Vehicle-mounted mobile platform radar measurement time-varying extension target multi-frame track-before-detect method

Abstract

The invention discloses a vehicle-mounted mobile platform radar measurement time-varying extended target multi-frame track-before-detect method, which comprises the following steps of: firstly, carrying out vehicle-mounted radar system parameter initialization and extended target modeling, then obtaining own vehicle navigation information and radar measurement data, constructing an intra-frame energy accumulation model with orientation self-adaptive capability, and carrying out multi-frame track-before-detect on a time-varying extended target; and finally, performing multi-frame joint accumulation processing based on moving platform motion compensation, outputting an extended target estimation trajectory and appearance parameters, and completing multi-frame track-before-detect processing of the vehicle-mounted moving platform radar measurement time-varying extended target. According to the method, key factors such as extended target measurement time varying and platform motion are comprehensively considered, it is ensured that effective accumulation of measurement time varying extended target echoes can still be achieved in a complex vehicle-mounted radar environment, compared with an existing method, the detection and tracking capacity of a vehicle-mounted moving platform radar on a measurement time varying extended target is remarkably improved, and the detection and tracking efficiency is improved. The method has better detection performance and stronger robustness, and can be applied to the fields of intelligent driving and the like.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of radar target detection and tracking, and particularly relates to a multi-frame detection pre-tracking method for a moving platform radar measuring time-varying extended target. BACKGROUND

[0002] An autonomous vehicle relies on multiple sensors to realize environment perception, and among the sensors, a radar occupies a core position due to its robustness in bad weather and cost advantage. The radar can not only provide accurate distance, speed and angle information, but also is widely used for detection and tracking of vehicles, pedestrians and traffic obstacles. With the improvement of radar resolution, multiple scattering centers of a target can be observed at the same time, and echo energy is distributed on multiple resolution cells to form an extended target. However, this extended characteristic reduces the average signal-to-noise ratio of the resolution cell, and makes the target more susceptible to clutter interference. At the same time, the relative motion between the ego vehicle and the target will cause the measurement to be time-varying, further increasing the difficulty of detection and tracking. It is still a key problem to be solved to realize robust perception of the measurement time-varying extended target under the moving platform condition. Most existing vehicle-mounted radars still adopt a "two-stage detection-tracking" framework: first, constant false alarm rate detection is performed on single-frame data to extract point tracks, and then clustering and filtering are used to estimate trajectories and shapes. However, in the case of low signal-to-noise ratio, the initial threshold processing often leads to information loss, and then causes weak target missed detection and shape estimation deviation, which seriously affects the detection and tracking performance of the measurement time-varying extended target.

[0003] Multi-frame track-before-detect (TBD) is an effective technique for detecting targets in low signal-to-noise ratio (SNR) scenarios. The biggest difference between multi-frame TBD and the existing two-stage detection-tracking method is that multi-frame TBD does not perform thresholding on single-frame data, but performs non-coherent energy accumulation on multi-frame raw radar echoes, separates real target echoes from noise and clutter by using the difference between them, and effectively avoids the information loss caused by single-frame echo data thresholding, so it can be used to detect weak target signals. The paper "Pseudo-spectrum based speed square filter for track-before-detect in range-doppler domain, IEEE Transactions on Signal Processing, 2019, 67(21): 5596-5610" proposes a pseudo-spectrum-based multi-frame TBD method for extended targets. Under the assumption of known target extension shape, this method uses the point spread function model to construct the pseudo-spectrum in the range-Doppler plane, and collects the pseudo-spectrum energy of multiple periods and multiple units to improve the detection performance. However, this method relies too much on initial information and prior extension model. This method relies on prior extension model, which is difficult to adapt to complex scenarios and is prone to model mismatch and insufficient energy accumulation. The paper "Multi-frame joint tracking and shape estimation method for weak extended targets, in 2021IEEE 24th International Conference on Information Fusion, 2021, pp.1-7" proposes a multi-frame joint detection and shape estimation method for elliptical extended targets. This method uses the correlation response map generated by the multi-scale kernel correlation filter as the multi-frame detection statistic to realize the joint estimation of trajectory and shape parameters under the condition that the target shape is unknown. However, the response map form is easily affected by the bandwidth, scale range and SNR of the kernel function. When the parameters are not properly set or the target is weak, the response map peak value may spread or be false, which will affect the detection and tracking performance. It should be noted that the above methods are mainly proposed for ground-based stationary radar platforms.The paper "Automotive Radar Multi-Frame Track-Before-Detect Algorithm Considering Self-Positioning Errors" in IEEE Transactions on Intelligent Transportation Systems, 2025 proposes a multi-frame detection pre-tracking method for vehicle-mounted mobile platform radar that considers the vehicle's positioning error. This method improves target detection performance by jointly modeling the vehicle's position, velocity, and yaw angle errors during energy accumulation and adaptively adjusting the detection threshold. However, this method does not adequately model the scale differences and shape changes of the extended target under the temporal distribution, which can still easily lead to a decrease in detection and tracking performance in complex vehicle scenarios.

[0004] In summary, most existing multi-frame pre-detection tracking methods neglect the impact of extended target orientation and position changes on measurement distribution, thus limiting the performance of extended target detection and tracking. Specifically: a) Target orientation impact: Existing methods typically assume that the energy distribution of the extended target remains constant across different scans, ignoring measurement changes caused by differences in target (e.g., vehicle) orientation and motion. This can easily lead to insufficient energy accumulation under low signal-to-noise ratio conditions; b) Inconsistent near and far-field scales: There is a nonlinear mapping relationship between the range-Doppler-azimuth data acquired by the radar system and the rigid body shape, resulting in differences between near-field and far-field observations, thus affecting energy accumulation and recognition consistency; c) Platform motion effect: The vehicle-mounted radar platform is in continuous motion, which causes field-of-view shift. Without compensation, this leads to velocity estimation errors and position drift, disrupting trajectory continuity and causing measurement misalignment. Therefore, the above methods have not yet effectively solved the problem of time-varying extended target detection for vehicle-mounted moving platform radars. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a multi-frame pre-detection tracking method for time-varying extended targets measured by vehicle-mounted moving platform radar. This method solves the problems of insufficient energy accumulation, trajectory drift, and decreased detection performance caused by existing extended target detection and tracking methods in vehicle-mounted radar applications, which do not fully consider key factors such as the time-varying nature of extended target measurements and platform motion.

[0006] The technical solution adopted in this invention is: a method for tracking time-varying extended targets before multi-frame detection using vehicle-mounted dynamic platform radar measurement, the specific steps of which are as follows:

[0007] Step 1: Initialize vehicle-mounted radar system parameters and expand target modeling;

[0008] Step 11: Setting up the scene and coordinate system;

[0009] The vehicle uses a frequency-modulated continuous wave radar system mounted on its vehicle to detect moving targets on the road. Both the vehicle and the target are set in the global coordinate system O. A -X A Y A The vehicle is in motion. The vehicle coordinate system is denoted as O. V -X V Y V Its reference point is the vehicle center O. V , where X V The axis is along the longitudinal direction of the vehicle, Y V The axis runs laterally along the vehicle. The radar coordinate system is denoted as O. R -X R Y R .

[0010] Step 12: Expand the target model construction;

[0011] Vehicle targets are described using a rectangular model, and the targets are assumed to maintain rigid body motion in a Cartesian coordinate system. A multi-category extended target model set {c} is introduced. j {j = 1, ..., J}, where J represents the total number of categories, and c is the target of each category. j =[l j ,w j [Corresponds to a fixed set of length parameters l] j With width parameter w j And the bounding box is determined by the target centroid state x in the k-th frame. k and its corresponding vertex coordinate set The only certainty.

[0012] in, These represent the coordinates of the four vertices of the rectangle relative to the centroid.

[0013] Step 13: Initialize system parameters;

[0014] Initialize the vehicle radar system parameters, including: radar inter-frame interval T. s Total number of observation frames K, radar range resolution Δr, radar azimuth resolution Δa, vehicle radar installation angle α, extended target category c j The initial centroid state x1, the initial centroid state y1 of the vehicle, the extended target motion angular velocity ω, and the detection threshold V calculated using Monte Carlo simulation experiments. T The current frame number k = 1.

[0015] Step 2: Acquisition of vehicle navigation information and radar measurement data;

[0016] In the vehicle radar system, the vehicle's current motion information is obtained through the navigation system. k and yaw angle β kSimultaneously, the target's motion information is obtained by radar measurements mounted on the vehicle, and the k-th frame echo measurement data acquired by the vehicle's radar receiver is denoted as z. k , z k ={z k (r,a),r∈[1,N r ], a∈[1,N a ]}.

[0017] Where r represents the distance cell number, N r N represents the total number of range-resolved units, 'a' represents the azimuth unit number, and N represents the total number of range-resolved units. a z represents the total number of resolvable elements in the azimuth dimension. k (r,a) represents the amplitude value of the radar measurement data corresponding to the range cell index r and the azimuth cell index a in the k-th frame.

[0018] Extending the spatial energy diffusion of the target, the radar echo energy is distributed across multiple adjacent resolution cells, thus forming an extended measurement set Ω. k The expression is as follows:

[0019]

[0020] in, Let represent the normalized scattering factor of the target amplitude of the measurement unit (r,a) in the k-th frame, and satisfy . when When the value is 0, it indicates that there is no target in that unit.

[0021] The target echo is then modeled as a Rice distribution, and its likelihood function is represented as follows:

[0022]

[0023] Where f(·) represents the conditional probability density function, and A represents the amplitude of the extended target centroid. c represents the background noise power. j I0(·) represents the extended target category, and I0(·) represents the first-order modified Bessel function of order 0.

[0024] Step 3: Model the extended target intra-frame likelihood ratio function and accumulate intra-frame energy based on the target motion orientation;

[0025] First, a set of multiple velocity assumptions is constructed, and the target state space is discretized. Then, the coordinate system of radar measurement and velocity assumption modeling is unified. Finally, target orientation information is introduced to determine the scattering area of ​​the extended target in the radar polar coordinate system, and an intra-frame energy accumulation model with orientation adaptation capability is constructed.

[0026] Step 4: Perform multi-frame joint accumulation processing based on motion compensation of the moving platform, output the estimated trajectory and shape parameters of the extended target, and complete the multi-frame pre-detection tracking processing of the time-varying extended target measured by the vehicle-mounted moving platform radar.

[0027] The problem of multi-frame accumulation under moving platform conditions is solved by modifying the coordinate system used for state-space representation. That is, the corresponding measurement data is searched along the relative motion trajectory of the current state to realize inter-frame state transition and effective energy accumulation, and output extended target estimated trajectory and shape parameters.

[0028] Furthermore, step 3 is specifically as follows:

[0029] Step 31: Construct a set of multiple velocity hypotheses and discretize the target state space;

[0030] In O A -X A Y A Set up a set of multiple velocity assumptions in the absolute coordinate system. Let the velocity be along X... A The range of the shaft's velocity values ​​is discretized as follows There are candidate values, denoted as _____. Along Y A The range of the shaft's velocity values ​​is discretized as follows There are candidate values, denoted as _____. Then define along X A axis and Y A The velocity of the axis is a set of multiple assumptions. Given radar measurement z k (r,a) and a set of velocity assumptions The discretized grid space of the target state Described as an expression:

[0031]

[0032] in, Let R represent any discretized state vector of the k-th frame, where R represents the radial distance and A represents the azimuth angle. and They represent along X A The assumed speed value, speed index, and speed resolution of the axis. and They represent along Y A The assumed speed value, speed index, and speed resolution of the axis; T indicates the transpose operation.

[0033] Step 32: Based on step 31, establish coordinate system one;

[0034] Radar measurements use the radar as a reference point at O R -X R Y RRepresented in relative polar coordinates, while velocity is assumed to be modeled in O. A -X A Y A In an absolute coordinate system, a uniform transformation is therefore performed on the coordinates.

[0035] First, the position coordinates in the radar polar coordinate system Transform to Cartesian coordinates, the corresponding state vector The expression is as follows:

[0036]

[0037] in, This represents a coordinate transformation operator used to convert polar coordinates to Cartesian coordinates.

[0038] Next, the transformed Cartesian coordinates are then used to consider the radar installation angle α and the vehicle's yaw angle β. k The influence of this process is represented as a rotation matrix operation. The rotation matrix R(θ) is defined as follows:

[0039]

[0040] Where θ represents the rotation angle.

[0041] Then, the motion of the self-driving platform is compensated to obtain... In O A -X A Y A The absolute position in the global coordinate system, and its state vector The expression is as follows:

[0042]

[0043] Where h(·) represents the transformation process function, which is the transformation process of mapping the radar polar coordinate state to the global Cartesian coordinate state.

[0044] Finally, combining the velocity assumption Get O A -X A Y A Final state vector in global coordinate system The expression is as follows:

[0045]

[0046] Step 33: Based on step 32, introduce target orientation information to determine O. R -X R Y RThe scattering region of the extended target in the radar polar coordinate system is used to construct an intra-frame energy accumulation model with orientation adaptation capability;

[0047] It is known that the orientation of the extended target is determined by the direction of its center-of-mass velocity, combined with the assumed velocity vector. Target orientation angle d k Represented as After introducing the orientation angle, the target type c is extended. j The positions of the vertices of the rectangle relative to the centroid are corrected using a rotation matrix, as shown in the following expression:

[0048]

[0049] in, This represents the set of vertices of the rectangle after correction in the k-th frame. These represent the coordinates of the i-th vertex after rotation correction.

[0050] Further compensation of the target centroid position yields O A -X A Y A Absolute position of the i-th vertex in the global coordinate system The expression is as follows:

[0051]

[0052] in, This indicates that the target centroid's position coordinates in the global coordinate system are determined by step 32. This yields the set of vertex coordinates of the target rectangle.

[0053] Then, through the inverse transformation process function h in step 32 -1 (·), representing the set of vertex coordinates Transform to the radar polar coordinate system to obtain the vertex coordinate set.

[0054] Among them, h -1 (·) indicates that step 32 starts from O A -X A Y A coordinate system to O R -X R Y R The function for the inverse transformation of the coordinate system.

[0055] After determining the set of vertex coordinates of the rectangular frame, the extended target scattering region is then derived. The extended target scattering region in the polar coordinate system of the vehicle-mounted radar is then obtained. From the state of the center of mass Car navigation information y k With target category c jThe region constraint, which is fully determined and used as the intra-frame likelihood ratio function, is expressed as follows:

[0056]

[0057] Where g(·) represents the implicit function generated in the scattering region.

[0058] Finally, based on the generalized likelihood ratio test and combined with equation (10), the intra-frame likelihood ratio function of the extended target frame after incorporating orientation angle information is derived. The expression is as follows:

[0059]

[0060] Furthermore, step 4 is specifically as follows:

[0061] Step 41: Initialize the value function and state transition relationship;

[0062] When k=1, use each discrete state Value function of echo data in the corresponding first frame Relationship with state transition Initialization, i.e. After initialization is complete, proceed to step 42.

[0063] Step 42: Inter-frame iterative accumulation;

[0064] When 2≤k≤K, the value function of each discrete state is the current frame state. The intra-frame likelihood ratio function and the set of all possible state transitions accumulated from the previous frame for this state. The sum of the value functions within, i.e.:

[0065]

[0066]

[0067] in, This represents any discretized grid state of the k-th frame. This represents the intra-frame likelihood ratio function derived in step 33. Describes the cumulative value function plane of the k-th frame. Used to store inter-frame state transition relationships. This indicates that the target transitions from frame (k-1) to frame k. The set of all possible state transitions.

[0068] for To obtain the current discrete state vector, firstly, calculate the current discrete state vector. In O A -X A YA Position coordinates in global coordinate system Then, based on the target motion model, its possible set of state transitions in the previous frame is inferred, as follows:

[0069] Where F represents the state transition equation, δ represents the preset boundary threshold vector. x and They represent X respectively A The permissible error range for position coordinates and velocity components along the axial direction, δ y and They represent Y respectively A The permissible error range for position coordinates and velocity components along the axial direction.

[0070] Step 43: Based on step 42, perform threshold decision on the accumulation value function;

[0071] When k = K, if the accumulated value function of the Kth frame exceeds the detection threshold η, then the target is determined to exist, and the state of the corresponding maximum value function is extracted. The optimal target state for the last frame is expressed as follows:

[0072]

[0073] Step 44: Based on step 43, perform trajectory backtracking on the accumulated value function;

[0074] When a target is detected, the state transition relationship function is traced back. The short track sequence in radar polar coordinates was recovered. The expression is as follows:

[0075]

[0076] Step 45: Based on step 44, perform coordinate transformation and output the estimated trajectory and shape parameters of the extended target;

[0077] First, the estimated short track sequence Perform the coordinate transformation in step 32, changing it from O R -X R Y R Radar relative polar coordinate system transformation to O A -X A Y A Using a global Cartesian coordinate system, obtain the corresponding target state sequence. Achieve accurate recovery of the time-varying extended target trajectory under vehicle-mounted dynamic platform conditions.

[0078] Then, from the short track sequence Extract velocity information to determine the target's orientation angle sequence. Combined with the extended target type c j It can obtain the estimated shape parameters of extended targets under time-varying measurement conditions, and realize multi-frame pre-detection tracking processing of extended targets measured by vehicle-mounted moving platform radar.

[0079] in, This represents the target orientation angle estimation result for the k-th frame.

[0080] The beneficial effects of this invention are as follows: This invention discloses a multi-frame pre-tracking method for time-varying extended targets measured by vehicle-mounted moving platform radar. First, the vehicle-mounted radar system parameters are initialized and the extended target is modeled. Then, the vehicle navigation information and radar measurement data are acquired to construct an intra-frame energy accumulation model with orientation adaptation capability. Finally, multi-frame joint accumulation processing based on moving platform motion compensation is performed to output the estimated trajectory and shape parameters of the extended target, thus completing the multi-frame pre-tracking processing for time-varying extended targets measured by vehicle-mounted moving platform radar. The method of this invention comprehensively considers key factors such as the orientation of the extended target, the difference in scale between near and far zones, and platform motion, ensuring effective accumulation of time-varying extended target echoes even in complex vehicle-mounted radar environments. It solves the problems of decreased detection and tracking performance caused by platform motion, scale deformation, and measurement mismatch between near and far zones in existing methods. It also overcomes the problem of performance degradation caused by the failure to consider the impact of platform motion and changes in the orientation of the extended target on the measurement distribution in existing methods. This significantly improves the detection and tracking capability of vehicle-mounted moving platform radar for time-varying extended targets, and has better detection performance and stronger robustness. It can be applied to fields such as intelligent driving and is an algorithm with high robustness and practical value. Attached Figure Description

[0081] Figure 1 This is a flowchart of a method for tracking a time-varying extended target before multi-frame detection using radar measurement on a vehicle-mounted moving platform, according to the present invention.

[0082] Figure 2 This is a schematic diagram of the vehicle-mounted radar sensing scenario used in an embodiment of the present invention.

[0083] Figure 3 This is a schematic diagram of the range-azimuth measurement plane of near-field target 1 in three consecutive frames in an embodiment of the present invention.

[0084] Figure 4 This is a schematic diagram of the range-azimuth measurement plane of the distant target 3 in three consecutive frames in an embodiment of the present invention.

[0085] Figure 5 This is a schematic diagram showing the estimated trajectory and shape results of near-field target 1 and far-field target 3 under the condition of a signal-to-noise ratio of 9dB, using the method of the present invention and the traditional multi-frame detection pre-tracking algorithm in an embodiment of the present invention.

[0086] Figure 6 This is a schematic diagram of the simulation performance curves of different extended target types in the radar near-field scenario in an embodiment of the present invention.

[0087] Figure 7 This is a schematic diagram of the simulation performance curves of different extended target types in the radar far-field scenario in an embodiment of the present invention. Detailed Implementation Plan

[0088] The method of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0089] This invention is primarily verified using simulation experiments, and all steps and conclusions have been verified correctly on the Matlab 2023a platform. To facilitate the description of the method of this invention, the following terms are first explained:

[0090] Term 1: Extended Target;

[0091] Extended targets refer to targets whose physical size is larger than a single resolution cell of a radar. Their echo energy is distributed across multiple resolution cells, and the echo characteristics are manifested as an energy distribution formed by multiple scattering points.

[0092] Term 2: Measurement of time variation;

[0093] This refers to the phenomenon where the echo energy distribution of a target changes over time during multi-frame observation. This change leads to inconsistencies in the number of resolution cells occupied by the target and the energy distribution pattern in different frames, and is a key factor affecting extended target energy accumulation and detection and tracking performance.

[0094] Term 3: Scale difference between near and far zones;

[0095] This refers to the phenomenon where the number of resolution cells occupied by a target in the near and far zones of a radar system is inconsistent. Since vehicle-mounted radar systems typically operate in a polar coordinate system, under constant angular resolution, the arc length increases with distance. This results in the same target having a larger projection range in the near zone, covering more resolution cells, while occupying only fewer resolution cells in the far zone.

[0096] Terminology; 4: Multiple Hypothesis Velocity Mapping;

[0097] This method involves setting multiple velocity assumptions when the target's velocity is unknown or uncertain, and mapping the measurement data to the corresponding velocity spaces to guide energy accumulation and trajectory estimation. This approach improves the adaptability of target motion state modeling and avoids energy accumulation bias and detection performance degradation caused by changes in target velocity.

[0098] Term 5: Accumulated value function;

[0099] The accumulation value function is a metric function formed during the energy accumulation process of tracking before multi-frame detection. It is used to characterize the probability of the target's existence and serves as the basis for subsequent detection decisions.

[0100] like Figure 1 The flowchart shown is a pre-detection tracking method for time-varying extended targets using radar measurement of a vehicle-mounted moving platform according to the present invention. The specific steps are as follows:

[0101] Step 1: Initialize vehicle-mounted radar system parameters and expand target modeling;

[0102] Step 11: Setting up the scene and coordinate system;

[0103] like Figure 2 As shown, this embodiment uses a frequency-modulated continuous wave radar system installed on the vehicle to detect moving targets on the road, with both the vehicle and the target set in the global coordinate system O. A -X A Y A The vehicle is in motion. The vehicle coordinate system is denoted as O. V -X V Y V Its reference point is the vehicle center O. V , where X V The axis is along the longitudinal direction of the vehicle, Y V The axis runs laterally along the vehicle. The radar coordinate system is denoted as O. R -X R Y R .

[0104] Step 12: Expand the target model construction;

[0105] like Figure 2 As shown, in this embodiment, vehicle targets are described using a rectangular model, and the targets are assumed to maintain rigid body motion in a Cartesian coordinate system. To accommodate extended targets of different sizes and structural characteristics, a multi-category extended target model set {c} is introduced. j {j = 1, ..., J}, where J represents the total number of categories, and c is the target of each category. j =[l j ,w j [Corresponds to a fixed set of length parameters l] j With width parameter w j And the bounding box is determined by the target centroid state x in the k-th frame. k and its corresponding vertex coordinate set The unique determination is as follows. In this embodiment, J = 2, then c1 = [4m, 2m] and c2 = [8m, 2m].

[0106] in, These represent the coordinates of the four vertices of the rectangle relative to the centroid.

[0107] Step 13: Initialize system parameters;

[0108] Initialize the vehicle radar system parameters, including: radar inter-frame interval T. s Total number of observation frames K, radar range resolution Δr, radar azimuth resolution Δa, vehicle radar installation angle α, extended target category c j The initial centroid state x1, the initial centroid state y1 of the vehicle, the extended target motion angular velocity ω, and the detection threshold V calculated using Monte Carlo simulation experiments. T The current frame number k = 1.

[0109] In this embodiment, the radar inter-frame interval T s =0.1s, total number of observation frames K=6, radar range resolution Δr=0.5m, radar azimuth resolution Δa=4.5°, vehicle radar installation angle α=-28°, extended target motion angular velocity ω=0.3rad / s, initial centroid state of near-field target c1 x1=[39m,2m / s,10m,24m / s] T The initial centroid state of target 2 in the near region c2 is x1 = [39m, 2m / s, 10m, 24m / s]. T The initial centroid state of target 3 in the far region c1 is x1 = [39m, 2m / s, 110m, 24m / s]. T The initial centroid state of target 4 in the far region c2 is x1 = [39m, 2m / s, 110m, 24m / s]. T The initial center of mass of the vehicle is y1 = [11m, 0.1m / s, 1.8m, 14m / s]. T The detection threshold V calculated using Monte Carlo simulation experiments T =112.47, current frame number k=1.

[0110] Step 2: Acquisition of vehicle navigation information and radar measurement data;

[0111] In the vehicle radar system, the vehicle's current motion information is obtained through the navigation system. k and yaw angle β k Simultaneously, the target's motion information is obtained by radar measurements mounted on the vehicle, and the k-th frame echo measurement data acquired by the vehicle's radar receiver is denoted as z. k , z k ={z k (r,a),r∈[1,N r ], a∈[1,N a ]}.

[0112] Where r represents the distance cell number, N r N represents the total number of range-resolved units, 'a' represents the azimuth unit number, and N represents the total number of range-resolved units.a z represents the total number of resolvable elements in the azimuth dimension. k (r,a) represents the amplitude value of the radar measurement data corresponding to the range cell index r and the azimuth cell index a in the k-th frame.

[0113] Due to the spatial energy diffusion of the extended target, the radar echo energy is distributed across multiple adjacent resolution cells, thus forming an extended measurement set Ω. k The expression is as follows:

[0114]

[0115] in, Let represent the normalized scattering factor of the target amplitude of the measurement unit (r,a) in the k-th frame, and satisfy . when When the value is 0, it indicates that there is no target in that unit.

[0116] The target echo is then modeled as a Rice distribution, and its likelihood function is represented as follows:

[0117]

[0118] Where f(·) represents the conditional probability density function, and A represents the amplitude of the extended target centroid. c represents the background noise power. j I0(·) represents the extended target category, and I0(·) represents the first-order modified Bessel function of order 0.

[0119] Step 3: Model the extended target intra-frame likelihood ratio function and accumulate intra-frame energy based on the target motion orientation;

[0120] First, a set of multiple velocity assumptions is constructed, and the target state space is discretized. Then, the coordinate system of radar measurement and velocity assumption modeling is unified. Finally, target orientation information is introduced to determine the scattering area of ​​the extended target in the radar polar coordinate system, and an intra-frame energy accumulation model with orientation adaptation capability is constructed.

[0121] Step 4: Perform multi-frame joint accumulation processing based on motion compensation of the moving platform, output the estimated trajectory and shape parameters of the extended target, and complete the multi-frame pre-detection tracking processing of the time-varying extended target measured by the vehicle-mounted moving platform radar.

[0122] Unlike existing methods that directly accumulate echo energy on the state-measurement plane, this embodiment solves the multi-frame accumulation problem under moving platform conditions by modifying the coordinate system used for state-space representation. That is, it searches for corresponding measurement data along the relative motion trajectory of the current state, realizes inter-frame state transition and effective energy accumulation, and outputs extended target estimated trajectory and shape parameters.

[0123] In this embodiment, step 3 is specifically as follows:

[0124] Step 31: Construct a set of multiple velocity hypotheses and discretize the target state space;

[0125] In O A -X A Y A Set up a set of multiple velocity assumptions in the absolute coordinate system. Let the velocity be along X... A The range of the shaft's velocity values ​​is discretized as follows There are candidate values, denoted as _____. Along Y A The range of the shaft's velocity values ​​is discretized as follows There are candidate values, denoted as _____. Then define along X A axis and Y A The velocity of the axis is a set of multiple assumptions. Given radar measurement z k (r,a) and a set of velocity assumptions The discretized grid space of the target state Described as an expression:

[0126]

[0127] in, Let R represent any discretized state vector of the k-th frame, where R represents the radial distance and A represents the azimuth angle. and They represent along X A The assumed speed value, speed index, and speed resolution of the axis. and They represent along Y A The assumed speed value, speed index, and speed resolution of the axis; T indicates the transpose operation.

[0128] Step 32: Based on step 31, establish coordinate system one;

[0129] Radar measurements use the radar as a reference point at O R -X R Y R Represented in relative polar coordinates, while velocity is assumed to be modeled in O. A -X A Y A In an absolute coordinate system, a uniform transformation is therefore performed on the coordinates.

[0130] First, the position coordinates in the radar polar coordinate system Transform to Cartesian coordinates, the corresponding state vector The expression is as follows:

[0131]

[0132] in, This represents a coordinate transformation operator used to convert polar coordinates to Cartesian coordinates.

[0133] Next, the transformed Cartesian coordinates are then used to consider the radar installation angle α and the vehicle's yaw angle β. k The influence of this process is represented as a rotation matrix operation. The rotation matrix R(θ) is defined as follows:

[0134]

[0135] Where θ represents the rotation angle.

[0136] Then, the motion of the self-driving platform is compensated to obtain... In O A -X A Y A The absolute position in the global coordinate system, and its state vector The expression is as follows:

[0137]

[0138] Where h(·) represents the transformation process function, which is the transformation process of mapping the radar polar coordinate state to the global Cartesian coordinate state.

[0139] Finally, combining the velocity assumption Get O A -X A Y A Final state vector in global coordinate system The expression is as follows:

[0140]

[0141] Step 33: Based on step 32, introduce target orientation information to determine O. R -X R Y R The scattering region of the extended target in the radar polar coordinate system is used to construct an intra-frame energy accumulation model with orientation adaptation capability;

[0142] It is known that the orientation of the extended target is determined by the direction of its center-of-mass velocity, combined with the assumed velocity vector. Target orientation angle d k Represented as After introducing the orientation angle, the target type c is extended. j The positions of the vertices of the rectangle relative to the centroid are corrected using a rotation matrix, as shown in the following expression:

[0143]

[0144] in, This represents the set of vertices of the rectangle after correction in the k-th frame. These represent the coordinates of the i-th vertex after rotation correction.

[0145] Further compensation of the target centroid position yields O A -X A Y A Absolute position of the i-th vertex in the global coordinate system The expression is as follows:

[0146]

[0147] in, This indicates that the target centroid's position coordinates in the global coordinate system are determined by step 32. This yields the set of vertex coordinates of the target rectangle.

[0148] Then, through the inverse transformation process function h in step 32 -1 (·), representing the set of vertex coordinates Transform to the radar polar coordinate system to obtain the vertex coordinate set.

[0149] Among them, h -1 (·) indicates that step 32 starts from O A -X A Y A coordinate system to O R -X R Y R The function for the inverse transformation of the coordinate system.

[0150] After determining the set of vertex coordinates of the rectangular frame, the extended target scattering region is then derived. The extended target scattering region in the polar coordinate system of the vehicle-mounted radar is then obtained. From the state of the center of mass Car navigation information y k With target category c j The region constraint, which is fully determined and used as the intra-frame likelihood ratio function, is expressed as follows:

[0151]

[0152] Where g(·) represents the implicit function generated in the scattering region.

[0153] Finally, based on the generalized likelihood ratio test and combined with equation (10), the intra-frame likelihood ratio function of the extended target frame after incorporating orientation angle information is derived. The expression is as follows:

[0154]

[0155] Therefore, the intra-frame likelihood ratio function is essentially a weighted accumulation of energy within the target scattering region, thus forming an intra-frame energy accumulation model to facilitate subsequent detection and tracking of extended targets.

[0156] In this embodiment, step 4 is specifically as follows:

[0157] Step 41: Initialize the value function and state transition relationship;

[0158] When k=1, use each discrete state Value function of echo data in the corresponding first frame Relationship with state transition Initialization, i.e. After initialization is complete, proceed to step 42.

[0159] Step 42: Inter-frame iterative accumulation;

[0160] When 2≤k≤K, the value function of each discrete state is the current frame state. The intra-frame likelihood ratio function and the set of all possible state transitions accumulated from the previous frame for this state. The sum of the value functions within, i.e.:

[0161]

[0162] in, This represents any discretized grid state of the k-th frame. This represents the intra-frame likelihood ratio function derived in step 33. Describes the cumulative value function plane of the k-th frame. Used to store inter-frame state transition relationships. This indicates that the target transitions from frame (k-1) to frame k. The set of all possible state transitions.

[0163] for To obtain the current discrete state vector, firstly, calculate the current discrete state vector. In O A -X A Y A Position coordinates in global coordinate system Then, based on the target motion model, its possible set of state transitions in the previous frame is inferred, as follows:

[0164] Where F represents the state transition equation, This represents a preset boundary threshold vector, used to adapt to discretization errors and random noise during motion; δ x and They represent X respectivelyA The permissible error range for position coordinates and velocity components along the axial direction, δ y and They represent Y respectively A The permissible error range for position coordinates and velocity components along the axial direction.

[0165] Step 43: Based on step 42, perform threshold decision on the accumulation value function;

[0166] When k = K, if the accumulated value function of the Kth frame exceeds the detection threshold η, then the target is determined to exist, and the state of the corresponding maximum value function is extracted. The optimal target state for the last frame is expressed as follows:

[0167]

[0168] Step 44: Based on step 43, perform trajectory backtracking on the accumulated value function;

[0169] When a target is detected, the state transition relationship function is traced back. The short track sequence in radar polar coordinates was recovered. The expression is as follows:

[0170]

[0171] Step 45: Based on step 44, perform coordinate transformation and output the estimated trajectory and shape parameters of the extended target;

[0172] First, the estimated short track sequence Perform the coordinate transformation in step 32, changing it from O R -X R Y R Radar relative polar coordinate system transformation to O A -X A Y A Using a global Cartesian coordinate system, obtain the corresponding target state sequence. Achieve accurate recovery of the time-varying extended target trajectory under vehicle-mounted dynamic platform conditions.

[0173] Then, from the short track sequence Extract velocity information to determine the target's orientation angle sequence. Combined with the extended target type c j It can obtain the estimated shape parameters of extended targets under time-varying measurement conditions, and realize multi-frame pre-detection tracking processing of extended targets measured by vehicle-mounted moving platform radar.

[0174] in, This represents the target orientation angle estimation result for the k-th frame.

[0175] To verify the effectiveness of the method of the present invention, a simulation experiment was also conducted in this embodiment, and a comparative analysis was performed with the traditional multi-frame detection pre-tracking algorithm.

[0176] like Figure 3 and Figure 4 As shown, the range-azimuth measurement plane distribution of near-field target 1 and far-field target 3 in three consecutive frames is presented. Figure 3 (a) Figure 4 (a) is frame 1. Figure 3 (b) Figure 4 (b) is the second frame. Figure 3 (c) Figure 4 (c) is frame 3. Due to the relative motion between the vehicle and the target, the spatial distribution of the extended target measurement exhibits obvious time-varying characteristics, with significant differences between the near and far scales. In the near scene, the target energy mainly extends along the azimuth dimension, while in the far scene, the extension mainly extends along the range dimension, resulting in a more elongated shape for the target.

[0177] like Figure 5 As shown, the method of the present invention and the traditional multi-frame detection pre-tracking algorithm estimate the trajectory and shape of near-field target 1 and far-field target 3 under the condition of signal-to-noise ratio of 9dB. Figure 5 (a) shows the estimated trajectory and shape of target 1 in the near area. Figure 5 (b) shows the estimated trajectory and shape of the distant target 3. Experimental results show that the method of this invention is significantly better than the traditional multi-frame pre-detection tracking method in terms of the estimation accuracy of the target shape and orientation. This is because the traditional multi-frame pre-detection tracking method usually uses a fixed range-azimuth extended target model for energy accumulation, which is difficult to adapt to the dynamic changes in the measurement distribution of moving targets, thus producing a large deviation in shape and orientation estimation.

[0178] Furthermore, to systematically evaluate the performance of the method of the present invention, a Monte Carlo simulation experiment was also conducted in this embodiment. For example... Figure 6 and Figure 7 As shown, the curves of detection probability and root mean square error versus signal-to-noise ratio (SNR) of the method of this invention and the traditional multi-frame pre-detection tracking algorithm are compared in radar near-field and far-field scenarios. In the near-field scenario, as... Figure 6 As shown in (a) (the curve of detection probability versus signal-to-noise ratio), when the detection probability is 0.6, the method of this invention achieves a signal-to-noise ratio gain of approximately 3 dB compared to the traditional multi-frame pre-detection tracking algorithm; in terms of tracking performance, as... Figure 6As shown in (b) (curve of root mean square error versus signal-to-noise ratio), the method of this invention maintains a lower root mean square error under different signal-to-noise ratios, demonstrating higher estimation accuracy for the target state. This performance improvement is mainly attributed to the significant geometric distortion generated by near-field targets under relative motion. Traditional fixed extended models cannot effectively align the measurement distribution, while the method of this invention, based on a scale-adaptive and orientation-aware measurement model, can flexibly characterize the target shape changes, thereby achieving sufficient accumulation of target energy and obtaining better detection and tracking performance.

[0179] In distant scenes, although target deformation is relatively minor, traditional multi-frame pre-detection tracking algorithms still suffer from energy mismatch due to fixed extended models. The method of this invention maintains superior performance by capturing scale changes across distance and accurately modeling energy distribution, such as... Figure 7 As shown in (a) (the curve of detection probability versus signal-to-noise ratio), it achieves a signal-to-noise ratio gain of approximately 1.5 dB compared to traditional multi-frame pre-detection tracking algorithms. Regarding tracking performance, as... Figure 7 As shown in (b) (curve of root mean square error versus signal-to-noise ratio), the method of the present invention also has a lower estimation error, effectively improving the tracking accuracy of distant targets. By combining a cross-frame registration mechanism of multi-hypothesis velocity mapping and moving platform motion compensation, the method of the present invention ensures reliable alignment of target measurements in the time dimension, and can still achieve stable detection and tracking performance even under moving platform and low signal-to-noise ratio conditions.

[0180] In summary, the method of this invention explicitly models the orientation and near-far scale variations of the extended target, and combines multiple hypothesis velocity mapping and motion compensation mechanisms to achieve adaptive multi-frame energy accumulation for time-varying extended targets measured by vehicle-mounted moving platform radar. This invention effectively solves the problems of insufficient energy accumulation, trajectory drift, and decreased detection performance caused by existing multi-frame pre-detection tracking methods for extended targets in vehicle-mounted radar applications, which fail to consider factors such as time-varying extended target measurements and platform motion. Therefore, it achieves stable detection, continuous tracking, and accurate shape estimation of extended targets, making it an algorithm with high robustness and practical value.

[0181] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A method for tracking a time-varying extended target before multi-frame detection using radar measurement on a vehicle-mounted moving platform, comprising the following specific steps: Step 1: Initialize vehicle-mounted radar system parameters and expand target modeling; Step 11: Setting up the scene and coordinate system; The vehicle uses a frequency-modulated continuous wave radar system mounted on its vehicle to detect moving targets on the road. Both the vehicle and the target are set in the global coordinate system O. A -X A Y A The vehicle coordinate system is denoted as O. V -X V Y V Its reference point is the vehicle center O. V , where X V The axis is along the longitudinal direction of the vehicle, Y V The axis runs laterally along the vehicle; the radar coordinate system is denoted as O. R -X R Y R ; Step 12: Expand the target model construction; Vehicle targets are described using a rectangular model, and the targets are assumed to maintain rigid body motion in a Cartesian coordinate system. A multi-category extended target model set {c} is introduced. j {j = 1, ..., J}, where J represents the total number of categories, and c is the target of each category. j =[l j ,w j [Corresponds to a fixed set of length parameters l] j With width parameter w j And the bounding box is determined by the target centroid state x in the k-th frame. k and its corresponding vertex coordinate set Uniquely certain; in, These represent the coordinates of the four vertices of the rectangle relative to the centroid; Step 13: Initialize system parameters; Initialize the vehicle radar system parameters, including: radar inter-frame interval T. s Total number of observation frames K, radar range resolution Δr, radar azimuth resolution Δa, vehicle radar installation angle α, extended target category c j The initial centroid state x1, the initial centroid state y1 of the vehicle, the extended target motion angular velocity ω, and the detection threshold V calculated using Monte Carlo simulation experiments. T Current frame number k = 1; Step 2: Acquisition of vehicle navigation information and radar measurement data; In the vehicle radar system, the vehicle's current motion information is obtained through the navigation system. k and yaw angle β k Simultaneously, the target's motion information is obtained by radar measurements mounted on the vehicle, and the k-th frame echo measurement data acquired by the vehicle's radar receiver is denoted as z. k , z k ={z k (r,a),r∈[1,N r ], a∈[1,N a ]}; Where r represents the distance cell number, N r N represents the total number of range-resolved units, 'a' represents the azimuth unit number, and N represents the total number of range-resolved units. a z represents the total number of resolvable elements in the azimuth dimension. k (r,a) represents the amplitude value of the radar measurement data corresponding to the range cell index r and the azimuth cell index a in the k-th frame; Extending the spatial energy diffusion of the target, the radar echo energy is distributed across multiple adjacent resolution cells, thus forming an extended measurement set Ω. k The expression is as follows: in, Let represent the normalized scattering factor of the target amplitude of the measurement unit (r,a) in the k-th frame, and satisfy . when When the time is right, it means there is no target in that unit; The target echo is then modeled as a Rice distribution, and its likelihood function is represented as follows: Where f(·) represents the conditional probability density function, and A represents the amplitude of the extended target centroid. c represents the background noise power. j I0(·) represents the extended target category, and I0(·) represents the 0th order first-class modified Bessel function; Step 3: Model the extended target intra-frame likelihood ratio function and accumulate intra-frame energy based on the target motion orientation; First, a set of multiple velocity assumptions is constructed and the target state space is discretized. Then, the coordinate system of radar measurement and velocity assumption modeling is unified. Finally, target orientation information is introduced to determine the scattering area of ​​the extended target in the radar polar coordinate system and to construct an intra-frame energy accumulation model with orientation adaptation capability. Step 4: Perform multi-frame joint accumulation processing based on motion compensation of the moving platform, output the estimated trajectory and shape parameters of the extended target, and complete the multi-frame pre-detection tracking processing of the time-varying extended target measured by the vehicle-mounted moving platform radar. The problem of multi-frame accumulation under moving platform conditions is solved by modifying the coordinate system used for state-space representation. That is, the corresponding measurement data is searched along the relative motion trajectory of the current state to realize inter-frame state transition and effective energy accumulation, and output extended target estimated trajectory and shape parameters.

2. The method for tracking a vehicle-mounted moving platform radar measurement time-varying extended target before multi-frame detection according to claim 1, characterized in that, Step 3 is described in detail below: Step 31: Construct a set of multiple velocity hypotheses and discretize the target state space; In O A -X A Y A Set up a set of multiple velocity assumptions in the absolute coordinate system; assume that along X... A The range of the shaft's velocity values ​​is discretized as follows There are candidate values, denoted as _____. Along Y A The range of the shaft's velocity values ​​is discretized as follows There are candidate values, denoted as _____. Then define along X A axis and Y A The velocity of the axis is a set of multiple assumptions. Given radar measurement z k (r,a) and a set of velocity assumptions The discretized grid space of the target state Described as an expression: in, Let R represent any discretized state vector of the k-th frame, where R represents the radial distance and A represents the azimuth angle. and They represent along X A The assumed speed value, speed index, and speed resolution of the axis. and They represent along Y A The assumed speed value, speed index, and speed resolution of the axis; T indicates the transpose operation. Step 32: Based on step 31, establish coordinate system one; Radar measurements use the radar as a reference point at O R -X R Y R Represented in relative polar coordinates, while velocity is assumed to be modeled in O. A -X A Y A In an absolute coordinate system, a uniform transformation is therefore required for the coordinates; First, the position coordinates in the radar polar coordinate system Transform to Cartesian coordinates, the corresponding state vector The expression is as follows: in, This represents a coordinate transformation operator used to convert polar coordinates to Cartesian coordinates; Next, the transformed Cartesian coordinates are then used to consider the radar installation angle α and the vehicle's yaw angle β. k The influence of this process is represented as a rotation matrix operation. The rotation matrix R(θ) is defined as follows: Where θ represents the rotation angle independent variable; Then, the motion of the self-driving platform is compensated to obtain... In O A -X A Y A The absolute position in the global coordinate system, and its state vector The expression is as follows: Where h(·) represents the transformation process function, which is the transformation process of mapping the radar polar coordinate state to the global Cartesian coordinate state; Finally, combining the velocity assumption Get O A -X A Y A Final state vector in global coordinate system The expression is as follows: Step 33: Based on step 32, introduce target orientation information to determine O. R -X R Y R The scattering region of the extended target in the radar polar coordinate system is used to construct an intra-frame energy accumulation model with orientation adaptation capability; It is known that the orientation of the extended target is determined by the direction of its center-of-mass velocity, combined with the assumed velocity vector. Target orientation angle d k Represented as After introducing the orientation angle, the target type c is extended. j The positions of the vertices of the rectangle relative to the centroid are corrected using a rotation matrix, as shown in the following expression: in, This represents the set of vertices of the rectangle after correction in the k-th frame. These represent the coordinates of the i-th vertex after rotation correction; Further compensation of the target centroid position yields O A -X A Y A Absolute position of the i-th vertex in the global coordinate system The expression is as follows: in, This indicates that the position coordinates of the target centroid in the global coordinate system are determined by step 32; thus, the set of vertex coordinates of the target rectangle is obtained. Then, through the inverse transformation process function h in step 32 -1 (·), representing the set of vertex coordinates Transform to the radar polar coordinate system to obtain the vertex coordinate set. Among them, h -1 (·) indicates that step 32 starts from O A -X A Y A coordinate system to O R -X R Y R The function for the inverse transformation of the coordinate system; After determining the set of vertex coordinates of the rectangular frame, the extended target scattering region is then derived. The extended target scattering region in the polar coordinate system of the vehicle-mounted radar is then obtained. From the state of the center of mass Car navigation information y k With target category c j The region constraint, which is fully determined and used as the intra-frame likelihood ratio function, is expressed as follows: Where g(·) represents the implicit function generated in the scattering region; Finally, based on the generalized likelihood ratio test and combined with equation (10), the intra-frame likelihood ratio function of the extended target frame after incorporating orientation angle information is derived. The expression is as follows:

3. The method for tracking a vehicle-mounted moving platform radar measurement time-varying extended target before multi-frame detection according to claim 1, characterized in that, Step 4 is described in detail below: Step 41: Initialize the value function and state transition relationship; When k=1, use each discrete state Value function of echo data in the corresponding first frame Relationship with state transition Initialization, i.e. After initialization is complete, proceed to step 42; Step 42: Inter-frame iterative accumulation; When 2≤k≤K, the value function of each discrete state is the current frame state. The intra-frame likelihood ratio function and the set of all possible state transitions accumulated from the previous frame for this state. The sum of the value functions within, i.e.: in, This represents any discretized grid state of the k-th frame. This represents the intra-frame likelihood ratio function derived in step 33. Describes the cumulative value function plane of the k-th frame. Used to store inter-frame state transition relationships. This indicates that the target transitions from frame (k-1) to frame k. The set of all possible state transitions; for To obtain the current discrete state vector, firstly, calculate the current discrete state vector. In O A -X A Y A Position coordinates in global coordinate system Then, based on the target motion model, its possible set of state transitions in the previous frame is inferred, as follows: Where F represents the state transition equation, δ represents the preset boundary threshold vector. x and They represent X respectively A The permissible error range for position coordinates and velocity components along the axial direction, δ y and They represent Y respectively A The permissible error range for position coordinates and velocity components along the axial direction; Step 43: Based on step 42, perform threshold decision on the accumulation value function; When k = K, if the accumulated value function of the Kth frame exceeds the detection threshold η, then the target is determined to exist, and the state of the corresponding maximum value function is extracted. The optimal target state for the last frame is expressed as follows: Step 44: Based on step 43, perform trajectory backtracking on the accumulated value function; When a target is detected, the state transition relationship function is traced back. The short track sequence in radar polar coordinates was recovered. The expression is as follows: Step 45: Based on step 44, perform coordinate transformation and output the estimated trajectory and shape parameters of the extended target; First, the estimated short track sequence Perform the coordinate transformation in step 32, changing it from O R -X R Y R Radar relative polar coordinate system transformation to O A -X A Y A Using a global Cartesian coordinate system, obtain the corresponding target state sequence. To achieve accurate recovery of the time-varying extended target trajectory under vehicle-mounted dynamic platform conditions; Then, from the short track sequence Extract velocity information to determine the target's orientation angle sequence. Combined with the extended target type c j It can obtain the estimated shape parameters of extended targets under time-varying measurement conditions, and realize multi-frame pre-detection tracking processing of extended targets measured by vehicle-mounted moving platform radar; in, This represents the target orientation angle estimation result for the k-th frame.

Images