Track management method based on multi-frame energy accumulation

Through multi-frame energy accumulation and prediction wavegate technology, the problem of high false track maintenance rate under low thresholds in radar technology is solved, and the detection performance and track maintenance probability of long-distance low signal-to-noise ratio targets are improved, achieving more stable Target tracking.

CN119044971AActive Publication Date: 2024-11-29XIDIAN UNIV

Patent Information

Application Number
CN202411186882.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-11-29
Estimated Expiration
2044-08-28

AI Technical Summary

Technical Problem

It is difficult for existing radar technology to effectively distinguish long-distance low signal-to-noise ratio targets and false tracks under low threshold detection, resulting in a high probability of false track maintenance and affecting target state analysis.

Method used

Using a track management method based on multi-frame energy accumulation, a prediction wave gate is established to perform low threshold detection by predicting the target state, and a false track maintenance probability threshold is set, and the track segments below the judgment threshold are identified as false tracks. Undo it to increase the probability of long-distance track maintenance.

Benefits of technology

On the premise of ensuring the maintenance probability of low false tracks, the detection performance and track maintenance probability of long-distance low signal-to-noise ratio targets are improved, the occurrence of false tracks is reduced, and the stability of target tracking is enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119044971A_ABST
    Figure CN119044971A_ABST
Patent Text Reader

Abstract

The invention discloses a track management method based on multi-frame energy accumulation. The method mainly solves the problem that a long-distance low-signal-to-noise-ratio target track is difficult to maintain under the condition of low false track maintenance probability in the prior art, and comprises the following steps: predicting a target state by using tracking information to establish a prediction gate; carrying out low-threshold detection in the prediction gate; performing data association on a detection result and a historical track; continuing to track or terminate the track according to different conditions of the associated track: if the associated track meets the n / m logic method criterion, continuing to track, otherwise terminating the track; for the continuous tracking track, judging whether the current association result is empty or not: if so, processing the next frame of data, otherwise, performing tracking filtering on the association result; and judging whether the filtered flight path is a false flight path or not: if the filtered flight path is the false flight path, cancelling the flight path, otherwise, continuing to track. According to the method, the target track maintenance probability under the lower false track maintenance probability can be improved, and the method can be used for detecting and tracking a low-signal-to-noise-ratio target.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of radar technology, in particular to a track management method which can be used for radar target detection and tracking. Technical Background

[0002] Traditional radar detection and tracking are one-way. In the detection step, there is usually no prior information about the target in its surveillance area. Generally, a higher detection threshold is required to reduce the number of false alarms, but a higher detection threshold makes it difficult to detect a low signal-to-noise ratio target at a long distance. After the track starts, the system already has sufficient prior information about the target. Based on the prior information of the target, more stable target tracking can be achieved. That is, the tracking information can be used to predict the target state at the next moment, establish a prediction gate, and perform low threshold detection within the gate. This method can improve the detection performance of targets with low signal-to-noise ratio at a long distance while controlling the number of false alarms, thereby increasing the probability of track maintenance. However, low threshold detection may cause the track to be associated with the false alarm point track, generate some false tracks, and affect the analysis of the target state. Therefore, a track management method is needed to maintain the target track with a low signal-to-noise ratio while canceling the false track.

[0003] The patent document with application number 202110224659.5 discloses a multi-target tracking algorithm based on a track management method, which measures the track life value and track score by the length of time that the track has not been updated. If it is updated for a short time, the track life value and track score will be increased, otherwise the track life value and track score will be reduced. When the track life value is 0 or the track score is negative, the track is terminated. If the number of historical associations with the terminated track is small, the track is determined to be a false track, otherwise it is determined to be a target track and output. By introducing the length of time that the track has not been updated, track interruption and loss are prevented; false tracks are suppressed by the information of the number of historical associations of the track. However, in the case of low threshold detection, the algorithm has an increase in the number of false point tracks in the wave gate, and the false point tracks are easily associated with the track, which leads to a shorter update time and makes it difficult to terminate the false track. At the same time, since the number of historical associations of the false track also increases, it is impossible to effectively revoke the false track by limiting the number of historical associations.

[0004] The patent document with application number 201811134571.9 discloses a heterogeneous multi-sensor multi-target tracking method, which adopts a variable evidence set method to define the track position state, carrier frequency, repetition rate, and pulse width as a valid evidence set. According to the type of candidate measurement, it is determined whether the candidate measurement has position, signal characteristics and other information, and the associated evidence set is automatically deleted according to the measurement type to determine the measurement associated evidence set. The position correlation function and the signal feature correlation function are calculated respectively, and the measurement comprehensive correlation function is obtained by combining evidence. The measurement comprehensive correlation function effectively utilizes the multi-dimensional measurement information of multiple sensors to calculate the track comprehensive score, so that the data association result is more accurate and the target tracking track continuity is better. However, as the detection threshold is reduced, the number of false point tracks detected by a single sensor increases. It is difficult to distinguish the target from the false point tracks through the position, signal characteristics and other information of the candidate measurement, which greatly increases the probability of the association between the false point tracks and the track, and the probability of the occurrence of false tracks will also increase accordingly, and the false track cannot be effectively cancelled under the condition of low threshold detection. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and propose a track management method based on multi-frame energy accumulation, so as to increase the detection performance of long-distance low signal-to-noise ratio targets and improve the track maintenance probability at long distances while ensuring a low false track maintenance probability.

[0006] The technical idea for achieving the purpose of the present invention is: through the tracking information of the track, the target state at the next moment is predicted, and a prediction gate is established, and a low threshold detection is performed within the gate to improve the detection performance of low signal-to-noise ratio targets; by accumulating multiple frames of energy of the track, a threshold of low false track maintenance probability is set, and the track below the judgment threshold is identified as a false track and revoked, so as to improve the target track maintenance probability at a long distance while ensuring the low false track maintenance probability.

[0007] According to the above ideas, the track management method based on multi-frame energy accumulation of the present invention includes the following steps:

[0008] (1) Estimated target state based on the last frame after the start of the track With its state covariance matrix D k|k , predict the target state of the next frame With its covariance matrix D l|k ;

[0009] (2) Based on forecast information and D l|k Establish a prediction gate and set the false alarm probability to P fa , perform low threshold detection on the data that has been processed and falls into the prediction wave gate, and obtain the detection result Z at the current time ll ;

[0010] (3) The detection result Z at the current time l l Perform data association with historical tracks to obtain the association result z l , and determine whether the associated track meets the n / m logic method criteria:

[0011] If not satisfied, the track is determined to be terminated;

[0012] If satisfied, proceed to step (4);

[0013] (4) Determine the association result z l Is it empty?

[0014] If the association result z l If it is empty, set time l = l + 1 and return to step (1);

[0015] If the association result z l If it is not empty, continue tracking and execute step (5);

[0016] (5) Track and filter the associated points to calculate the estimated target state at the current time l With the covariance matrix D l|l , and update the track;

[0017] (6) Assume that the accumulated energy window length is N, and perform multi-frame energy accumulation on the track segment of the updated track within the accumulated energy window to obtain the energy accumulation value Y l ;

[0018] (7) Set the false track maintenance probability as P t|fa , calculate the corresponding track decision threshold as v t_track ;

[0019] (8) Determine the energy accumulation value Y l Is it below the track decision threshold v? t_track

[0020] If it is lower than the threshold, the track segment in the window is considered to be a false track and will be cancelled;

[0021] Otherwise, continue tracking, set time l=l+1, and return to step (1).

[0022] Compared with the prior art, the present invention has the following advantages

[0023] Firstly, the present invention accumulates the multi-frame energy of the track and cancels the track below the false track judgment threshold, thereby ensuring a low probability of maintaining a false track;

[0024] Secondly, the present invention uses tracking information to assist in establishing a prediction gate for low threshold detection, which can ensure the detection performance of low signal-to-noise ratio targets while controlling a low number of false alarms, thereby improving the probability of track maintenance at long distances. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Flow chart for realizing the present invention

[0026] Figure 2 This is a comparison chart of the track maintenance probability at different distances between the present invention and the traditional detection and tracking algorithm. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only embodiments of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without creative work should all fall within the protection scope of the present invention.

[0028] It should be noted that the step numbers in the specification and claims of the present invention are only for a clear description of the implementation scheme of the present invention to facilitate understanding, and the order of the step numbers is not limited.

[0029] Reference Figure 1 , the implementation steps of this example are as follows:

[0030] Step 1: Estimate the state of the target in the last frame after the track starts With its state covariance matrix D k|k , predict the target state of the next frame With its covariance matrix D l|k .

[0031] The prediction refers to obtaining the possible state of the target at time l according to the motion state of the target at time k and its motion trend, thereby obtaining the prediction information of the target at time l.

[0032] The specific implementation of this step is as follows:

[0033] (1.1) Estimated state based on the target at time k Predict the target state at time l

[0034]

[0035] Among them, F l|k is the target state transfer matrix from time k to time l;

[0036] (1.2) According to the target state covariance matrix D at time k k|k , calculate the target prediction state covariance matrix D at time l l|k :

[0037]

[0038] Among them, Q k|k is the process noise covariance matrix from time k to time l.

[0039] Step 2: Based on the forecast information and D l|k Establish a prediction wave gate, perform low threshold detection on the data that has been processed and falls into the prediction wave gate, and obtain the detection result Z at the current time l l .

[0040] The prediction gate refers to the target prediction information at time l. and D l|k , estimate the area where the target may appear. Since the prediction gate is only a part of the input observation space, although low threshold detection will lead to a higher false alarm probability, the number of false alarms is limited, while improving the detection performance of low signal-to-noise ratio targets.

[0041] The specific implementation of this step is as follows:

[0042] (2.1) The target prediction state Convert to the measurement space and record it as

[0043]

[0044] Among them, h l is the target measurement function at time l;

[0045] (2.2) According to the target prediction state covariance matrix D at time l l|k , calculate the target measurement state The prediction covariance matrix B l|k :

[0046]

[0047] Among them, H l Expressed as the target measurement function h at time l l The Jacobian matrix of Indicates H l The transpose of

[0048] (2.3) Measure the state according to the predicted target and the predicted covariance matrix B l|k , establish the following prediction gate:

[0049]

[0050] Among them, y l Indicates the possible location of the target in the measurement space, Represents the matrix B l|k The inverse,

[0051] g is the prediction gate parameter, which represents the gate size and is used to ensure that the target falls into the prediction gate with a higher probability;

[0052] (2.4) Assume the false alarm probability is P fa , calculate the corresponding detection threshold v t

[0053] v t =-ln(P fa );

[0054] (2.5) Compare the data after signal processing in the prediction gate with the detection threshold v t Size:

[0055] If the energy of the point trace in the data exceeds the detection threshold v t , then this part of the trace is recorded as is the jth point at time l, and m is the number of points detected at time l;

[0056] If the energy of the point trace in the data does not exceed the detection threshold v t , it is considered not to be the target and this part of data is discarded.

[0057] Step 3: The detection result Z at the current time l l Perform data association with historical tracks to obtain the association result z l , and determine whether the associated track meets the n / m logic method criteria.

[0058] The data association mentioned above refers to an algorithm for associating detection point tracks with historical tracks, including but not limited to nearest neighbor, global nearest neighbor, probabilistic data association algorithm and joint probabilistic data association algorithm.

[0059] The n / m logic criterion is to judge whether the track is successfully associated with at least n target points within the final m frames. If it fails to be successfully associated with at least n target points, it means that the target has not been effectively observed within the m frames, and therefore the track can be terminated.

[0060] The specific implementation of this step is as follows:

[0061] (3.1) Use the association algorithm to associate the detection result at the current time l with the historical track data to obtain the association result z l ,

[0062]

[0063] in, is the historical track information, Z l is the detection result at the current moment l, and association represents the data association algorithm.

[0064] (3.2) For the association result z l and historical tracks The total number of target frames s detected from frame l-m+1 to frame l is counted as a whole, where m is the length of the time window in the n / m logic method criterion;

[0065] (3.3) Compare the target frame number s with the detection number threshold n in the n / m logic method criterion:

[0066] If s≥n, the n / m logic rule is satisfied, and step 4 is executed.

[0067] If s<n, the n / m logic rule is not met and the track is terminated.

[0068] Step 4: Determine the association result z l If it is empty, choose to return to step 1 or proceed to step 5

[0069] If the current association result is empty, it means that no target is detected at the current moment, and the data of the next moment needs to be detected. Therefore, it is necessary to return to step 1 to process the data of the next moment.

[0070] The specific implementation of this step is as follows:

[0071] If the association result z l If it is empty, set time l = l + 1 and return to step (1);

[0072] If the association result z l If it is not empty, continue tracking and execute step (5);

[0073] Step 5: Track and filter the associated points to calculate the estimated target state at the current time l With the covariance matrix D l|l , and update the track.

[0074] The tracking filter described in this step refers to a tracking filter algorithm, including but not limited to Kalman filtering, extended Kalman filtering, unscented Kalman filtering and other tracking filter algorithms.

[0075] The specific implementation of this step is as follows:

[0076] (5.1) Target estimated state

[0077] (5.1.1) According to the noise covariance matrix R measured at time l l The predicted covariance matrix B with the target measurement state l|k Calculate the innovation covariance matrix: S l =B l|k +R l ;

[0078] (5.1.2) According to the target prediction state covariance matrix D at time l l|k , target measurement function h at time l l The Jacobian matrix H l and the innovation covariance matrix S l , calculate the Kalman gain:

[0079] (5.1.3) According to the target prediction state at time l Kalman gain K l , the association result z at time l l The target predicted state of the measurement space Calculate the posterior mean of the target state at time l

[0080]

[0081] (5.2) According to the Kalman gain K l , target measurement function h at time l l The Jacobian matrix H l The covariance matrix D of the target predicted state at time l l|k , calculate the posterior covariance matrix D of the target state at time l l|l

[0082] D l|l =D l|k -K l H l D l|k

[0083] (5.3) Update track information

[0084] Step 6, performing multi-frame energy accumulation on the track segment of the updated track within the accumulation energy window to obtain the energy accumulation value Y l .

[0085] The accumulated energy window described in this step refers to the frame number interval composed of N points successfully associated at the end of the track. Its significance lies in that if only the energy of the last point is considered, it is difficult to distinguish whether it is a false point or a target point with a low signal-to-noise ratio; if the points of the entire track are considered, when the track is associated with a false point, it will be affected by many real target points in the track, which will lead to untimely cancellation of the false track.

[0086] The specific implementation of this step is as follows:

[0087] Assume that the accumulated energy window length is N, calculate the multi-frame energy accumulation value Y l :

[0088]

[0089] where γ i is the point track energy corresponding to time i in the track. If the track is not associated with a point track in a frame, the point track energy of that frame is considered to be 0. l-num(l) represents the frame number of the first target among the N consecutive successfully associated targets including time l.

[0090] Step 7: Set the false track maintenance probability to P t|fa , calculate the corresponding track decision threshold as v t_track .

[0091] The meaning of the false track maintenance probability described in this step is that when the point track corresponding to the accumulated multi-frame energy is a false point track, that is, the track segment is a false track, it is still determined to be the target track. Since the false track will affect the judgment of the target track, it is necessary to control the false track maintenance probability, and its specific implementation steps include the following:

[0092] (7.1) The noise signal-to-noise ratio SNR is modeled, which obeys an exponential distribution with a parameter of 1, that is:

[0093] f(SNR)=e -SNR ,(SNR>0)

[0094] (7.2) will pass through the first detection threshold v t The false point trace signal-to-noise ratio is recorded as ξ, and its probability density function is the truncation of the noise signal-to-noise ratio probability density function f(SNR). According to the form of the noise signal-to-noise ratio probability density function f(SNR), the false point traces that pass the first detection threshold v t The probability density function of the false point trace signal-to-noise ratio ξ is expressed as:

[0095]

[0096] (7.3) Let the intermediate variable μ i =ξ i -vt,i , where ξ i represents the signal-to-noise ratio of the false points in the i-th frame that pass the first detection threshold, v t,i represents the first detection threshold of the i-th frame, according to the first detection threshold v t The probability distribution form of the false point trace signal-to-noise ratio ξ, the intermediate variable μ i The probability density distribution is expressed as:

[0097]

[0098] That is, the intermediate variable μ i It follows an exponential distribution with parameter 1;

[0099] (7.4) Let the sum of the signal-to-noise ratios of multiple frames of false traces be Then it is related to the intermediate variable μ i The relationship is:

[0100]

[0101] Among them, v t,i is the threshold for detecting the data after signal processing at time i.

[0102] (7.5) Let the sum of the intermediate variables be is η, due to μ i They obey an exponential distribution with parameter 1 and are independent and identically distributed, so η obeys a gamma distribution with parameter (N,1);

[0103] (7.6) Assume the false alarm probability of η to be the false track maintenance probability P t|fa , according to its probability density distribution, the false alarm probability P of η t|fa It is expressed as:

[0104]

[0105] Among them, the lower limit of the integral is the preliminary threshold value, Γ(N)=(N-1)!

[0106] (7.7) According to step (7.4), we get According to this relationship, the probability of maintaining a false track is calculated as P t|fa In the case of , the track decision threshold of the sum of the signal-to-noise ratio of multiple frames of false points θ is:

[0107]

[0108] Among them, v t,i is the threshold for detecting the data after signal processing at time i.

[0109] Step 8, determine the energy accumulation value Yl Is it below the track decision threshold v? t_track , select whether to cancel the track.

[0110] If the multi-frame energy accumulation value of the track is lower than the track decision threshold, it means that it has 1-P t|fa The probability of being a false track is that since false tracks need to be suppressed, the value of the false track maintenance probability is generally set low. Therefore, the track has a high probability of being a false track and needs to be revoked.

[0111] The specific implementation of this step is to convert the energy accumulation value Y l and track decision threshold v t_track For comparison:

[0112] If l <v t_track , the track segment in the window is considered to be a false track and is cancelled;

[0113] If l ≥v t_track , continue tracking, set time l = l + 1, and return to step (1).

[0114] The effect of the present invention can be further illustrated by the following simulation test:

[0115] 1. Simulation conditions

[0116] Set the radar operating frequency to 1GHz, bandwidth to 1MHz, θ 3dB is 2°, the data rate is 2s, the prediction gate parameter g is 4, the first detection threshold is 12.8dB at the beginning of the track; the initial position of the target is [80km, 80km] T , the azimuth angle is 45°, and it moves away from the radar at a speed of 200m / s. The RCS fluctuation model of the target is Swilling I type. The mean signal-to-noise ratio of the target at 200km is set to 12dB, and the mean signal-to-noise ratio of the target is inversely proportional to the fourth power of the distance.

[0117] The data association algorithm used in the present invention selects the nearest neighbor algorithm, adopts the 4 / 10 logic method criterion, and the track filtering algorithm adopts the conversion measurement Kalman filtering algorithm. In the track judgment, the multi-frame energy accumulation window length N=4.

[0118] The traditional single-frame detection and tracking algorithm assumes that the false alarm probability corresponding to the detection threshold is 10 -6 , its data association algorithm selects the nearest neighbor algorithm, its track filtering algorithm selects the conversion measurement Kalman filtering algorithm, and if no target is detected for three consecutive frames, the track is determined to be terminated.

[0119] 2. Simulation content

[0120] Under the above simulation conditions, 100,000 Monte Carlo experiments were performed using the present invention and the traditional detection and tracking algorithm, and the relationship between the track maintenance probability and the distance was statistically analyzed. The results are as follows: Figure 2 As shown, the horizontal axis is the distance and the vertical axis is the track maintenance probability. Curve 1, Curve 2, Curve 3, and Curve 4 are respectively the statistical results of the relationship between the track maintenance probability and the distance under different frame false alarm probabilities and false track maintenance probabilities of the present invention; Curve 5 is the statistical result of traditional detection and tracking.

[0121] By comparing curve 1 with curve 3, it can be seen that the larger the first detection threshold, that is, the lower the probability of frame false alarm, the worse the track maintenance effect;

[0122] From the comparison between curve 1 and curve 2, it can be seen that the larger the track decision threshold, that is, the lower the probability of false track maintenance, the worse the track maintenance effect;

[0123] By comparing curve 1 with curve 5, we can see that when the track maintenance probability is 0.5 and the frame false alarm probability is 10 -3 , the probability of maintaining a false track is 10 -3 Under the condition of , the distance maintained by the invention is 186.5km, while the distance maintained by the traditional detection and tracking algorithm is 132.3km. Compared with the traditional algorithm, the track maintenance distance of the invention is pushed 40.1% further; when the track maintenance probability is 0.9 and the frame false alarm probability is 10 -3 , the probability of maintaining a false track is 10 -3 Under the condition of , the distance maintained by the present invention is 149.8km, and the distance maintained by the traditional detection and tracking algorithm is 121.5km. Compared with the traditional algorithm, the track maintenance distance of the present invention is pushed further by 23.3%;

[0124] By comparing curve 4 with curve 5, we can see that when the track maintenance probability is 0.5 and the frame false alarm probability is 10 -4 , the probability of maintaining a false track is 10 -4 Under the condition of , the distance maintained by the invention is 178.1km, while the distance maintained by the traditional detection and tracking algorithm is 132.3km. Compared with the traditional algorithm, the distance maintained by the invention is 34.6% farther; when the probability of track maintenance is 0.9 and the probability of frame false alarm is 10 -4 , the probability of maintaining a false track is 10 -4 Under the condition of , the distance maintained by the present invention is 146.7km, and the distance maintained by the traditional detection and tracking algorithm is 121.5km. Compared with the traditional algorithm, the track maintenance distance of the present invention is pushed 20.7% further;

[0125] In summary, compared with the traditional single-frame detection and tracking algorithm, the present invention can improve the track maintenance probability at a long distance while ensuring a low false track maintenance probability.

Claims

1. A track management method based on multi-frame energy accumulation, characterized in that: include: (1) Estimated target state based on the last frame after the start of the track With its state covariance matrix D k|k , predict the target state of the next frame With its covariance matrix D l|k ; (2) Based on forecast information and D l|k Establish a prediction gate and set the false alarm probability to P fa , perform low threshold detection on the data that has been processed and falls into the prediction wave gate, and obtain the detection result Z at the current time l l ; (3) The detection result Z at the current time l l Perform data association with historical tracks to obtain the association result z l , and determine whether the associated track meets the n / m logic method criteria: If not satisfied, the track is determined to be terminated; If satisfied, proceed to step (4); (4) Determine the association result z l Is it empty? If the association result z l If it is empty, set time l = l + 1 and return to step (1); If the association result z l If it is not empty, continue tracking and execute step (5); (5) Track and filter the associated points to calculate the estimated target state at the current time l With the covariance matrix D l|l , and update the track; (6) Assume that the accumulated energy window length is N, and perform multi-frame energy accumulation on the track segment of the updated track within the accumulated energy window to obtain the energy accumulation value Y l ; (7) Set the false track maintenance probability as P t|fa , calculate the corresponding track decision threshold as v t_track ; (8) Determine the energy accumulation value Y l Is it below the track decision threshold v? t_track If it is lower than the threshold, the track segment in the window is considered to be a false track and will be cancelled; Otherwise, continue tracking, set time l=l+1, and return to step (1).

2. The method according to claim 1, characterized in that Step (1) Estimated state of the target in the last frame after the start of the track With its state covariance matrix D k|k , predict the target state of the next frame With its covariance matrix D l|k , implemented as follows: (1a) Estimated state based on the target at time k Predict the target state at time l Among them, F l|k is the target state transfer matrix from time k to time l; (1b) According to the target state covariance matrix D at time k k|k , calculate the target prediction state covariance matrix D at time l l|k : Among them, Q k|k is the noise covariance matrix of the process from time k to time l.

3. The method according to claim 1, characterized in that In step (2), based on the prediction information and D l|k , establish a prediction wave gate, and implement it as follows: (2a) Convert the target prediction state to the measurement space, denoted as Among them, h l is the target measurement function at time l; (2b) Calculate the predicted covariance matrix of the target measurement state: B l|k =H l D l|k H l T , Among them, H l Expressed as the target measurement function h at time l l The Jacobian matrix, H l T Indicates H l The transpose of (2c) Measure the state according to the predicted target and the predicted covariance matrix B l|k , establish the following prediction gate: Among them, y l Indicates the possible location of the target in the measurement space, Represents the matrix B l|k The inverse, g is the prediction gate parameter, which represents the gate size and is used to ensure that the target falls into the prediction gate with a higher probability.

4. The method according to claim 1, characterized in that: In step (2), the false alarm probability of the data that has been processed and falls into the prediction gate is P fa The low threshold detection is implemented as follows: (2d) Calculate the false alarm probability P fa The corresponding detection threshold v t v t =-ln(P fa ); (2e) Compare the data after signal processing within the prediction gate with the detection threshold v t Size: If the energy of the point trace in the data exceeds the detection threshold v t , then this part of the trace is recorded as is the jth point at time l, and m is the number of points detected at time l; If the energy of the point trace in the data does not exceed the detection threshold v t , it is considered not to be the target and this part of data is discarded.

5. The method according to claim 1, characterized in that: Step (3) associate the detection result at the current time l with the historical track data to obtain the association result z l , expressed as follows in, is the historical track information, Z l is the detection result at the current moment l, and association represents the data association algorithm.

6. The method according to claim 1, characterized in that In step (3), it is determined whether the associated track meets the n / m logic method criteria, which is implemented as follows: (3a) For the association result z l and historical tracks The total number of target frames s detected from frame l-m+1 to frame l is counted as a whole, where m is the length of the time window in the n / m logic method criterion; (3b) Compare the target frame number s with the detection number threshold n in the n / m logic rule: s≥n, then the n / m logic rule is satisfied, If s<n, the n / m logic rule is not met.

7. The method according to claim 1, characterized in that In step (5), the target estimated state at the current time l is calculated With the covariance matrix D l|l , and update the track, which can be achieved as follows, (5a) Target estimated state (5a1) According to the noise covariance matrix R measured at time l l The predicted covariance matrix B with the target measurement state l|k Calculate the innovation covariance matrix: S l =B l|k +R l ; (5a2) According to the target prediction state covariance matrix D at time l l|k , target measurement function h at time l l The Jacobian matrix H l and the innovation covariance matrix S l , calculate the Kalman gain: (5a3) According to the target prediction state at time l Kalman gain K l , the association result z at time l l The target predicted state of the measurement space Calculate the posterior mean of the target state at time l (5b) According to the Kalman gain K l , target measurement function h at time l l The Jacobian matrix H l The covariance matrix D of the target predicted state at time l l|k , calculate the posterior covariance matrix D of the target state at time l l|l D l|l =D l|k -K l H l D l|k (5c) Update track information 8. The method according to claim 1, characterized in that The energy accumulation value Y is obtained in step (6) l , which is expressed as follows, where γ i is the point track energy corresponding to time i in the track. If the track is not associated with a point track in a frame, the point track energy of that frame is considered to be 0. N is the accumulated energy window length of the track. l-num(l) represents the frame number of the first target in the N consecutive successfully associated targets including time l.

9. The method according to claim 1, characterized in that: In step (7), the false track maintenance probability is set to P t|fa , calculate the corresponding track decision threshold as v t_track , implemented as follows, (7a) Set the false track maintenance probability P t|fa It is expressed as: Among them, the lower limit of the integral is the preliminary threshold value, Γ(N)=(N-1)! (7b) Based on the preliminary threshold Calculate the track decision threshold v t_track Among them, v t,i is the threshold for detecting the data after signal processing at time i.

Citation Information

Patent Citations

  • Multi-feature fusion based infrared small dim moving target track starting method

    CN101930072A

  • Track-before-detect method based on flight path inspection

    CN102419437A

  • Detection and tracking integration method for frame constant false-alarm target

    CN103809173A

  • Radar weak object detection method based on image characteristics

    CN105022046A

  • First-initiation second-confirmation radar track initiation method

    CN106054150A

Cited By

  • Dense multi-unmanned aerial vehicle target real-time tracking method based on GM-PHD filtering

    CN121880962A