence. The motion position time distribution information refers to the position of the target in the image of the image sequence at each time point. Step S3: Based on the known imaging parameters, determine the actual motion position time distribution information of the target according to the motion position time distribution information of the target in the image sequence; Step S2 specifically includes: Step S21: Perform time-frequency transformation on the grayscale signal of the image in the image sequence to obtain the spectrum at different positions, and further filter the spectrum to obtain the phase map of motion contrast imaging and the motion trajectory of the target. Step S22: Perform image fusion between the phase map of the motion contrast imaging obtained in step S21 and the motion trajectory of the target to obtain the target motion trajectory containing time information, which serves as the time distribution information of the target's motion position in the image sequence; Step S21 specifically includes: Step S211: Perform time-frequency transformation on the grayscale signals at all locations in the image to obtain the spectrum at different locations in the image; Step S212: The spectrum is filtered by a first bandpass filter, the passband of which is the spectral range with the highest contrast in the spectrum, to obtain the amplitude map of motion contrast imaging; and the spectrum is filtered by a second bandpass filter, the passband of which is a single specific frequency, to obtain the phase map of motion contrast imaging. Step S213: Determine the target position based on the linear trajectory on the amplitude map of the motion contrast imaging obtained in step S212. Obtain its spectrum by performing non-stationary signal time-frequency transformation on the gray signal at the target position, and then determine the passband of the third bandpass filter. The passband of the third bandpass filter covers the frequency range with the highest target frequency contrast. Step S214: Select the third bandpass filter to filter the spectrum obtained in step S211 to obtain the target's motion trajectory.

2. The imaging method for tracking small targets with weak signals in complex systems according to claim 1, characterized in that, Step S1 specifically includes: Step S11: Select a target area that can contain the target's motion trajectory, and install an imaging device at the location aligned with the target area for data acquisition; Step S12: Set the focal length and aperture of the imaging device to enable the imaging device to capture clear images, and set the frame rate of the imaging device as needed; Step S13: Start the imaging device, aim at the target area and start taking pictures, continuously recording the image sequence.

3. The imaging method for tracking small targets with weak signals in complex systems according to claim 1, characterized in that, In step S211, the time-frequency transformation method used for the time-frequency transformation includes one of the following: Fourier transform, short-time Fourier transform, Gabor transform, synchronous compression transform, Laplace transform, Z-transform, Wigner-Ville distribution, Hilbert-Huang transform, and wavelet transform.

4. The imaging method for tracking small targets with weak signals in complex systems according to claim 1, characterized in that, Time-frequency transformation of image sequences is performed using GPU parallel computing.

5. The imaging method for tracking small targets with weak signals in complex systems according to claim 1, characterized in that, The target's motion position time distribution information in the image sequence includes the length of the target's motion trajectory in the image and the corresponding time interval, and the target's actual motion position time distribution information includes the target's actual motion direction and the target's actual average speed.

6. The imaging method for tracking small targets with weak signals in complex systems according to claim 5, characterized in that, The known imaging parameters include the lens focal length of the imaging device, the shooting frame rate of the imaging device, and the pixel size of the imaging device. The known imaging parameters also include the imaging distance of the imaging device; or, step S3 specifically includes: determining the imaging distance of the imaging device based on the initial value of the target's actual average velocity, and then determining the target's actual motion position time distribution information based on the imaging distance of the imaging device and the known imaging parameters.

7. The imaging method for tracking small targets with weak signals in complex systems according to claim 6, characterized in that, The actual average rate of the target for: ， in, The actual average speed of the target is represented by ; n represents the length of the target's trajectory in the image, in pixels; a represents the pixel size of the imaging device; L represents the imaging distance of the imaging device; and f represents the focal length of the lens of the imaging device.

8. The imaging method for tracking small targets with weak signals in complex systems according to claim 6, characterized in that, The imaging distance L of the imaging device is: ， in, The actual average speed of the target is represented by ; n represents the length of the target's trajectory in the image, in pixels; a represents the pixel size of the imaging device; L represents the imaging distance of the imaging device; and f represents the focal length of the lens of the imaging device.

Images