A millimeter wave radar tracking method in a multi-target complex environment

Abstract

The application belongs to the technical field of radar signal back-end processing, and relates to a millimeter wave radar tracking method in a multi-target complex environment, which comprises the following steps: step one: processing a radio frequency signal received by a radar to generate millimeter wave radar point cloud data; step two: demarcating a radar action range, eliminating point cloud data beyond the range, and then performing point cloud clustering through a DBSCAN clustering method to obtain merge points; and step three: performing track initiation, matching measurement data with a track through data association gate judgment, updating a target state if successful association, and then outputting the track, or otherwise performing track loss or maintenance operation through a scoring board mechanism. The application can effectively improve the accuracy of target tracking, reduce the false tracking rate of multi-target tracking, reduce the calculation complexity of the algorithm, and greatly reduce the calculation time.

Description

Technical Field

[0001] This invention belongs to the field of radar signal back-end processing technology, and relates to a millimeter-wave radar tracking method in complex multi-target environments. Background Technology

[0002] With the increasing informatization of modern society, millimeter-wave radar technology is mainly applied in two areas: military and civilian. In the civilian sector, it is often used for signal processing, such as measuring target distance, speed, and azimuth, and for backend processing like clustering and tracking using the generated radar point cloud data. It is also used for functions such as gesture recognition, fall detection, and heartbeat detection through micro-Doppler features. All of these rely on its core component, the millimeter-wave radar sensor. Therefore, the radar industry has developed rapidly, achieving significant success in military, transportation, automotive, medical, elderly care, and drone fields.

[0003] Multi-target tracking radar is mainly used in scenarios such as transportation and security for the detection and tracking of unknown personnel or vehicles. Due to the strong anti-interference capability of millimeter-wave radar, it can accurately identify moving targets in the dark or in harsh environments.

[0004] Since target tracking is performed in cluttered environments, the most critical issues are the accuracy of single-target tracking and the problem of mistracking in multi-target tracking. When two or more objects are too close, such as moving in parallel or intersecting directions, common data association algorithms, such as associating the measurement data of the current frame with the trajectory of the previous frame, often result in mistracking. Existing technologies often use distance-based data association algorithms, such as NNDA (Nearest Neighbor Association) and GNNDA (Global Nearest Neighbor Association). These methods treat the nearest observation point as the true target point. This can lead to situations where, when two targets are close together, a target point is mistakenly identified as being associated with two different trajectories, resulting in mistracking. Summary of the Invention

[0005] To address the aforementioned technical problems in the existing technology, this invention proposes a millimeter-wave radar tracking method for complex multi-target environments, the specific technical solution of which is as follows:

[0006] A millimeter-wave radar tracking method for multi-target complex environments includes the following steps:

[0007] Step 1: Process the radio frequency signals received by the radar to generate millimeter-wave radar point cloud data;

[0008] Step 2: Define the radar's effective range, remove point cloud data that exceeds the range, and then use the DBSCAN clustering method to cluster the point cloud to obtain merged points;

[0009] Step 3: Start the track. Match the measurement data with the track by using the data association gate. If the association is successful, update the target status and output the track. Otherwise, use the scoreboard mechanism to perform track loss or maintenance operations.

[0010] Furthermore, step one specifically includes the following sub-steps:

[0011] Step 1.1: Separate the data from different channels in the radio frequency signal, and improve the signal-to-noise ratio (SNR) of the signal by removing the DC component, filtering and noise reduction, and matched filtering.

[0012] Step 1.2: Perform a Fast Fourier Transform on the signal in the distance dimension, and process the echo signal in the time domain using pulse compression technology to calculate the target's distance information;

[0013] Step 1.3: Perform Doppler FFT processing on the slow-time data to extract the Doppler frequency of the target, thereby obtaining the radial velocity of the target;

[0014] Step 1.4: Remove static targets using the Moving Target Display (MTI) filtering technique;

[0015] Step 1.5: Perform two-dimensional constant false alarm rate (CFAR) detection to filter out target points and their intensity information;

[0016] Step 1.6: Perform direction of arrival (DOA) estimation to obtain target angle information.

[0017] Furthermore, in step two, the process of defining the radar's effective range and removing point cloud data that exceeds the range involves setting a starting origin, an angle measurement range, and a distance measurement range. Based on the calculated Euclidean distance between the measured point and the starting origin, as well as the angle, it is determined whether the data is within the range, thus obtaining the processed point cloud.

[0018] Furthermore, in step three, the trajectory initiation adopts the m / n logic method, specifically: acquiring all measurements at the initial moment, i.e., the measured values, and establishing a ring gate for the maximum and minimum distances based on the estimated maximum and minimum speeds and the sampling interval;

[0019] The second frame measurement that falls within the initial correlation gate is correlated with the initial measurement to form a candidate track;

[0020] The candidate track formed in the previous moment is extrapolated using the Kalman filter KF algorithm. The extrapolation point is used as the center, and the size of the subsequent gate is determined by the track extrapolation error covariance, thus forming the sampling confirmation area for the next moment. If the measurement of the third frame falls into the relevant gate, it is associated. If multiple measurements fall into the gate, the measurement closest to the extrapolation point is taken.

[0021] Continue extrapolating. If no measurement falls within the gate of the extrapolation point, record the number of failures and use the extrapolation point as the predicted value of the current frame for extrapolation. If the number of failures is greater than nm, it proves to be a failure. Continue extrapolating until a stable track is formed and the track starts successfully.

[0022] Furthermore, the Kalman filter (KF) algorithm specifically includes:

[0023] The target state vector of the entire multi-target tracking process is represented as:

[0024] X k =[x,y,x',y']

[0025] Where x, y represent the target's position information, and x', y' represent the target's velocity information;

[0026] Prediction step: Predict the state at the next moment based on the final predicted value at the previous moment;

[0027] State transition equation: X k|k-1 =FX k-1|k-1 +w k

[0028] Predicted covariance matrix: P k|k-1 =FP k-1|k-1 F T +Q

[0029] In the formula, X k-1|k-1 w represents the final predicted value at the previous moment. k Let Q be the process noise, Q be the process noise covariance matrix, and F be the state transition matrix;

[0030] Update step: Adjust the predicted value of the prediction step based on the measured value to obtain the final state prediction value;

[0031] Measurement equation: Z k =HX k +u k

[0032] Measurement covariance matrix: S k =HP k|k-1 H T +R

[0033] Kalman gain:

[0034] Status Update: X k|k =X k|k-1 +K k (Z k -HX k|k-1 )

[0035] Update covariance matrix: Pk|k =(IK k H)P k|k-1

[0036] In the formula, R is the measurement noise covariance matrix, and Z k -HX k|k-1 For the information at this moment, the measurement matrix H is:

[0037]

[0038] X the current updated state k|k As input for the next moment, the prediction and update process is repeated to achieve radar target tracking.

[0039] Furthermore, in step three, the matching of measurement data with the track through data association gate determination is implemented using the GJPDA algorithm, which specifically includes the following steps:

[0040] Step 1: Using Mahalanobis distance as the criterion, make a prediction based on the estimated value from the previous time step, and establish a gate with the center of the predicted value to obtain the effective measurement falling within the gate.

[0041] Step 2: Obtain the confirmation matrix Ω from the relationship between the effective measurement and the target confirmation gate;

[0042] Step 3: Calculate the interconnection probability β between the target and the measurement based on the likelihood function. jt ;

[0043] Step 4: Calculate the final combination information v k The target state estimate and covariance matrix at the current time can be obtained based on the Kalman filter update step, and are ready for use in the next iteration.

[0044] Furthermore, in step 2, the process of establishing the confirmation matrix Ω is as follows:

[0045]

[0046] Where w jt Represents the measurement j (j=1,2,...,m) k A flag indicating whether w is associated with the target t (t = 0, 1, 2, ..., T), with a value of 0 or 1; when w jt When w = 0, it means that the j-th measurement is not related to the t-th target; when w jt When m = 1, it indicates that the j-th measurement is associated with the t-th target; if the first column of the confirmation matrix is ​​all 1s, it indicates that at any given time no target is associated with the measurement, and all measurements are spurious; kIt is the number of measurements that fall into the tracking gate at time k, and T is the number of targets.

[0047] Furthermore, in step 3, the likelihood function is used instead of the method of establishing multiple interconnection matrices and calculating the conditional probability of each interconnection matrix to obtain the interconnection probability of measurement j and target t. In this algorithm, the final formula for calculating the association probability is as follows;

[0048]

[0049] In the formula G jt It can be viewed as the probability that the j-th measurement is associated with the t-th target;

[0050]

[0051] In the formula The information function is used to calculate the information between the j-th measurement and the t-th target; S jt (k) represents the information covariance corresponding to the information function; in actual tracking situations, the value of B is related to the clutter density of the environment.

[0052] Furthermore, in step 4, the combined information is calculated as follows:

[0053]

[0054] Based on Kalman filtering, the final state estimate of target t at time k is calculated, and the posterior covariance is updated.

[0055] Z k|k =X k|k-1 +K k V t

[0056] P k∣k =P k∣k-1 -(1-β 0t )K k S k K k T +K k (Z k|k -V t V t T )K k T

[0057] If step 1 results in a failed trajectory, then X will be... k|k-1 The final state estimate is input into the next iteration, using the covariance matrix from the previous time step as the covariance matrix P for this time step. k∣k .

[0058] Furthermore, the scoring board mechanism specifically involves: recording the number of times the target is lost and deciding whether to continue maintaining the track; if the number of times the target is lost exceeds a set threshold n, the target is considered lost, a track loss operation is performed, and the tracking of the track is terminated; otherwise, the track maintenance phase is entered, and subsequent measurement data is waited for to determine whether the target still exists.

[0059] Beneficial effects:

[0060] The edge track initiation method in this invention can effectively solve the problem of generating repetitive tracks that are similar to the real target tracks due to excessive clutter density or excessively small track gate. By utilizing the common knowledge that new targets only appear at the edge of the radar's effective range, the computational complexity of the algorithm can be effectively reduced.

[0061] The GJPDA proposed in this invention optimizes the interconnect matrix construction step based on the commonly used JPDA algorithm for multi-target tracking. It replaces the conditional probability calculation of the interconnect matrix with a likelihood function, effectively solving the problem of exponential growth in the interconnect matrix due to increased clutter density leading to an increase in observations. This significantly reduces computation time while retaining the confirmation matrix step, thus preserving the multi-target tracking characteristics of JPDA. When deployed on embedded devices, it can also address real-time performance issues to some extent.

[0062] This invention effectively matches new measurement data with the correct target track by using gate judgment and m / n association rules, reducing the possibility of erroneous association. A scoreboard mechanism is employed to evaluate track stability, preventing premature termination of target tracking due to short-term measurement loss. Attached Figure Description

[0063] Figure 1 This is a flowchart of radar multi-target tracking according to an embodiment of the present invention;

[0064] Figure 2 This is a flowchart of radar signal processing according to an embodiment of the present invention;

[0065] Figure 3 This is a schematic diagram of the DBSCAN algorithm according to an embodiment of the present invention;

[0066] Figure 4 This is a flowchart of the trajectory management process according to an embodiment of the present invention;

[0067] Figure 5 This is a schematic diagram of the start of the m / n logical track according to an embodiment of the present invention;

[0068] Figure 6 This is a schematic diagram illustrating the process of establishing the confirmation matrix using the GJPDA algorithm according to an embodiment of the present invention;

[0069] Figure 7 This is a dataset collection image of pedestrian cross-movement according to an embodiment of the present invention;

[0070] Figure 8 This is a display of pedestrian cross-movement radar point cloud data according to an embodiment of the present invention;

[0071] Figure 9 This is a diagram showing the results of multi-target tracking of pedestrian cross-movement according to an embodiment of the present invention;

[0072] Figure 10 This is a dataset collection diagram of a single vehicle's close-range parallel motion according to an embodiment of the present invention;

[0073] Figure 11 This is a display of radar point cloud data for a single vehicle in close-range parallel motion according to an embodiment of the present invention;

[0074] Figure 12 This is a diagram showing the results of multi-target tracking of a single vehicle moving in close parallel motion according to an embodiment of the present invention. Detailed Implementation

[0075] To make the objectives, technical solutions, and technical effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0076] like Figure 1 As shown in the figure, this embodiment discloses a millimeter-wave radar tracking method in a complex multi-target environment, including the following steps:

[0077] Step 1: Process the radio frequency data received by the radar to generate millimeter-wave radar point cloud data.

[0078] like Figure 2 As shown, firstly, in the data preprocessing stage, the MIMO radar, through the configuration of N transmit antennas and M receive antennas, can approximately simulate N×M virtual channels. The formation of these virtual channels significantly improves the radar's spatial resolution and target detection capability. In this stage, the main task of data preprocessing is to separate the data from different channels in the radio frequency signal, improving the signal-to-noise ratio (SNR) through methods such as removing DC components, filtering and noise reduction, and matched filtering, thus providing a better data foundation for subsequent processing.

[0079] After data preprocessing, the signal enters the range-Doppler processing stage for incoherent accumulation. In this stage, a Fast Fourier Transform (FFT) is first performed on the signal in the range dimension. Pulse compression technology is then used to process the echo signal in the time domain to accurately calculate the target's range information. Next, Doppler FFT processing is performed on the slow-time data to extract the target's Doppler frequency, thereby obtaining the target's radial velocity. Simultaneously, to eliminate interference signals in static scenes, such as echoes from the ground or fixed objects, the radar system uses Moving Target Indication (MTI) filtering technology to remove static targets, making the signals of moving targets more prominent. This dynamic filtering process not only improves the system's sensitivity to moving targets but also significantly reduces the impact of background noise.

[0080] After MTI filtering, static clutter in the signal has been effectively removed, and the signal then enters the two-dimensional constant false alarm rate (CFAR) detection stage. The core idea of ​​the CFAR detection algorithm is to maintain a constant false alarm probability during target detection by dynamically adjusting the detection threshold to ensure system stability under different environments. Commonly used two-dimensional CFAR algorithms include cell-averaged CFAR (CA-CFAR) and maximum value selection CFAR (OS-CFAR), which can adaptively identify the target point's location in range-Doppler two-dimensional data. During detection, the CFAR algorithm constructs a sliding window, performs statistical analysis of the noise environment around the target, calculates the detection threshold based on the set false alarm probability, and ultimately filters out possible target points and their intensity information.

[0081] After target detection, to further refine the target's spatial location information, the radar system enters the Direction of Arrival (DOA) estimation stage. In this stage, the commonly used spatial spectrum estimation algorithm, MUSIC (Multiple Signal Classification), is based on the separation of the signal and noise subspaces. It constructs a spatial spectrum function and scans the spatial angles to deduce the target's angular information using a maximization method. The MUSIC algorithm can provide extremely high angular resolution under high signal-to-noise ratio conditions, making it particularly suitable for MIMO radar systems with numerous receiving channels.

[0082] The above steps can yield relatively accurate millimeter-wave radar point cloud information.

[0083] Step 2: Define the radar's effective range, remove observations outside the range, and then use the DBSCAN clustering method to cluster the point cloud and obtain the ideal merge points.

[0084] Specifically, a preprocessing module is set up to read the point cloud information obtained from radar signal processing, cluster the point cloud, and use the average value as the ideal merge point. The clustering method adopts the DBSCAN clustering method based on density clustering, and uses the K nearest neighbor distance as the radius in DBSCAN to achieve adaptive clustering. At the same time, the radar range is automatically defined, and when a new measurement value is input, the observation value that exceeds the radar range is directly discarded. In addition, since in reality, after the track start, if a new target or track appears, it will generally enter from the edge of the radar detection range, an edge track start mechanism is proposed. The data preprocessing module is reused, and the function that identifies the track that exceeds the radar measurement range is overloaded.

[0085] For example, setting the starting origin {0, 0}, the angle measurement range {10, 150}, and the distance measurement range {0, 50}, the point cloud is obtained by calculating the Euclidean distance between the measured point and the starting origin, and determining whether the angle is within the range. The Euclidean distance formula is as follows:

[0086]

[0087] Next, clustering is performed using the DBSCAN algorithm, which is determined by the following two parameters:

[0088] ε(eps): Defines the size range of the neighborhood;

[0089] MinPts: The minimum number of neighborhood points that a point is considered to be a core point.

[0090] The DBSCAN algorithm is a density-based clustering algorithm. If p is a core point, then there are at least MinPts points within the range ε. If q is also a core point and q is within the range ε ​​of p, then q is said to be density-accessible from p. This process continues to form a cluster consisting of multiple core points and their neighboring points. Figure 3 As shown, unlike ordinary Euclidean clustering which forms regular circles, this method has better recognition of objects that do not belong to regular shapes, and it does not require prior knowledge of the number of targets, automatically identifying them into different numbers of clusters.

[0091] Next, the ideal merge point is obtained by averaging all x and y values ​​of these clusters.

[0092] Step 3: As Figure 4As shown, based on existing track information and measurement input (i.e., new measurement data provided by sensors), it is determined whether the current measurement value is within the track gate. The track gate is a threshold region generated based on the predicted target state; only measurements falling within this region can be associated with the track. If the measurement value is within the track gate, a filtering prediction step is executed, using methods such as Kalman filtering to update the target state and improve the accuracy of subsequent data association. Otherwise, it further checks whether the current measurement value meets the m / n logical association condition. If the condition is met, a new track initiation is executed, i.e., a new target track is created.

[0093] The next step involves data association gate judgment, which checks whether the measured value falls within a more stringent data association gate range to determine if it can be matched with an existing track. If the measured value is within the data association gate, data association is performed, matching the measured data with an existing track to update the target state. After successful association, a filtering update step is executed to correct the target state, improve the accuracy of track estimation, and finally output the track, providing stable target trajectory information.

[0094] If the measured values ​​cannot be correlated with any existing track, a scoreboard mechanism is used to manage and continuously evaluate the stability and reliability of the track. This mechanism records the number of target losses and determines whether to continue maintaining the track. If the number of target losses exceeds a set threshold n, the target is considered lost, a track loss operation is performed, and tracking of the track is terminated; otherwise, the track maintenance phase begins, waiting for subsequent measurement data to determine if the target still exists. The entire process is executed cyclically, enabling stable target tracking in noisy environments and is suitable for various applications such as radar and visual target tracking.

[0095] Based on the above method, this embodiment designs a track management module, as well as a tracking filtering module and a data association module that run in parallel therein.

[0096] like Figure 5 As shown, the process of starting the track is the process of generating the track header. In this embodiment, an m / n logical track start is used, which means that if the same target is observed at least m times within n frames, the target is considered to exist. An initial track is generated in the order of m frames, and the covariance matrix of the last frame is obtained by Kalman filtering.

[0097] The specific process for initiating the m / n logical track is as follows:

[0098] Acquire all measurements, i.e., the measured values, at the initial moment, and based on the estimated maximum velocity V max and minimum speed V min And the sampling interval T establishes the maximum distance V max *T and minimum distance V min*T's ring gate:

[0099] d ij (t) = max[0, z j (k+1)-z i (k)-v max t]+max[0,-z j (k+1)+z i (k)+v min t]

[0100]

[0101] The second frame measurement that falls within the gate is correlated with the initial measurement to form a candidate track.

[0102] The candidate track formed in the previous moment is extrapolated using the Kalman filter (KF) algorithm. The extrapolation point is used as the center, and the size of the subsequent gate is determined by the track extrapolation error covariance, thus forming the sampling confirmation area for the next moment. If the measurement in the third frame falls into the relevant gate, it is correlated. If multiple measurements fall into the gate, the measurement closest to the extrapolation point is taken.

[0103] Continue extrapolating. If no measurement falls within the gate of the extrapolation point, record the number of failures and use the extrapolation point as the predicted value of the current frame for extrapolation. If the number of failures is greater than nm, it proves to be a failure. Continue extrapolating until a stable track is formed and the track starts successfully.

[0104] In this embodiment, the filtering prediction step executed by the tracking filtering module predicts the target's current position based on the target's position and covariance matrix from the previous moment. Kalman filtering (KF) is used as the main algorithm for target tracking. First, a state model is established, including state variables such as position and velocity, and a state transition matrix is ​​defined to predict the target's position at the next moment. Then, the prediction uncertainty P is calculated using the process noise covariance matrix. After the sensor acquires the measurement value, the measurement residual is calculated using the measurement matrix H. Next, the Kalman gain is calculated to weight and fuse the predicted and measured values, updating the state estimate and adjusting the uncertainty. This process is continuously iterated to make the target trajectory estimation more accurate. The KF algorithm implementation process specifically includes:

[0105] The target state vector of the entire multi-target tracking process is represented as:

[0106] X k =[x,y,x',y']

[0107] Where x, y represent the target's position information, and x', y' represent the target's velocity information.

[0108] Prediction step: Predict the state at the next moment based on the final predicted value at the previous moment.

[0109] State transition equation: X k|k-1 =FX k-1|k-1 +w k

[0110] Predicted covariance matrix: P k|k-1 =FP k-1|k-1 F T +Q

[0111] In the formula, X k-1|k-1 w represents the final predicted value at the previous moment. k Let Q be the process noise, Q be the process noise covariance matrix, and F be the state transition matrix. Based on the target state vector, we can obtain:

[0112]

[0113] Where T is the radar sampling interval.

[0114] Update step: Adjust the predicted value of the prediction step based on the measured value to obtain the final state prediction value.

[0115] Measurement equation: Z k =HX k +u k

[0116] Measurement covariance matrix: S k =HP k|k-1 H T +R

[0117] Kalman gain:

[0118] Status Update: X k|k =X k|k-1 +K k (Z k -HX k|k-1 )

[0119] Update covariance matrix: P k|k =(IK k H)P k|k-1

[0120] In the formula, R is the measurement noise covariance matrix, and Z k -HX k|k-1 For the information at this moment, the measurement matrix H is:

[0121]

[0122] X the current updated state k|k As input for the next moment, the prediction and update process is repeated to achieve radar target tracking.

[0123] In the data association module, the measured value at the current moment is associated with the predicted value, and multiple observations are accurately matched with multiple target predicted values ​​in a multi-target environment. For multiple objects running in parallel or intersecting, distance alone is often insufficient for accurate determination. Therefore, a data association algorithm specifically for multi-target tracking, JPDA (Joint Probability Density Data Association), is selected. This algorithm is an advanced version of PDA (Probability Density Data Association). PDA is only suitable for single-target cases. Based on this, an m-row n-column confirmation matrix is ​​generated. With m targets and n measurements, multiple interconnection matrices can be generated according to certain rules. By calculating the probability of each interconnection matrix, it is finally determined which measurement ultimately belongs to which target. However, the drawback of this algorithm is that as the number of targets and measurements increases, the number of interconnection matrices increases exponentially. If deployed on hardware devices, it cannot meet the real-time requirements. Therefore, this embodiment proposes a method to approximate the interconnection probability using a likelihood function to reduce the number of association cases to consider, thereby improving the efficiency of the algorithm. This algorithm is called GJPDA. The GJPDA algorithm, when calculating correlation probabilities, exhibits characteristics of the JPDA algorithm. It assigns higher weights to measurements appearing only within a single target tracking gate, while reducing the weights of measurements within overlapping regions of multiple target tracking gates. This weight allocation strategy effectively handles overlapping target-related gates, improving the algorithm's robustness. The implementation process of the GJPDA algorithm includes the following steps:

[0124] Step 1: Make a prediction based on the estimated value from the previous moment, and establish a gate with the center of the predicted value to obtain the effective measurement falling into the gate.

[0125] Step 2, as follows Figure 6 As shown, the confirmation matrix Ω is derived from the relationship between the effective measurement and the target confirmation gate.

[0126] Step 3: Calculate the interconnection probability β between the target and the measurement based on the likelihood function. jt .

[0127] Step 4: Calculate the final combination information v k The target state estimate and covariance matrix at the current time can be obtained based on the Kalman filter update step, and are ready for use in the next iteration.

[0128] The formula for establishing the gate based on the center of the predicted value in step 1 is shown below. Here, Mahalanobis distance is used as the criterion, the predicted value at the previous moment is used as the center of the gate, and γ is used as the radius.

[0129]

[0130] In the formula, S-1 (k) is the covariance matrix of the target prediction at the previous time step. The predicted value for this moment is obtained based on the target value from the previous moment. i (k) represents the observed value at this moment.

[0131] In step 2, as Figure 6 As shown, the process of establishing the confirmation matrix Ω is as follows:

[0132]

[0133] Where w jt Represents the measurement j (j=1,2,...,m) k This flag indicates whether w is associated with the target t (t = 0, 1, 2, ..., T), and its value is either 0 or 1. Specifically, when w jt When w = 0, it means that the j-th measurement is not related to the t-th target; when w jt When m = 1, it indicates that the j-th measurement is associated with the t-th target. If the first column of the confirmation matrix is ​​all 1s, it means that at any given time, no target is associated with any measurement, and all measurements are spurious. k It is the number of measurements that fall into the tracking gate at time k, and T is the number of targets.

[0134] In step 3, the likelihood function is used instead of the method of establishing multiple interconnection matrices and calculating the conditional probability of each interconnection matrix to obtain the interconnection probability of measurement j and target t. The final formula for calculating the association probability in this algorithm is as follows;

[0135]

[0136] In the formula G jt It can be viewed as the probability that the j-th measurement is associated with the t-th target;

[0137]

[0138] In the formula This represents the information function used to calculate the information between the j-th measurement and the t-th target. jt (k) represents the information covariance corresponding to the information function. In actual tracking situations, the value of B is related to the clutter density of the environment. If the clutter density is not very high, the constant B can usually be set to 0.

[0139] In step 4, the combined information is calculated as follows:

[0140]

[0141] Based on Kalman filtering, the final state estimate of target t at time k is calculated, and the posterior covariance is updated.

[0142] Z k|k =X k|k-1 +K k V t

[0143] P k∣k =P k∣k-1 -(1-β 0t )K k S k K k T +K k (Z k|k -V t V t T )K k T

[0144] If step 1 results in a failed trajectory, then X will be... k|k-1 The final state estimate is input into the next iteration, using the covariance matrix from the previous time step as the covariance matrix P for this time step. k∣k .

[0145] Test data and algorithm verification:

[0146] This invention uses a 77G millimeter-wave radar, some of whose parameters are as follows:

[0147]

[0148] Depend on Figure 7 As shown, the scene is a road scene, and two pedestrians are moving in a crisscrossing pattern. Figure 8 Given 2D point cloud data after radar signal processing, a radar multi-target tracking algorithm can be used to obtain... Figure 9 The results show that the algorithm did not mistrack the target before and after the two pedestrians crossed, and the trajectory did not start repeatedly. The overall performance was good, and the simplification of the JPDA calculation process did not result in any loss of accuracy. Furthermore, the algorithm did not lose track of either pedestrian throughout the entire process, confirming the effectiveness of the tracking method. Figure 10 As shown, the scene is a road scene with two bicycle targets and one pedestrian target. The two bicycles are moving in parallel, and the closest distance between them is less than 0.3 meters. According to... Figure 12 The results show that the clustering method in the preprocessing stage can effectively distinguish closely clustered objects, and the subsequent data association will not result in trajectory overlap.

[0149] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the implementation process of the present invention has been described in detail above, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A millimeter-wave radar tracking method for multi-target complex environments, characterized in that, Includes the following steps: Step 1: Process the radio frequency signals received by the radar to generate millimeter-wave radar point cloud data; Step 2: Define the radar's effective range, remove point cloud data that exceeds the range, and then use the DBSCAN clustering method to cluster the point cloud to obtain merged points; Step 3: Start the track. Match the measurement data with the track by using the data association gate. If the association is successful, update the target status and output the track. Otherwise, use the scoreboard mechanism to perform track loss or maintenance operations. In step three, the trajectory is initiated using the m / n logic method, specifically: all measurements at the initial moment are obtained, and a ring gate for the maximum and minimum distances is established based on the estimated maximum and minimum speeds and the sampling interval; The second frame measurement that falls within the initial correlation gate is correlated with the initial measurement to form a candidate track; The candidate track formed in the previous moment is extrapolated using the Kalman filter KF algorithm. The extrapolation point is used as the center, and the size of the subsequent gate is determined by the track extrapolation error covariance, thus forming the sampling confirmation area for the next moment. If the measurement of the third frame falls into the relevant gate, it is associated. If multiple measurements fall into the gate, the measurement closest to the extrapolation point is taken. Continue extrapolating. If no measurement falls within the gate of the extrapolation point, record the number of failures and use the extrapolation point as the predicted value of the current frame for extrapolation. If the number of failures is greater than nm, it proves to be a failure. Continue extrapolating until a stable track is formed and the track starts successfully. In step three, the matching of measurement data with the track through data association gate determination is implemented using the GJPDA algorithm, which specifically includes the following steps: Step 1: Using Mahalanobis distance as the criterion, make a prediction based on the estimated value from the previous moment, and establish a gate with the center of the predicted value to obtain the effective measurement falling into the gate. ; Step 2: Derive the confirmation matrix from the relationship between the effective measurement and the target confirmation gate. ; Step 3: Calculate the interconnection probability between the target and the measurement based on the likelihood function. ; Step 4: Calculate the final combination information The target state estimate and covariance matrix at the current time can be obtained based on the Kalman filter update step, and are ready for use in the next iteration. In step 2, the confirmation matrix is... The establishment process is as follows: in Indicates measurement Is it in line with the target? The associated flag has a value of 0 or 1; when When, it means the first The first measurement and the first The goals are unrelated; when When, it means the first The first measurement and the first There is a correlation between the targets; if the first column of the confirmation matrix is ​​all 1, it means that at any given time no target is associated with the measurement, and all measurements are spurious measurements. yes The number of measurements falling within the tracking gate at any given time. It refers to the number of targets; In step 3, the likelihood function is used instead of the method of establishing multiple interconnection matrices and calculating the conditional probability of each interconnection matrix to obtain the measurement. and target The interconnection probability is calculated using the following formula in this algorithm: In the formula Consider it as the probability that the j-th measurement is associated with the t-th target; In the formula Represents the information function, used to calculate the first... The first measurement and the first Information about the target; This represents the information covariance corresponding to the information function; in actual tracking situations, The value of is related to the clutter density of the environment; In step 4, the combined information is calculated as follows: Based on Kalman filtering, calculate Momentary Goal The final state estimate is obtained, and the posterior covariance is updated: If step 1 results in a failed trajectory, then... The final state estimate is input into the next iteration, using the covariance matrix from the previous time step as the covariance matrix for this time step. .

2. The tracking method as described in claim 1, characterized in that, Step one specifically includes the following sub-steps: Step 1.1: Separate the data from different channels in the radio frequency signal, and improve the signal-to-noise ratio (SNR) of the signal by removing the DC component, filtering and noise reduction, and matched filtering. Step 1.2: Perform a Fast Fourier Transform on the signal in the distance dimension, and process the echo signal in the time domain using pulse compression technology to calculate the target's distance information; Step 1.3: Perform Doppler FFT processing on the slow-time data to extract the Doppler frequency of the target, thereby obtaining the radial velocity of the target; Step 1.4: Remove static targets using the Moving Target Display (MTI) filtering technique; Step 1.5: Perform two-dimensional constant false alarm rate (CFAR) detection to filter out target points and their intensity information; Step 1.6: Perform direction of arrival (DOA) estimation to obtain target angle information.

3. The tracking method as described in claim 1, characterized in that, In step two, defining the radar's effective range and removing point cloud data that exceeds the range involves setting a starting origin, an angle measurement range, and a distance measurement range. Based on the calculated Euclidean distance between the measurement point and the starting origin, as well as the angle, it is determined whether the data is within the range, thus obtaining the processed point cloud.

4. The tracking method as described in claim 1, characterized in that, The Kalman filter (KF) algorithm specifically includes: The target state vector of the entire multi-target tracking process is represented as: in , Indicates the target's location information. , Indicates the target's speed information; Prediction step: Predict the state at the next moment based on the final predicted value at the previous moment; State transition equation: Predicting the covariance matrix: In the formula, This represents the final predicted value at the previous moment. For process noise, It is the process noise covariance matrix. This is the state transition matrix; Update step: Adjust the predicted value of the prediction step based on the measured value to obtain the final state prediction value; Measurement equation: Measurement covariance matrix: Kalman gain: Status Update: Update the covariance matrix: In the formula, To measure the noise covariance matrix, For information at this moment, the measurement matrix for: The current updated state As input for the next moment, the prediction and update process is repeated to achieve radar target tracking.

5. The tracking method as described in claim 1, characterized in that, The scoring board mechanism specifically records the number of times the target is lost and decides whether to continue maintaining the track. If the number of times the target is lost exceeds a set threshold n, the target is considered lost, a track loss operation is performed, and the tracking of the track is terminated. Otherwise, the track maintenance phase is entered, and subsequent measurement data is waited for to determine whether the target still exists.

Images