Target reconstruction method and system based on passive sensing radar echo signal

Abstract

The invention provides a target reconstruction method and system based on passive sensing radar echo signals, and the method comprises the steps: obtaining a passive sensing radar echo signal set of a target scene, the passive sensing radar echo signal set comprising a plurality of echo signal sequences; performing signal preprocessing operation on the passive sensing radar echo signal set to obtain a preprocessed radar echo signal set; multi-level feature extraction operation is executed on the preprocessed radar echo signal set, a multi-dimensional state feature set of the target object is generated, and the multi-dimensional state feature set comprises scattering characteristic features, motion correlation features and environment interference suppression features; and calling a pre-trained target state analysis model to carry out state transition modeling processing on the multi-dimensional state feature set, generating a reconstructed state vector of the target object, and carrying out reverse mapping based on the reconstructed state vector to obtain a geometric parameter set and a motion parameter set of the target object.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, and more specifically, to a target reconstruction method and system based on passive sensing radar echo signals. Background Technology

[0002] In the field of radar echo signal analysis and target reconstruction, traditional techniques aim to obtain relevant information about target objects through radar echo signals. With the continuous expansion of application scenarios and the increasing demands for target reconstruction accuracy, the need for precise reconstruction of target objects in complex environments is growing daily.

[0003] Analysis revealed that existing technologies are insufficient in removing noise and interference from radar echo signals, resulting in poor signal quality and affecting the accuracy of subsequent analysis. Furthermore, current technologies are limited in feature extraction, analyzing only a few dimensions and failing to comprehensively capture the state characteristics of the target object. In addition, existing technologies lack effective state transition modeling methods during target reconstruction, making it difficult to accurately reflect the actual state changes of the target object. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a target reconstruction method and system based on passive sensing radar echo signals.

[0005] A first aspect of this application provides a target reconstruction method based on passive sensing radar echo signals, applied to a target reconstruction system, the method comprising: Acquire a set of passive sensing radar echo signals of the target scene. The set of passive sensing radar echo signals includes multiple echo signal sequences. Each echo signal sequence includes a scattered signal formed after at least one transmitting signal source is reflected to the target object. The passive sensing radar echo signal set is subjected to signal preprocessing operation to obtain a preprocessed radar echo signal set. A multi-level feature extraction operation is performed on the preprocessed radar echo signal set to generate a multi-dimensional state feature set of the target object. The multi-dimensional state feature set includes scattering characteristic features, motion correlation features, and environmental interference suppression features. The pre-trained target state analysis model is invoked to perform state transition modeling on the multi-dimensional state feature set, generating a reconstructed state vector of the target object. Based on the reconstructed state vector, the geometric parameter set and motion parameter set of the target object are obtained by inverse mapping.

[0006] Optionally, the step of performing signal preprocessing on the passive sensing radar echo signal set to obtain a preprocessed radar echo signal set includes: Noise suppression processing is performed on each scattered signal in the echo signal sequence to obtain a set of denoised scattered signals. The denoised scattered signal set is subjected to signal segmentation processing to obtain multiple signal segment sets, each signal segment set corresponding to the scattered signal within the same time window; A time-domain alignment operation is performed on the multiple signal segment sets to eliminate the time delay differences between different transmitted signal sources and generate a time-domain aligned radar echo signal set. The amplitude of the time-domain aligned radar echo signal set is normalized to obtain a preprocessed radar echo signal set.

[0007] Optionally, the step of performing signal segmentation processing on the denoised scattered signal set to obtain multiple signal segment sets includes: The signal segment length threshold is determined based on the target object's maximum speed and the radar signal sampling frequency. Based on the signal segment length threshold, the denoised scattered signal set is segmented by a sliding window to generate an initial signal segment set. Energy density detection is performed on each signal segment in the initial signal segment set. If the energy density of a signal segment is lower than a preset energy threshold, the signal segment is removed. The remaining signal segments are checked for overlap rate. If the overlap rate of adjacent signal segments exceeds a preset overlap threshold, the adjacent signal segments are merged to generate an optimized set of signal segments. The optimized signal segment set is divided into multiple signal segment sets, each corresponding to the scattered signal within the same time window.

[0008] Optionally, the step of performing multi-level feature extraction on the preprocessed radar echo signal set to generate a multi-dimensional state feature set of the target object includes: Perform time-frequency joint transformation processing on the preprocessed radar echo signal set to generate a time-frequency distribution feature set; Environmental interference compensation processing is performed on the time-frequency distribution feature set to eliminate interference components caused by ground reflection and atmospheric attenuation, resulting in an environmentally suppressed time-frequency feature set. Scattering characteristic analysis is performed on the time-frequency feature set after environmental suppression to extract the scattering intensity distribution characteristics and multipath scattering path characteristics of the target object surface; Motion trajectory correlation processing is performed on the preprocessed radar echo signal set to extract the velocity change features and acceleration correlation features of the target object; The scattering intensity distribution features, the multipath scattering path features, the velocity change features, and the acceleration correlation features are fused to generate a multidimensional set of state features of the target object.

[0009] Optionally, the scattering characteristic analysis processing performed on the time-frequency feature set after environmental suppression to extract the scattering intensity distribution characteristics and multipath scattering path characteristics of the target object surface includes: The set of time-frequency features after environmental suppression is processed by scattering center detection to determine the set of potential scattering center locations on the surface of the target object. Cluster analysis is performed on the set of potential scattering center locations to remove isolated scattering centers and generate an optimized set of scattering centers. The optimized set of scattering centers is subjected to scattering intensity distribution modeling processing to generate scattering intensity distribution features of the target object surface; Based on the optimized set of scattering centers and radar signal propagation path model, the set of multipath reflection path lengths between the target object and the emitted signal source is determined. The multipath scattering path features are generated by calculating the path phase difference set based on the set of multipath reflection path lengths and the radar signal wavelength.

[0010] Optionally, the step of calling the pre-trained target state analysis model to perform state transition modeling on the multi-dimensional state feature set to generate a reconstructed state vector of the target object includes: The multi-dimensional state feature set is input into the feature encoding module of the pre-trained target state analysis model to generate the initial state vector of the target object; The state transition module of the target state analysis model is invoked to perform time series modeling on the initial state vector, generating the state transition matrix of the target object within a continuous time window. The initial state vector is corrected based on the state transition matrix to generate a set of intermediate state vectors for the target object. The intermediate state vector set is subjected to spatial consistency verification to remove state vectors that conflict with the motion law of the target object, and an optimized intermediate state vector set is generated. The optimized intermediate state vector set is input into the feature decoding module of the target state analysis model to generate the reconstructed state vector of the target object.

[0011] Optionally, the step of performing spatial consistency verification on the intermediate state vector set to remove state vectors that conflict with the motion law of the target object includes: A set of state transition constraints is generated based on the kinematic equations of the target object. The set of state transition constraints includes velocity continuity constraints, acceleration range constraints, and motion direction smoothness constraints. Constraint matching is performed on each state vector in the intermediate state vector set, and the matching degree score between each state vector and the state transition constraint set is calculated. If the matching score is lower than the preset matching threshold, the state vector is determined to conflict with the motion law of the target object, and a set of conflicting state vectors is generated. The conflicting state vector set is removed from the intermediate state vector set to generate an optimized intermediate state vector set.

[0012] Optionally, obtaining the set of geometric parameters and the set of motion parameters of the target object based on the inverse mapping of the reconstructed state vector includes: The predefined inverse mapping model is invoked to perform geometric structure analysis on the reconstructed state vector, generating a set of three-dimensional spatial coordinates and surface curvature distribution characteristics of the target object; The three-dimensional spatial coordinate set is subjected to topological connection processing to determine the geometric boundary features and vertex connection relationships of the target object, and a set of geometric parameters is generated. The reconstructed state vector is subjected to kinematic analysis to extract the instantaneous velocity vector, acceleration vector, and motion direction angle of the target object, thereby generating a set of motion parameters; The geometric parameter set and the motion parameter set are spatiotemporally aligned to generate a complete reconstructed parameter set for the target object.

[0013] Optionally, the step of performing topological connection processing on the three-dimensional spatial coordinate set to determine the geometric boundary features and vertex connection relationships of the target object includes: Perform a neighborhood search on the set of three-dimensional spatial coordinates to determine the set of neighboring points for each spatial coordinate point; Based on the set of neighboring points, local topological connections between spatial coordinate points are generated, and an initial topological connection graph is generated. Redundant edges are removed from the initial topology graph to eliminate cross connections and loop connections, generating an optimized topology graph. Based on the optimized topology connection graph, extract the set of geometric boundary line segments and vertex connection relationships of the target object to generate a set of geometric parameters.

[0014] Optionally, after obtaining the set of geometric parameters and the set of motion parameters of the target object based on the inverse mapping of the reconstructed state vector, the method further includes: obtaining the set of three-dimensional spatial coordinates in the set of geometric parameters and the instantaneous velocity vector in the set of motion parameters; performing geometric symmetry detection processing on the set of three-dimensional spatial coordinates to generate the symmetry axis distribution features and the set of symmetry error coefficients on the surface of the target object; performing coordinate correction processing on the set of symmetry error coefficients to generate a corrected set of three-dimensional spatial coordinates; and performing motion trajectory smoothness constraint processing on the corrected set of three-dimensional spatial coordinates and the instantaneous velocity vector to eliminate abrupt components in the acceleration vector and generate an optimized set of motion parameters and the set of geometric parameters.

[0015] Optionally, after obtaining the set of geometric parameters and the set of motion parameters of the target object based on the inverse mapping of the reconstructed state vector, the method further includes: acquiring the set of Doppler frequency shift features and external environment map data from the preprocessed radar echo signal set; performing spatiotemporal alignment processing on the set of Doppler frequency shift features and the velocity change features in the set of motion parameters to generate a frequency shift-velocity correlation matrix; performing occlusion region compensation processing on the set of three-dimensional spatial coordinates in the set of geometric parameters according to the external environment map data to generate environment-corrected geometric boundary features; and performing joint optimization processing on the frequency shift-velocity correlation matrix and the environment-corrected geometric boundary features to generate the fused motion trajectory and anti-occlusion geometric parameter set of the target object.

[0016] A second aspect of this application provides a target reconstruction system, including: a processor and a memory and a bus connected to the processor; the processor and the memory communicate with each other through the bus; the processor is used to call a computer program in the memory to execute the above-described target reconstruction method based on passive sensing radar echo signals.

[0017] A third aspect of this application provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the above-described target reconstruction method based on passive sensing radar echo signals.

[0018] The target reconstruction method and system based on passive sensing radar echo signals provided in this application first acquire a set of passive sensing radar echo signals of the target scene, completely preserving the scattering signals formed by multi-source reflections, providing a comprehensive data foundation for subsequent analysis. Secondly, through signal preprocessing, noise and interference are effectively removed, improving signal quality and making the preprocessed radar echo signal set purer and more accurate, providing reliable data for feature extraction. Next, multi-level feature extraction operations generate a multi-dimensional state feature set from multiple dimensions such as scattering characteristics, motion correlation, and environmental interference suppression, comprehensively and meticulously depicting the target object's state, providing rich and accurate feature information for target reconstruction. Then, a pre-trained target state analysis model is called to perform state transition modeling, fully exploring the complex relationships between features to generate an accurate reconstructed state vector. Finally, based on this inverse mapping, the set of geometric parameters and motion parameters of the target object are obtained, achieving accurate reconstruction of the target object's state. This allows for a more comprehensive and accurate understanding of the target object's characteristics and motion, effectively improving the accuracy and reliability of target reconstruction.

[0019] Therefore, the embodiments of this application comprehensively acquire the passive sensing radar echo signal set, perform fine signal preprocessing and multi-level feature extraction, and use a pre-trained model for state transition modeling, which effectively solves the problems of inaccurate signal processing, incomplete feature extraction and inaccurate target reconstruction in the prior art, and significantly improves the quality and reliability of target reconstruction. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a target reconstruction method based on passive sensing radar echo signals, provided in an embodiment of this application.

[0022] Figure 2 This is a schematic diagram of a target reconstruction system provided in an embodiment of this application. Detailed Implementation

[0023] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0024] To better understand the above technical solutions, the technical solutions of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this application and the specific features in the embodiments are detailed descriptions of the technical solutions of this application, rather than limitations on the technical solutions of this application. In the absence of conflict, the embodiments of this application and the technical features in the embodiments can be combined with each other.

[0025] Please see Figure 1 This is a flowchart of a target reconstruction method based on passive sensing radar echo signals provided in an embodiment of this application. The method is applied to a target reconstruction system, and the specific content of the method includes steps 110-150.

[0026] Step 110: Obtain the passive sensing radar echo signal set of the target scene. The passive sensing radar echo signal set includes multiple echo signal sequences, and each echo signal sequence includes a scattered signal formed after at least one transmitting signal source is reflected to the target object.

[0027] In this embodiment, taking a monitoring area as an example, multiple signal sources are set up within this area, such as signal source A and signal source B. The target object is a moving vehicle within the area. The signal sources continuously emit signals, which are reflected upon encountering the target vehicle, forming scattered signals. Each signal source generates a corresponding echo signal sequence, and multiple echo signal sequences together constitute a passive sensing radar echo signal set. For example, the signal emitted by signal source A, after being reflected by the target vehicle, forms an echo signal sequence A containing multiple scattered signals; the signal emitted by signal source B, after being reflected by the target vehicle, forms an echo signal sequence B containing multiple scattered signals. The scattered signals in these echo signal sequences carry relevant information about the target vehicle, such as its position and motion state. Each scattered signal has corresponding characteristics, such as signal strength and frequency, which change with the state of the target vehicle.

[0028] Step 120: Perform signal preprocessing on the passive sensing radar echo signal set to obtain a preprocessed radar echo signal set.

[0029] Next, signal preprocessing is performed on the passive sensing radar echo signal set obtained above.

[0030] Step 121: Perform noise suppression processing on each scattered signal in the echo signal sequence to obtain a set of denoised scattered signals.

[0031] In this monitoring area, noise is introduced into the scattered signals of the echo signal sequence due to environmental factors. For example, electromagnetic interference from surrounding electronic devices and atmospheric clutter can contaminate the scattered signals. To remove this noise, appropriate noise suppression algorithms are employed, such as wavelet transform-based noise suppression algorithms. Taking a scattered signal from echo signal sequence A as an example, this scattered signal represents a series of discrete values ​​in the time domain. Through wavelet transform, this scattered signal is decomposed into different frequency sub-bands, with noise mainly concentrated in certain high-frequency sub-bands. Then, the coefficients in these high-frequency sub-bands are thresholded, setting coefficients below the threshold to zero, thereby removing the noise component. After this processing, the noise in the scattered signal is effectively suppressed. The same operation is performed on all scattered signals in echo signal sequence A to obtain a denoised scattered signal set A. The same noise suppression processing is also performed on the scattered signals in other echo signal sequences to obtain their respective denoised scattered signal sets.

[0032] Step 122: Perform signal segmentation processing on the denoised scattering signal set to obtain multiple signal segment sets, each signal segment set corresponding to the scattering signal within the same time window.

[0033] After denoising, the scattered signal needs to be segmented.

[0034] Step 1221: Determine the signal segment length threshold based on the target object's maximum speed and the radar signal sampling frequency.

[0035] For the target vehicle, its maximum speed is Vmax, and the radar signal sampling frequency is Fs. Determining the signal segment length threshold needs to consider the ability to completely capture the target vehicle's motion state changes over a short period. According to relevant principles, the signal segment length threshold L is related to the target vehicle's maximum speed Vmax and the radar signal sampling frequency Fs. Specifically, to ensure that the target vehicle's motion state changes within a signal segment are not excessive, the signal segment length threshold L should satisfy a certain relationship, ensuring that the distance the target vehicle travels within a signal segment is within an acceptable range. For example, if the target vehicle's maximum speed Vmax is high, the signal segment length threshold L needs to be reduced accordingly to ensure accurate capture of its motion state changes; conversely, a higher radar signal sampling frequency Fs allows for the resolution of more details, and the signal segment length threshold L can be adjusted accordingly.

[0036] Step 1222: Perform sliding window segmentation on the denoised scattered signal set based on the signal segment length threshold to generate an initial signal segment set.

[0037] Taking the denoised scattered signal set A as an example, a sliding window segmentation algorithm is employed using a predetermined signal segment length threshold L. A sliding window is defined with a size equal to the signal segment length threshold L. Starting from the beginning of the scattered signal set A, the window slides with a certain step size (e.g., a step size of 1). Each slide treats the scattered signal within the window as a signal segment, thus generating a series of signal segments that together constitute the initial signal segment set A. For example, the initial signal segment set A may contain signal segments a1, a2, etc., each corresponding to a portion of the scattered signal set A within a different time window.

[0038] Step 1223: Perform energy density detection on each signal segment in the initial signal segment set. If the energy density of a signal segment is lower than a preset energy threshold, then remove the signal segment.

[0039] For each signal segment in the initial signal segment set A, such as signal segment a1, its energy density is calculated. The energy density can be calculated by operating on the signal intensity within the signal segment, for example, by summing the squares of the signal intensities at each discrete point within the segment and then dividing by the length of the signal segment. A preset energy threshold E0 is used. If the energy density value of signal segment a1 is less than E0, it indicates that the signal segment may be severely interfered with or does not contain valid information, and it is removed from the initial signal segment set A. This energy density detection and removal operation is performed on all signal segments in the initial signal segment set A to obtain a set of signal segments that have undergone energy filtering.

[0040] Step 1224: Perform overlap rate verification on the remaining signal segments. If the overlap rate of adjacent signal segments exceeds the preset overlap threshold, merge the adjacent signal segments to generate an optimized signal segment set.

[0041] For the set of signal segments after energy filtering, taking adjacent signal segments a2 and a3 as examples, their overlap rate is calculated. The overlap rate can be calculated by determining the length of the overlapping portion of the two signal segments on the time axis and then dividing it by the length of the shorter signal segment. A preset overlap threshold R0 is set. If the overlap rate of signal segments a2 and a3 is greater than R0, it indicates that these two signal segments have a lot of repetitive information, and they are merged into a new signal segment. This overlap rate verification and merging operation is performed on all adjacent signal segments to generate an optimized signal segment set A. The same signal segmentation processing is performed on other denoised scattered signal sets to obtain their respective optimized signal segment sets.

[0042] Step 1225: Divide the optimized signal segment set into multiple signal segment sets, each signal segment set corresponding to the scattered signal within the same time window.

[0043] The optimized signal segment set A is divided according to time windows. For example, a fixed time interval T is used as a time window. Signal segments within the same time window in the optimized signal segment set A are grouped together, resulting in multiple signal segment sets. Each signal segment set corresponds to the scattered signal within the same time window. For example, signal segment sets A1, A2, etc., are obtained. The signal segments in signal segment set A1 all come from scattered signals within a certain time window, while the signal segments in signal segment set A2 all come from scattered signals within another time window. The same division operation is performed on other optimized signal segment sets.

[0044] Step 123: Perform time-domain alignment on the multiple signal segment sets to eliminate the time delay differences between different transmitted signal sources and generate a time-domain aligned radar echo signal set.

[0045] Due to the varying distances from different signal sources to the target vehicle, and the influence of various factors during signal propagation, time delay differences exist between signal segment sets corresponding to different signal sources. To eliminate these time delay differences, a time-domain alignment algorithm is employed. Taking signal segment set A and signal segment set B as examples, these two sets originate from signal source A and signal source B, respectively. A time-domain alignment algorithm (e.g., a cross-correlation-based time-domain alignment algorithm) is used to calculate the time delay between signal segment set A and signal segment set B. Specifically, by calculating the cross-correlation function of the signals in the two sets, the time delay value corresponding to the maximum value of the cross-correlation function is found; this time delay value represents the time delay difference between the two signal segment sets. Then, signal segment set B is time-shifted according to this time delay value to align it with signal segment set A in the time domain. This time-domain alignment operation is performed on all signal segment sets corresponding to different signal sources, generating a time-domain aligned radar echo signal set.

[0046] Step 124: Perform amplitude normalization processing on the time-domain aligned radar echo signal set to obtain a preprocessed radar echo signal set.

[0047] After time-domain alignment, to ensure comparability between different signals, amplitude normalization is performed on the time-domain aligned radar echo signal set. For each signal in the time-domain aligned radar echo signal set, its amplitude range is calculated. For example, for a signal s, its maximum value M and minimum value m are found, and then the amplitude of the signal is mapped to a fixed range, such as [0, 1], using a normalization algorithm (e.g., subtracting the minimum value m from each value in signal s and then dividing by the difference between the maximum value M and the minimum value m). This amplitude normalization process is performed on all signals in the time-domain aligned radar echo signal set to obtain the preprocessed radar echo signal set.

[0048] Step 130: Perform multi-level feature extraction on the preprocessed radar echo signal set to generate a multi-dimensional state feature set of the target object. The multi-dimensional state feature set includes scattering characteristic features, motion correlation features, and environmental interference suppression features.

[0049] After obtaining the preprocessed radar echo signal set, multi-level feature extraction operations are performed.

[0050] Step 131: Perform time-frequency joint transformation processing on the preprocessed radar echo signal set to generate a time-frequency distribution feature set.

[0051] In this step, a time-frequency joint transformation algorithm, such as the short-time Fourier transform algorithm, can be used. For each signal in the preprocessed radar echo signal set, taking a single signal as an example, it is divided into multiple shorter time segments, and a Fourier transform is performed on the signal within each time segment. Through the Fourier transform, the time-domain signal is transformed into the frequency domain, obtaining the frequency distribution of the signal within that time segment. As time progresses, the frequency distributions of different time segments combine to form the time-frequency distribution of the signal. This time-frequency joint transformation process is performed on all signals in the preprocessed radar echo signal set to generate a time-frequency distribution feature set. This time-frequency distribution feature set contains the distribution information of each signal at different times and frequencies, such as which frequencies of the signal have higher energy at a certain time point.

[0052] Step 132: Perform environmental interference compensation processing on the time-frequency distribution feature set to eliminate interference components caused by ground reflection and atmospheric attenuation, and obtain the environmentally suppressed time-frequency feature set.

[0053] In real-world monitoring environments, ground reflection and atmospheric attenuation interfere with radar echo signals. To eliminate this interference, an environmental interference compensation algorithm is employed. Taking a time-frequency distribution from the time-frequency distribution feature set as an example, a corresponding interference model is established by analyzing the characteristics of ground reflection and atmospheric attenuation. For instance, based on factors such as ground material and atmospheric humidity, the influence of ground reflection and atmospheric attenuation on signal strength and frequency is determined. Then, based on this interference model, the interference components in the time-frequency distribution are estimated and compensated. For example, if the interference model indicates that the signal strength is reduced by a certain proportion due to ground reflection and atmospheric attenuation within a certain frequency range, then the signal strength within that frequency range in the time-frequency distribution is correspondingly increased to eliminate the interference. This environmental interference compensation process is applied to all time-frequency distributions in the time-frequency distribution feature set to obtain an environmentally suppressed time-frequency feature set.

[0054] Step 133: Perform scattering characteristic analysis on the time-frequency feature set after environmental suppression to extract the scattering intensity distribution characteristics and multipath scattering path characteristics of the target object surface.

[0055] Step 1331: Perform scattering center detection processing on the time-frequency feature set after environmental suppression to determine the set of potential scattering center locations on the surface of the target object.

[0056] For the time-frequency feature set after environmental suppression, a scattering center detection algorithm is employed. By analyzing information such as the signal intensity distribution and frequency changes in the time-frequency feature set, the possible locations of scattering centers on the target object's surface are determined. For example, in the time-frequency feature set, the signal intensity in certain regions is significantly higher than that in the surrounding regions, and the frequency characteristics also show corresponding changes; the locations on the target object's surface corresponding to these regions are likely scattering centers. By setting an algorithm (e.g., an algorithm based on threshold detection and cluster analysis), these possible scattering center locations are determined, forming a set of potential scattering center locations on the target object's surface.

[0057] Step 1332: Perform cluster analysis on the set of potential scattering center locations, remove isolated scattering centers, and generate an optimized set of scattering centers.

[0058] Cluster analysis is performed on the set of potential scattering center locations using a clustering algorithm, such as K-means clustering. The locations in the set are clustered according to features such as distance between them. For example, locations that are closer together are grouped into one cluster. During the clustering process, some isolated scattering centers are identified. These locations are far from other scattering centers and do not exhibit obvious clustering characteristics. Based on preset rules (e.g., setting a distance threshold; if a scattering center is more than a certain distance from its nearest cluster center, it is considered an isolated scattering center), these isolated scattering centers are removed, generating an optimized set of scattering centers.

[0059] Step 1333: Perform scattering intensity distribution modeling on the optimized scattering center set to generate the scattering intensity distribution characteristics of the target object surface.

[0060] For the optimized set of scattering centers, a scattering intensity distribution model is established. Taking each scattering center in the optimized set as an example, the signal intensity distribution within a certain range around it is analyzed. Using a pre-defined modeling algorithm (e.g., an interpolation and fitting-based algorithm), a function is constructed based on the signal intensity data around the scattering center to describe the scattering intensity distribution around it. This modeling process is performed on all scattering centers, and the scattering intensity distribution models of each scattering center are combined to generate the scattering intensity distribution characteristics of the target object's surface. This characteristic describes the scattering intensity at different locations on the target object's surface.

[0061] Step 1334: Based on the optimized set of scattering centers and radar signal propagation path model, determine the set of multipath reflection path lengths between the target object and the emitted signal source.

[0062] Based on the optimized set of scattering centers and the radar signal propagation path model, the multipath reflection path length between the target object and the transmitting signal source is calculated. Taking a single transmitting signal source and a scattering center on the target object's surface as an example, the radar signal propagation path model describes the propagation mode of the signal from the transmitting signal source to the scattering center on the target object's surface, including possible reflections and refractions. Using this model, combined with the location information of the scattering center and the transmitting signal source, the direct path length and possible multipath reflection path lengths from the transmitting signal source to the scattering center are calculated. This calculation is performed for all transmitting signal sources and all scattering centers in the optimized set of scattering centers to obtain the set of multipath reflection path lengths between the target object and the transmitting signal source.

[0063] Step 1335: Calculate the path phase difference set based on the multipath reflection path length set and the radar signal wavelength to generate multipath scattering path features.

[0064] Given the set of multipath reflection path lengths and the radar signal wavelength, calculate the path phase difference. For each path length in the set of multipath reflection path lengths, calculate the corresponding path phase difference based on the relationship between the path phase difference, path length, and radar signal wavelength (e.g., the path phase difference equals the ratio of the path length to the radar signal wavelength multiplied by 2π). Combine all the calculated path phase differences to generate a multipath scattering path feature. This feature contains the phase difference information of different multipath reflection paths between the target object and the transmitting signal source.

[0065] Step 134: Perform motion trajectory correlation processing on the preprocessed radar echo signal set to extract the velocity change features and acceleration correlation features of the target object.

[0066] A motion trajectory correlation algorithm is employed to analyze a pre-processed set of radar echo signals. By observing the characteristic changes of the signals at different times, such as the Doppler frequency shift, the motion state of the target object is inferred. Taking a pre-processed set of radar echo signals over a period of time as an example, the frequency changes of the signals acquired at different times are analyzed. If the signal frequency changes with time, the velocity change of the target object can be calculated based on the Doppler effect. By further analyzing the velocity calculation results at multiple time points, such as calculating the rate of change of velocity, the acceleration correlation characteristics of the target object can be extracted. Through this motion trajectory correlation processing, the velocity change characteristics and acceleration correlation characteristics of the target object are obtained.

[0067] Step 135: Perform feature fusion processing on the scattering intensity distribution features, the multipath scattering path features, the velocity change features, and the acceleration correlation features to generate a multi-dimensional state feature set of the target object.

[0068] A feature fusion algorithm is employed to fuse the aforementioned scattering intensity distribution features, multipath scattering path features, velocity variation features, and acceleration correlation features. For example, a weighted fusion algorithm is used, assigning a weight to each feature based on its importance in describing the target object's state. The scattering intensity distribution features, multipath scattering path features, velocity variation features, and acceleration correlation features are then weighted and concatenated according to their respective weights to generate a multi-dimensional state feature set of the target object. This multi-dimensional state feature set contains information on the target object's scattering characteristics, motion state, and environmental interference suppression, among other aspects.

[0069] Step 140: Call the pre-trained target state analysis model to perform state transition modeling on the multi-dimensional state feature set, generate the reconstructed state vector of the target object, and obtain the geometric parameter set and motion parameter set of the target object by inverse mapping based on the reconstructed state vector.

[0070] The multi-dimensional set of state features is input into a pre-trained target state analysis model for processing.

[0071] Step 141: Input the multi-dimensional state feature set into the feature encoding module of the pre-trained target state analysis model to generate the initial state vector of the target object.

[0072] The pre-trained target state analysis model includes a feature encoding module. A multi-dimensional set of state features is input into the feature encoding module, which employs encoding algorithms such as those used in Convolutional Neural Networks (CNNs). Taking one feature from the multi-dimensional state feature set as an example, this feature might be a multi-dimensional array. The feature encoding module extracts and compresses this feature through convolutional operations, pooling operations, etc., converting it into a low-dimensional vector representation. This processing is performed on all features in the multi-dimensional state feature set, and then the processed vectors are concatenated to generate the initial state vector of the target object. This initial state vector contains the encoded information from the multi-dimensional state feature set.

[0073] Step 142: Call the state transition module of the target state analysis model to perform time series modeling processing on the initial state vector to generate the state transition matrix of the target object within a continuous time window.

[0074] The state transition module of the target state analysis model employs a time series modeling algorithm, such as the Long Short-Term Memory (LSTM) algorithm. The initial state vector is input into the state transition module, which learns the state transition patterns of the target object based on the changes in the initial state vector at different time points. By analyzing and processing the initial state vectors within multiple consecutive time windows, a state transition matrix for the target object within those consecutive time windows is generated. This state transition matrix describes the state changes of the target object at different time points.

[0075] Step 143: Based on the state transition matrix, modify the initial state vector to generate a set of intermediate state vectors for the target object.

[0076] When correcting the initial state vector using the generated state transition matrix, this is achieved through predefined matrix operation rules. Taking an element in the initial state vector as an example, weighted summations and other operations are performed on that element based on the relationships between corresponding elements in the state transition matrix. For instance, some elements in the state transition matrix represent the influence weights between different elements in the initial state vector at a corresponding time step. Each element in the initial state vector is calculated according to these weights to obtain the corrected vector elements. This calculation is performed on all elements in the initial state vector to generate corrected vectors. These corrected vectors together constitute the intermediate state vector set of the target object. This intermediate state vector set considers the state change trend of the target object within a continuous time window and, compared to the initial state vector, more accurately reflects the state of the target object at different times.

[0077] Step 144: Perform spatial consistency verification on the intermediate state vector set, remove state vectors that conflict with the motion law of the target object, and generate an optimized intermediate state vector set.

[0078] Step 1441: Generate a set of state transition constraints based on the kinematic equations of the target object. The set of state transition constraints includes velocity continuity constraints, acceleration range constraints, and motion direction smoothness constraints.

[0079] For the target vehicle, its kinematic equations are established based on kinematic principles. Regarding velocity continuity constraints, considering that the target vehicle's velocity will not change drastically instantaneously during actual motion, a threshold range for velocity change is set. For example, if the target vehicle's velocity is V1 at the previous moment and V2 at the next moment, then the velocity change V2-V1 should be within a reasonable range; exceeding this range violates the velocity continuity constraint. Acceleration range constraints determine the maximum and minimum values ​​of acceleration based on the target vehicle's dynamic performance and physical limitations. For example, the target vehicle's acceleration 'a' should satisfy amin ≤ a ≤ amax. Motion direction smoothness constraints ensure that the target vehicle's motion direction will not suddenly change drastically; this is limited by setting an angle change threshold. For example, if the motion direction angle is θ1 at the previous moment and θ2 at the next moment, then |θ2-θ1| should be less than a preset angle threshold. These constraints are integrated to form a set of state transition constraints.

[0080] Step 1442: Perform constraint matching processing on each state vector in the intermediate state vector set, and calculate the matching degree score between each state vector and the state transition constraint set.

[0081] For each state vector in the intermediate state vector set, the state vector contains motion state information of the target object at a certain moment, such as velocity, acceleration, and direction of motion. Taking one state vector as an example, its velocity information is compared with the velocity continuity constraint to calculate whether the velocity change is within the allowable threshold range, and a score is given based on the degree of compliance. Similarly, the acceleration information is compared with the acceleration range constraint; if the acceleration value is within the specified range of amin ≤ a ≤ amax, a corresponding score is given. For the direction of motion information, it is compared with the direction of motion smoothness constraint, and a score is given based on whether the angle change is less than a preset threshold. Combining these scores, and using preset calculation rules (such as weighted summation, assigning different weights to the scores of velocity, acceleration, and direction of motion), the matching degree score between the state vector and the set of state transition constraints is calculated. This constraint matching process and matching degree score calculation are performed on all state vectors in the intermediate state vector set.

[0082] Step 1443: If the matching score is lower than the preset matching threshold, it is determined that the state vector conflicts with the motion law of the target object, and a set of conflicting state vectors is generated.

[0083] A matching threshold M0 is preset, and the matching score calculated for each state vector is compared with M0. If the matching score of a state vector is less than M0, it means that the motion state of the target object represented by that state vector does not conform to the preset kinematic constraints, i.e., it conflicts with the motion law of the target object. All state vectors with matching scores lower than M0 are collected to form a conflict state vector set.

[0084] Step 1444: Remove the conflicting state vector set from the intermediate state vector set to generate an optimized intermediate state vector set.

[0085] After determining the set of conflicting state vectors, the conflicting state vectors in the intermediate state vector set are removed. The remaining state vectors constitute the optimized intermediate state vector set. The state vectors in this optimized intermediate state vector set all conform to the kinematic laws of the target object and can more accurately reflect the true state of the target object.

[0086] Step 145: Input the optimized intermediate state vector set into the feature decoding module of the target state analysis model to generate the reconstructed state vector of the target object.

[0087] The optimized set of intermediate state vectors is input into the feature decoding module of the target state analysis model. The feature decoding module employs a decoding algorithm corresponding to that of the feature encoding module, such as deconvolution. Taking one vector from the optimized set of intermediate state vectors as an example, the feature decoding module performs a deconvolution operation on this vector, transforming it from a low-dimensional encoded representation back to a high-dimensional feature representation. This decoding process is performed on all vectors in the optimized set of intermediate state vectors. Then, the decoded vectors are concatenated to generate the reconstructed state vector of the target object. This reconstructed state vector integrates various aspects of the target object after state transition modeling and verification, providing a foundation for subsequent inverse mapping to obtain the set of geometric and motion parameters of the target object.

[0088] Step 150: Obtain the set of geometric parameters and the set of motion parameters of the target object by inverse mapping based on the reconstructed state vector.

[0089] The reconstructed state vector is processed by a predefined inverse mapping model to obtain the set of geometric parameters and motion parameters of the target object.

[0090] Step 151: Call the predefined inverse mapping model to perform geometric structure analysis on the reconstructed state vector to generate the three-dimensional spatial coordinate set and surface curvature distribution characteristics of the target object.

[0091] The reconstructed state vector contains comprehensive state information of the target object. The inverse mapping model analyzes geometrically relevant elements in the reconstructed state vector, such as elements that may be associated with the target object's position in different directions. Using pre-defined mapping rules (e.g., mapping relationships obtained through pre-training), these elements are converted into the target object's three-dimensional spatial coordinates. For example, several elements in the reconstructed state vector, after calculation and transformation, yield the target object's coordinates on the X, Y, and Z axes; numerous such coordinates constitute the set of the target object's three-dimensional spatial coordinates. Simultaneously, the inverse mapping model also analyzes other relevant elements in the reconstructed state vector, such as elements related to the target object's surface shape, and uses appropriate algorithms (e.g., algorithms based on surface fitting and differential calculation) to calculate the curvature distribution characteristics of the target object's surface.

[0092] Step 152: Perform topological connection processing on the three-dimensional spatial coordinate set to determine the geometric boundary features and vertex connection relationships of the target object, and generate a set of geometric parameters.

[0093] Step 1521: Perform a neighborhood search on the three-dimensional spatial coordinate set to determine the set of neighboring points for each spatial coordinate point.

[0094] For each coordinate point in the set of three-dimensional spatial coordinates, such as coordinate point P(x, y, z), a neighborhood search algorithm is used. A search radius R is set, and other coordinate points are searched within a spherical region of radius R centered on coordinate point P. All coordinate points within this region are considered as neighbors of coordinate point P, forming a set of neighbors for coordinate point P. This neighborhood search process is performed on all coordinate points in the set of three-dimensional spatial coordinates to determine the set of neighbors for each coordinate point.

[0095] Step 1522: Generate local topological connection relationships between spatial coordinate points based on the set of neighboring points, and generate an initial topological connection graph.

[0096] Based on the set of neighboring points of each coordinate point, the connections between coordinate points are determined. For example, if coordinate point P1's set of neighboring points includes coordinate point P2, then a connection is established between coordinate points P1 and P2. In this way, the connections between all coordinate points are determined, forming local topological connections between spatial coordinate points. These connections are then represented graphically to generate an initial topological connection diagram. This initial topological connection diagram shows the preliminary connections between the three-dimensional spatial coordinate points of the target object.

[0097] Step 1523: Perform redundant edge removal processing on the initial topology connection graph to eliminate cross connections and loop connections, and generate an optimized topology connection graph.

[0098] The initial topology graph is analyzed to identify redundant edges. Redundant edges include cross connections and loop connections. For example, if two edges AB and CD cross in space, such cross connections may affect the accurate description of the target object's geometry and should be removed. For loop connections, such as a loop consisting of coordinate points A, B, and C, relevant algorithms (e.g., graph theory-based algorithms) are used to analyze which edges constitute the loop and remove these redundant edges. After this redundant edge removal process, an optimized topology graph is generated.

[0099] Step 1524: Extract the set of geometric boundary line segments and vertex connection relationships of the target object based on the optimized topology connection graph, and generate a set of geometric parameters.

[0100] From the optimized topology graph, line segments located on the geometric boundaries of the target object are identified. For example, by analyzing the endpoints of edges in the graph, it is determined which edges are at boundary positions, and these boundary segments are collected to form a set of geometric boundary segments. Simultaneously, the connection relationships between vertices in the graph are analyzed; for example, which vertices are directly connected to vertex A, and these vertex connection relationships are compiled to form a set of vertex connection relationships. The set of geometric boundary segments and the set of vertex connection relationships are combined to generate a set of geometric parameters.

[0101] Step 153: Perform kinematic analysis on the reconstructed state vector to extract the instantaneous velocity vector, acceleration vector and motion direction angle of the target object, and generate a set of motion parameters.

[0102] The reconstructed state vector is analyzed using a kinematic analytical algorithm. The reconstructed state vector contains the motion state information of the target object. By analyzing the velocity-related elements in the vector—for example, some elements may be related to the rate of change of velocity of the target object in different directions—the instantaneous velocity components of the target object in each direction are calculated according to relevant calculation rules (e.g., rules based on derivative calculations). These components are then combined to form the instantaneous velocity vector of the target object. Similarly, the acceleration-related elements are analyzed, and the acceleration components of the target object in each direction are calculated using similar calculation rules, and then combined to form the acceleration vector. For the motion direction angle, the motion direction angle of the target object is calculated by analyzing the direction-related elements in the vector, such as using trigonometric functions. The instantaneous velocity vector, acceleration vector, and motion direction angle are combined to generate a set of motion parameters.

[0103] Step 154: Perform spatiotemporal alignment processing on the geometric parameter set and the motion parameter set to generate a complete reconstruction parameter set for the target object.

[0104] Spatiotemporal alignment is performed on the geometric parameter set and the motion parameter set. Since the geometric parameter set describes the spatial structure information of the target object, and the motion parameter set describes its motion state information, they need to be unified in time and space. For example, the time points corresponding to the geometric parameter set and the motion parameter set are determined, and the geometric parameters and motion parameters at different time points are matched. Spatially, the positional information in the motion parameters is associated with the three-dimensional spatial coordinates in the geometric parameters. Using a spatiotemporal alignment algorithm (such as an algorithm based on timestamp matching and spatial coordinate mapping), the geometric parameter set and the motion parameter set are integrated to generate a complete reconstructed parameter set for the target object. This complete reconstructed parameter set comprehensively describes the geometric structure and motion state of the target object.

[0105] After obtaining the set of geometric parameters and the set of motion parameters of the target object through inverse mapping based on the reconstructed state vector, there are still some further processing steps.

[0106] Step 210: Obtain the three-dimensional spatial coordinate set in the geometric parameter set and the instantaneous velocity vector in the motion parameter set.

[0107] A three-dimensional spatial coordinate set is extracted from the generated set of geometric parameters. This set contains information about the coordinates of the target object in three-dimensional space. Simultaneously, an instantaneous velocity vector is obtained from the set of motion parameters. This instantaneous velocity vector describes the magnitude and direction of the target object's velocity at a given moment. For example, the three-dimensional spatial coordinate set contains a series of coordinate points P1(x1, y1, z1), P2(x2, y2, z2), etc., and the instantaneous velocity vector V = (Vx, Vy, Vz), where Vx, Vy, and Vz represent the velocity components in the X, Y, and Z directions, respectively.

[0108] Step 220: Perform geometric symmetry detection processing on the three-dimensional spatial coordinate set to generate the symmetry axis distribution characteristics and symmetry error coefficient set of the target object surface.

[0109] A geometric symmetry detection algorithm is employed to analyze a set of three-dimensional spatial coordinates. Taking a target vehicle as an example, for the vehicle's three-dimensional spatial coordinate set, relevant algorithms (such as those based on principal component analysis and symmetry transformation) are used to find possible axes of symmetry. For instance, if the target vehicle exhibits a certain degree of symmetry, the algorithm analyzes the distribution of coordinate points to calculate the directions in which these points exhibit symmetrical distribution. The direction and position of the axes of symmetry are determined, forming the distribution characteristics of the symmetry axes on the target object's surface. Simultaneously, symmetry error coefficients are calculated. For example, for coordinate points on both sides of the axis of symmetry, indices such as the difference in their distances to the axis of symmetry are calculated. Through pre-defined calculation rules (such as statistical analysis of these distance differences), a set of symmetry error coefficients is obtained. This set of symmetry error coefficients reflects the degree of difference between the actual geometric shape and the ideal symmetrical shape of the target object.

[0110] Step 230: Perform coordinate correction processing on the three-dimensional spatial coordinate set according to the set of symmetry error coefficients to generate a corrected three-dimensional spatial coordinate set.

[0111] The three-dimensional spatial coordinate set is corrected based on a set of symmetry error coefficients. For each coordinate point in the three-dimensional spatial coordinate set, adjustments are made according to the deviation reflected by the symmetry error coefficients. For example, if the symmetry error coefficients indicate a deviation in a certain direction, the coordinate point is displaced in the corresponding direction based on the magnitude and direction of the deviation. In this way, all coordinate points in the three-dimensional spatial coordinate set are corrected, generating a corrected three-dimensional spatial coordinate set. This corrected three-dimensional spatial coordinate set makes the geometry of the target object more consistent with its actual symmetry characteristics.

[0112] Step 240: Based on the corrected three-dimensional spatial coordinate set and the instantaneous velocity vector, perform motion trajectory smoothness constraint processing to eliminate abrupt components in the acceleration vector and generate an optimized set of motion parameters and a set of geometric parameters.

[0113] By combining the corrected 3D spatial coordinate set and the instantaneous velocity vector, the acceleration vector is subjected to motion trajectory smoothness constraints. Since the acceleration of a target object does not change abruptly in actual motion, the changes in coordinate points over time and the instantaneous velocity vector in the corrected 3D spatial coordinate set are analyzed to determine if there are abrupt components in the acceleration vector. For example, if a component of the acceleration vector shows an abnormally large change at a certain moment, which should not occur according to the motion trajectory and velocity changes of the coordinate point, then this abrupt component is adjusted. A smoothing algorithm (such as a filtering and interpolation-based algorithm) is used to process the acceleration vector, eliminating abrupt components and making the acceleration vector smoother. Simultaneously, based on the adjustment of the acceleration vector, other elements in the motion parameter set (such as the motion direction angle) are fine-tuned accordingly, generating an optimized motion parameter set. For the geometric parameter set, since the coordinate points have already been corrected, and the adjustment of the motion state may also affect the description of the geometric structure, some necessary updates and adjustments are made to the geometric parameter set based on the optimized motion parameter set, generating an optimized geometric parameter set.

[0114] After obtaining the set of geometric parameters and the set of motion parameters of the target object through inverse mapping based on the reconstructed state vector, there is another series of subsequent processing steps.

[0115] Step 310: Obtain the Doppler frequency shift feature set and external environment map data from the preprocessed radar echo signal set.

[0116] The Doppler frequency shift feature set is extracted from the preprocessed radar echo signal set. This set contains information about the frequency changes of the radar echo signal caused by the movement of the target object. For example, the frequency of the radar echo signal increases when the target vehicle approaches the transmitting signal source, and decreases when it moves away. The Doppler frequency shift feature set is obtained by performing frequency analysis on the preprocessed radar echo signal set. Simultaneously, external environment map data is acquired. This data can include information such as the terrain and building distribution of the monitored area. For example, the map data may contain information such as the location and height of buildings, which is crucial for accurately describing the movement and geometry of the target object.

[0117] Step 320: Perform spatiotemporal alignment processing on the Doppler frequency shift feature set and the velocity change features in the motion parameter set to generate a frequency shift-velocity correlation matrix.

[0118] Spatiotemporal alignment of velocity variation features within the Doppler frequency shift feature set and motion parameter set is performed. First, the time points corresponding to the Doppler frequency shift features and velocity variation features are determined. For example, the Doppler frequency shift value in the radar echo signal acquired at a certain moment, and the velocity variation value of the target object at the same moment, are found. Then, the Doppler frequency shift feature set and velocity variation features are matched according to the time points. Using relevant algorithms (e.g., time series matching and interpolation-based algorithms), the Doppler frequency shift and velocity variation values ​​at different time points are correlated. These correlations are represented in matrix form, generating a frequency shift-velocity correlation matrix. This matrix reflects the correspondence between Doppler frequency shift and target object velocity variation at different time points.

[0119] Step 330: Perform occlusion area compensation processing on the three-dimensional spatial coordinate set in the geometric parameter set based on the external environment map data to generate environmentally corrected geometric boundary features.

[0120] This method combines external environment map data with occlusion compensation processing for a 3D spatial coordinate set. Taking buildings in the external environment map as an example, if a building is near a target object (such as a vehicle), it may occlude part of the target object's geometry. Based on the location and shape information of buildings in the map data, and the coordinate information of the target object in the 3D spatial coordinate set, it determines which coordinate points are likely to be in occluded areas. For coordinate points in occluded areas, an adaptive compensation algorithm (such as an algorithm based on projection and interpolation) is used to adjust these coordinate points. For example, if a coordinate point is occluded by a building, based on the building's boundary and the target object's movement direction, the possible location of the coordinate point in the unoccluded state is inferred, and corresponding compensation is performed. Through this processing, all potentially occluded coordinate points in the 3D spatial coordinate set are processed to generate environment-corrected geometric boundary features. These environment-corrected geometric boundary features take into account the influence of the external environment on the target object's geometry.

[0121] Step 340: Perform joint optimization processing on the frequency shift-velocity correlation matrix and the environmentally corrected geometric boundary features to generate the fused motion trajectory and anti-occlusion geometric parameter set of the target object.

[0122] A joint optimization algorithm is employed to process the frequency-shift-velocity correlation matrix and the environment-corrected geometric boundary features. This algorithm considers both the target object's motion information reflected in the frequency-shift-velocity correlation matrix and the target object's geometric structure information reflected in the environment-corrected geometric boundary features. For example, analyzing the frequency-shift-velocity correlation matrix reveals the target object's velocity changes at different times, while the environment-corrected geometric boundary features provide information about the target object's geometric shape after considering external environmental occlusion. The algorithm integrates and optimizes these two information sources through optimization strategies (e.g., strategies based on minimizing the objective function). For instance, when generating the target object's trajectory, it considers both the impact of velocity changes on the trajectory and the constraints imposed by geometric boundary features to prevent the trajectory from crossing occluded areas. Through this joint optimization process, a fused motion trajectory and an anti-occlusion geometric parameter set for the target object are generated. This fused motion trajectory integrates multiple factors such as Doppler frequency shift, velocity changes, and the external environment, while the anti-occlusion geometric parameter set more accurately describes the target object's geometry under occlusion conditions. Specifically, the joint optimization algorithm continuously adjusts the motion trajectory and geometric parameters in multiple iterations to achieve the best matching and optimization results.

[0123] During the iteration process, the algorithm first predicts the possible positional changes of the target object at different times based on the velocity information provided by the frequency-shift-velocity correlation matrix. For example, if the frequency-shift-velocity correlation matrix shows that the velocity of the target object gradually increases within a certain time period, the displacement of the target object within that time period is increased accordingly when predicting the motion trajectory. Simultaneously, the algorithm references the environmentally corrected geometric boundary features to ensure that the predicted motion trajectory does not conflict with occluded areas. If the predicted trajectory might pass through occluded areas such as buildings, the algorithm adjusts the trajectory to bypass the occluded areas.

[0124] For the set of geometric parameters, the joint optimization algorithm updates accordingly based on the adjustment of the motion trajectory. For example, if the motion trajectory changes, it may affect the geometric boundary features of the target object. The algorithm will re-evaluate and adjust the set of geometric boundary line segments and vertex connection relationships in the set of geometric parameters to ensure the consistency between the geometric structure and the motion state.

[0125] In each iteration, the algorithm calculates an objective function value that comprehensively considers factors such as the accuracy of the frequency shift-velocity correlation and the matching degree between the motion trajectory and the geometric boundary. By continuously adjusting the motion trajectory and geometric parameters, the objective function value is gradually reduced until a certain convergence condition is reached. For example, when the objective function value changes very little in consecutive iterations, or reaches a preset minimum value, the joint optimization process is considered to have converged. The motion trajectory and geometric parameters obtained at this point are the optimized results, which are the fused motion trajectory and anti-occlusion geometric parameter set of the target object.

[0126] Fusion of motion trajectories more accurately reflects the motion of the target object in the real environment, taking into account the velocity information brought about by Doppler frequency shift and the constraints of the external environment on motion. The anti-occlusion geometric parameter set provides a more accurate description of the target object's geometry in the presence of occlusion, making the reconstruction of the target object in complex environments more reliable and comprehensive. This approach not only improves the accuracy of target reconstruction but also enhances the adaptability and stability of the entire target reconstruction method in practical applications, better meeting the needs for accurate reconstruction and analysis of target objects in various real-world scenarios.

[0127] Through the detailed embodiments described above, those skilled in the art can clearly understand and implement the target reconstruction method based on passive sensing radar echo signals, from signal acquisition and preprocessing to feature extraction, state analysis, and finally target parameter reconstruction and optimization.

[0128] In summary, this embodiment first acquires the passive sensing radar echo signal set of the target scene, completely preserving the scattering signals formed by multi-source reflections, providing a comprehensive data foundation for subsequent analysis. Secondly, through signal preprocessing, noise and interference are effectively removed, improving signal quality and making the preprocessed radar echo signal set purer and more accurate, providing reliable data for feature extraction. Next, multi-level feature extraction operations generate a multi-dimensional state feature set from multiple dimensions such as scattering characteristics, motion correlation, and environmental interference suppression, comprehensively and meticulously depicting the target object's state and providing rich and accurate feature information for target reconstruction. Then, a pre-trained target state analysis model is invoked to perform state transition modeling, fully exploring the complex relationships between features to generate an accurate reconstructed state vector. Finally, based on this inverse mapping, the geometric parameter set and motion parameter set of the target object are obtained, achieving accurate reconstruction of the target object's state. This allows for a more comprehensive and accurate understanding of the target object's characteristics and motion, effectively improving the accuracy and reliability of target reconstruction.

[0129] Therefore, the embodiments of this application comprehensively acquire the passive sensing radar echo signal set, perform fine signal preprocessing and multi-level feature extraction, and use a pre-trained model for state transition modeling, which effectively solves the problems of inaccurate signal processing, incomplete feature extraction and inaccurate target reconstruction in the prior art, and significantly improves the quality and reliability of target reconstruction.

[0130] Please refer to the following: Figure 2 This application also provides a target reconstruction system 100, including a processor 111, a memory 112, and a bus 113 connected to the processor 111. The processor 111 and the memory 112 communicate with each other via the bus 113. The processor 111 is used to call program instructions in the memory 112 to execute the aforementioned target reconstruction method based on passive sensing radar echo signals.

[0131] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, product, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, product, or system that includes that element.

[0132] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A target reconstruction method based on passive sensing radar echo signals, characterized in that, include: Acquire a set of passive sensing radar echo signals of the target scene. The set of passive sensing radar echo signals includes multiple echo signal sequences. Each echo signal sequence includes a scattered signal formed after at least one transmitting signal source is reflected to the target object. The passive sensing radar echo signal set is subjected to signal preprocessing operation to obtain a preprocessed radar echo signal set. A multi-level feature extraction operation is performed on the preprocessed radar echo signal set to generate a multi-dimensional state feature set of the target object. The multi-dimensional state feature set includes scattering characteristic features, motion correlation features, and environmental interference suppression features. The pre-trained target state analysis model is invoked to perform state transition modeling on the multi-dimensional state feature set, generating a reconstructed state vector of the target object. Based on the reconstructed state vector, the geometric parameter set and motion parameter set of the target object are obtained by inverse mapping.

2. The target reconstruction method according to claim 1, characterized in that, The preprocessing operation on the passive sensing radar echo signal set to obtain a preprocessed radar echo signal set includes: Noise suppression processing is performed on each scattered signal in the echo signal sequence to obtain a set of denoised scattered signals. The denoised scattered signal set is subjected to signal segmentation processing to obtain multiple signal segment sets, each signal segment set corresponding to the scattered signal within the same time window; A time-domain alignment operation is performed on the multiple signal segment sets to eliminate the time delay differences between different transmitted signal sources and generate a time-domain aligned radar echo signal set. The amplitude of the time-domain aligned radar echo signal set is normalized to obtain a preprocessed radar echo signal set.

3. The target reconstruction method according to claim 2, characterized in that, The denoised scattered signal set is subjected to signal segmentation processing to obtain multiple signal segment sets, including: The signal segment length threshold is determined based on the target object's maximum speed and the radar signal sampling frequency. Based on the signal segment length threshold, the denoised scattered signal set is segmented by a sliding window to generate an initial signal segment set. Energy density detection is performed on each signal segment in the initial signal segment set. If the energy density of a signal segment is lower than a preset energy threshold, the signal segment is removed. The remaining signal segments are checked for overlap rate. If the overlap rate of adjacent signal segments exceeds a preset overlap threshold, the adjacent signal segments are merged to generate an optimized set of signal segments. The optimized signal segment set is divided into multiple signal segment sets, each corresponding to the scattered signal within the same time window.

4. The target reconstruction method according to claim 1, characterized in that, The process of performing multi-level feature extraction on the preprocessed radar echo signal set to generate a multi-dimensional state feature set of the target object includes: Perform time-frequency joint transformation processing on the preprocessed radar echo signal set to generate a time-frequency distribution feature set; Environmental interference compensation processing is performed on the time-frequency distribution feature set to eliminate interference components caused by ground reflection and atmospheric attenuation, resulting in an environmentally suppressed time-frequency feature set. Scattering characteristic analysis is performed on the time-frequency feature set after environmental suppression to extract the scattering intensity distribution characteristics and multipath scattering path characteristics of the target object surface; Motion trajectory correlation processing is performed on the preprocessed radar echo signal set to extract the velocity change features and acceleration correlation features of the target object; The scattering intensity distribution features, the multipath scattering path features, the velocity change features, and the acceleration correlation features are fused to generate a multidimensional set of state features of the target object.

5. The target reconstruction method according to claim 4, characterized in that, The process of performing scattering characteristic analysis on the time-frequency feature set after environmental suppression to extract the scattering intensity distribution characteristics and multipath scattering path characteristics of the target object surface includes: The set of time-frequency features after environmental suppression is processed by scattering center detection to determine the set of potential scattering center locations on the surface of the target object. Cluster analysis is performed on the set of potential scattering center locations to remove isolated scattering centers and generate an optimized set of scattering centers. The optimized set of scattering centers is subjected to scattering intensity distribution modeling processing to generate scattering intensity distribution features of the target object surface; Based on the optimized set of scattering centers and radar signal propagation path model, the set of multipath reflection path lengths between the target object and the emitted signal source is determined. The multipath scattering path features are generated by calculating the path phase difference set based on the set of multipath reflection path lengths and the radar signal wavelength.

6. The target reconstruction method according to claim 1, characterized in that, The step of calling the pre-trained target state analysis model to perform state transition modeling on the multi-dimensional state feature set to generate a reconstructed state vector of the target object includes: The multi-dimensional state feature set is input into the feature encoding module of the pre-trained target state analysis model to generate the initial state vector of the target object; The state transition module of the target state analysis model is invoked to perform time series modeling on the initial state vector, generating the state transition matrix of the target object within a continuous time window. The initial state vector is corrected based on the state transition matrix to generate a set of intermediate state vectors for the target object. The intermediate state vector set is subjected to spatial consistency verification to remove state vectors that conflict with the motion law of the target object, and an optimized intermediate state vector set is generated. The optimized intermediate state vector set is input into the feature decoding module of the target state analysis model to generate the reconstructed state vector of the target object.

7. The target reconstruction method according to claim 6, characterized in that, The spatial consistency verification process for the intermediate state vector set, which removes state vectors that conflict with the motion law of the target object, includes: A set of state transition constraints is generated based on the kinematic equations of the target object. The set of state transition constraints includes velocity continuity constraints, acceleration range constraints, and motion direction smoothness constraints. Constraint matching is performed on each state vector in the intermediate state vector set, and the matching degree score between each state vector and the state transition constraint set is calculated. If the matching score is lower than the preset matching threshold, the state vector is determined to conflict with the motion law of the target object, and a set of conflicting state vectors is generated. The conflicting state vector set is removed from the intermediate state vector set to generate an optimized intermediate state vector set.

8. The target reconstruction method according to claim 1, characterized in that, The set of geometric parameters and the set of motion parameters of the target object obtained by inverse mapping based on the reconstructed state vector include: The predefined inverse mapping model is invoked to perform geometric structure analysis on the reconstructed state vector, generating a set of three-dimensional spatial coordinates and surface curvature distribution characteristics of the target object; The three-dimensional spatial coordinate set is subjected to topological connection processing to determine the geometric boundary features and vertex connection relationships of the target object, and a set of geometric parameters is generated. The reconstructed state vector is subjected to kinematic analysis to extract the instantaneous velocity vector, acceleration vector, and motion direction angle of the target object, thereby generating a set of motion parameters; The geometric parameter set and the motion parameter set are spatiotemporally aligned to generate a complete reconstructed parameter set for the target object. The topological connection processing of the three-dimensional spatial coordinate set to determine the geometric boundary features and vertex connection relationships of the target object includes: Perform a neighborhood search on the set of three-dimensional spatial coordinates to determine the set of neighboring points for each spatial coordinate point; Based on the set of neighboring points, local topological connections between spatial coordinate points are generated, and an initial topological connection graph is generated. Redundant edges are removed from the initial topology graph to eliminate cross connections and loop connections, generating an optimized topology graph. Based on the optimized topology connection graph, extract the set of geometric boundary line segments and vertex connection relationships of the target object to generate a set of geometric parameters.

9. A target reconstruction system, characterized in that, It includes a processor and a memory and a bus connected to the processor; the processor and the memory communicate with each other through the bus; the processor is used to call a computer program in the memory to execute the target reconstruction method based on passive sensing radar echo signals as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the target reconstruction method based on passive sensing radar echo signals as described in any one of claims 1-8.

Images