Multi-target tracking method and system for low-transmission-bandwidth sensor network
By quantifying and transmitting the target measuring value residues of the sensor network under the Karman filtering framework, the problem of large information transmission and low security in the traditional multi -target tracking method is solved, and more efficient transmission and stronger resistance have been achieved Interference.
Patent Information
- Application Number
- CN202411700859.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-26
- Publication Date
- 2025-01-28
AI Technical Summary
In the prior art, the traditional multi-target tracking method is suitable for conventional sensor networks, which directly transmits the quantized target measurement value, resulting in large amounts of information transmission and severe radiation, which are easy to be discovered, interfered with and attacked by the other party, especially at low transmission bandwidth. There are security and efficiency problems in the sensor network.
Under the framework of Karman filtering, the residues of the target measurement value are quantified, and the transmission of quantitative residues, rather than the target measurement value after direct transmission quantification. After receiving quantitative residues at the fusion center, the quantitative residual matrix is formed and allocated to process each target measurement value and each target trace to form tracking results.
By reducing the amount of transmission information and radiation of the sensor network, the transmission efficiency is improved, while reducing the probability of being discovered, interference and attack by the other party, and avoiding the decrease in sensor performance caused by interference.
Smart Images
Figure CN119383567A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information processing technology, and in particular to a multi-target tracking method and system for a low transmission bandwidth sensor network. Background Art
[0002] To accurately describe the information in this article, the following definitions are given:
[0003] The low transmission bandwidth referred to in this application is the amount of information transmitted between 1 bit / s and 3 bit / s.
[0004] Multi-target tracking refers to tracking more than two targets.
[0005] With the development of technology, the confrontation environment of various sensors is becoming more and more complex, and the active or passive interference they are subject to is also increasing. For sensor networks, tracking the target of interest and forming a track is one of its main functions.
[0006] Traditional multi-target tracking methods are suitable for conventional sensor networks. Conventional sensors directly quantify the target measurement values and transmit them to the fusion center. After receiving the quantified data, the fusion center tracks the target and forms a tracking result. Conventional sensors directly transmit quantified target measurement values, which has the characteristics of large information transmission volume and high radiation. Especially for sensors that transmit wireless information, it is easy to be discovered, interfered and attacked by the other party. Therefore, multi-target tracking methods for low transmission bandwidth sensor networks need to be studied urgently. Summary of the invention
[0007] In order to solve the problems in the prior art and achieve the purpose of multi-target tracking for low transmission bandwidth sensor networks, the present application provides a multi-target tracking method and system for low transmission bandwidth sensor networks.
[0008] In a first aspect, a multi-target tracking method and system for a low transmission bandwidth sensor network are provided, which adopts the following technical solutions:
[0009] The fusion center obtains a one-step prediction and sends it to the sensor, where the one-step prediction includes a one-step prediction mean and a one-step prediction mean square error;
[0010] The sensor forms a quantized residual according to its own target measurement value and the one-step prediction, and sends the quantized residual to the fusion center;
[0011] The fusion center forms a quantized residual matrix from the quantized residual and distributes it;
[0012] The fusion center processes each target measurement value and each target track in the assigned quantized residual matrix to form a tracking processing result.
[0013] Furthermore, when the fusion center obtains the one-step prediction, the method adopted is as follows:
[0014] The fusion center sets initial parameters, including initial target state, initial covariance matrix, measurement equation, measurement error, quantization threshold and number of quantization intervals;
[0015] The fusion center calculates the one-step prediction mean: Calculate the first The one-step prediction mean of the target tracks The method is as follows:
[0016]
[0017] in,
[0018] is the measurement equation used to convert the target state into the target measurement value, where Indicates the moment; is the state transfer matrix, which is used to predict the state at the current moment based on the state at the previous moment; for The predicted mean value of the target state at the moment; for The estimated mean of the target state at time ; The serial number representing the target measurement value;
[0019] The fusion center calculates the mean square error of the one-step prediction value :
[0020]
[0021] in, The serial number representing the target measurement value; is the measurement equation used to convert the target state into the target measurement value, where Indicates the moment; for The target state covariance matrix prediction value at the moment; for The transpose of , T is the transpose; is the measurement error.
[0022] Furthermore, when the sensor forms a quantized residual based on its own target measurement value and the one-step prediction, the method adopted is as follows:
[0023] The sensor sets a quantization threshold and a number of quantization intervals, and the quantization threshold and the number of quantization intervals are consistent with the quantization threshold and the number of quantization intervals set by the fusion center;
[0024] The residual is calculated as follows:
[0025]
[0026] in, is the residual, Indicates the time, The serial number representing the target measurement value; Measured by the sensor The target measurement value at the moment; The sensor receives the data sent by the fusion center. The one-step forecast mean at time; The sensor receives the data sent by the fusion center. The mean square error of the one-step forecast value at time;
[0027] The sensor quantizes the residual according to the quantization threshold and the number of quantization intervals to obtain a quantized residual.
[0028] Furthermore, when the fusion center forms the quantized residual into a quantized residual matrix and distributes it, the quantized residual matrix is distributed using the Hungarian algorithm.
[0029] Furthermore, the fusion center processes each target measurement value and each target track in the allocated quantized residual matrix to form a tracking processing result, and the method used is as follows:
[0030] There are three situations for the assigned target measurement values and target tracks:
[0031] Case I: Target track assigned to target measurement value
[0032] No. The target track is assigned to target measurement values, according to the quantized residual , calculate the Target tracks currently The estimated mean value of the target state at time and the target state covariance matrix estimate , that is, complete the Update of target track;
[0033] Case II: Target track is not assigned to target measurement value
[0034] For the first If there is a target track, the target track cannot be updated and the target track is marked with the preparation deletion flag D_flg. If the target track has not been updated for T consecutive times, that is, when the D_flg of the target track is equal to T, the target track is deleted from the track file;
[0035] Case III: Target measurement value has no corresponding target track
[0036] For the first target measurements, based on the quantized residual , and the residual before quantization is obtained by inverse quantization , and then get the first Target measurements ; For the dead reckoning that is not assigned a target track Target measurements , start the track, create a temporary track file, and mark the temporary track with a confirmation flag C_flg. If the temporary track is updated for H consecutive times, that is, when the C_flg of the temporary track is equal to H, the temporary track is confirmed in the track file and is considered to be not a temporary track but a confirmed track.
[0037] In a second aspect, a multi-target tracking system for a low transmission bandwidth sensor network is provided, which is used to implement the multi-target tracking method for a low transmission bandwidth sensor network described in the first aspect, and the system includes:
[0038] Fusion center: used to obtain one-step prediction and send it to the sensor; receive the quantized residual sent by the sensor to form a quantized residual matrix, distribute the quantized residual matrix, process each distributed target measurement value and each target track, and form a tracking processing result.
[0039] Sensor: Receives the one-step prediction sent by the fusion center, combines it with its own target measurement value to form a quantized residual, and sends the quantized residual to the fusion center.
[0040] The beneficial effects of the present invention are as follows:
[0041] The multi-target tracking method in the prior art is applicable to conventional sensor networks. Conventional sensors directly quantify the target measurement values and transmit them to the fusion center. The fusion center receives the quantized data and processes it to form a tracking result. The prior art directly transmits the quantized target measurement values, which has the characteristics of large information transmission volume and high radiation. In particular, sensors that perform wireless information transmission are easily discovered, interfered with, and attacked by the other party.
[0042] This application is for low transmission bandwidth sensor networks. It does not directly transmit the quantized target measurement value, but quantizes the residual of the target measurement value under the framework of Kalman filtering and transmits the quantized residual. It reduces the amount of information and radiation transmitted by the sensor network, and while the transmission efficiency is higher, it also reduces the probability of being discovered, interfered with and attacked by the other party, avoiding the degradation of sensor performance due to interference. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1is a flow chart of a multi-target tracking method for a low transmission bandwidth sensor network according to an embodiment of the present invention;
[0044] Figure 2 It is the tracking processing result of two targets in the embodiment of the present invention. DETAILED DESCRIPTION
[0045] The present invention is further described below.
[0046] The present invention provides a multi-target tracking method for a low transmission bandwidth sensor network. The sensor network consists of a fusion center and sensors. The sensor does not directly quantize the target measurement value and transmit it to the fusion center, but quantizes the residual of the target measurement value based on the Kalman filter framework and then transmits it. After receiving the quantized residual, the fusion center tracks the target and forms a tracking processing result. This can effectively reduce the amount of information transmitted by the sensor network and reduce the probability of being discovered, interfered with and attacked by the other party.
[0047] In this embodiment, the detection data of two civil air traffic control radars are used for simulation processing, one of which is set as the fusion center and the other is set as the sensor. The flowchart of the multi-target tracking method for low transmission bandwidth sensor networks is as follows: Figure 1 shown.
[0048] Step 1: The fusion center obtains a one-step prediction and sends it to the sensor. The one-step prediction includes a one-step prediction mean and a one-step prediction mean square error.
[0049] 1-1. Initial parameter setting at the fusion center
[0050] The initial parameters include:
[0051] Initial target state ;
[0052] Initial covariance matrix ;
[0053] Measurement equation , which is used to convert the target state into the target measurement value, where Indicates the moment;
[0054] Measurement error ,in Indicates the moment;
[0055] Quantization Threshold ,in is an integer, , ;
[0056] Number of quantization intervals ,in , is an integer;
[0057] 1-2. The fusion center calculates the one-step prediction mean and the one-step prediction mean square error
[0058] Calculate the Target track The one-step forecast mean at time :
[0059] (1.1)
[0060] in, (1.2)
[0061] is the measurement equation used to convert the target state into the target measurement value, where Indicates the moment;
[0062] is the state transfer matrix, which is used to predict the state at the current moment based on the state at the previous moment;
[0063] for The predicted mean value of the target state at the moment;
[0064] for The estimated mean of the target state at time ;
[0065] Indicates the sequence number of the target measurement value.
[0066] Calculate the mean square error of the one-step forecast :
[0067] (1.3)
[0068] in, The serial number representing the target measurement value;
[0069] is the measurement equation used to convert the target state into the target measurement value, where Indicates the moment;
[0070] for The target state covariance matrix prediction value at the moment;
[0071] for The transpose of (T is the transpose);
[0072] is the measurement error.
[0073] Step 2: The sensor forms a quantized residual based on its own target measurement value and one-step prediction, and sends the quantized residual to the fusion center.
[0074] 2-1. Initial parameter setting of the sensor
[0075] The initial parameters include:
[0076] Quantization Threshold ,in is an integer, , ;
[0077] Number of quantization intervals ,in , is an integer;
[0078] Among them, the quantization threshold and quantization interval number set by the sensor The quantization threshold and quantization interval number to be set with the fusion center Be consistent.
[0079] 2-2. Sensor calculation residual
[0080] The residual is calculated as follows:
[0081] (1.4)
[0082] in, is the residual, Indicates the time, The serial number representing the target measurement value;
[0083] Measured by the sensor Moment target measurement value;
[0084] The sensor receives the data sent by the fusion center. The one-step forecast mean at time;
[0085] The sensor receives the data sent by the fusion center. The mean square error of the one-step forecast at time .
[0086] 2-3. The sensor quantizes the residual to obtain a quantized residual.
[0087] According to the quantization threshold and quantization interval number set in step 2-1 , for the residual Quantize and get the quantized residual .
[0088] 2-4. The sensor will quantify the residual After encoding, it is sent to the fusion center.
[0089] Step 3: The fusion center converts the quantized residual into a quantized residual matrix and distributes it
[0090] The fusion center decodes the received encoded quantized residual to obtain the quantized residual .
[0091] When the sensor obtains multiple target measurements After that, For the moment, is the target measurement value sequence number. According to step 2-2 to step 2-3, the fusion center obtains the quantized residuals of multiple target measurement values and multiple target tracks. ,in For the moment, is the target measurement value number, is the target track number; if the current At time t, there are M target measurements and N target tracks, and the fusion center will receive M*N quantized residuals , generate an M*N quantized residual matrix.
[0092] The M*N quantized residual matrix is subjected to Hungarian allocation, and each target measurement value is paired with each target track to obtain the allocation result.
[0093] Step 4: The fusion center processes each target measurement value and each target track in the allocated quantized residual matrix to form a tracking processing result.
[0094] There are three situations for the assigned target measurement values and target tracks:
[0095] Case I: Target track assigned to target measurement value
[0096] For example, The target track is assigned to target measurement values, according to the quantized residual , calculate the Target tracks currently The estimated mean value of the target state at time and the target state covariance matrix estimate , that is, complete the Update of target track.
[0097] Among them, Target tracks currently The estimated mean value of the target state at time The calculation method is as follows:
[0098] The Fusion Center The quantized residual of the target track and the set quantization threshold ,get The quantization interval ; Under the Kalman filter architecture, calculate the target state , the calculation formula is as follows:
[0099] (1.5)
[0100] (1.6)
[0101] in for The predicted mean value of the target state at the moment;
[0102] is the measurement equation used to convert the target state into the target measurement value, where Indicates the moment;
[0103] for The target state covariance matrix prediction value at the moment;
[0104] for The transpose of (T is the transpose);
[0105] is the measurement error;
[0106] is an exponential function;
[0107] is the distribution function that obeys the standard normal distribution;
[0108] and is the boundary of the quantization interval.
[0109] No. Target tracks The estimated value of the target state covariance matrix at time The calculation method is as follows:
[0110] (1.7)
[0111]
[0112] in, for The target state covariance matrix prediction value at the moment;
[0113] for The transpose of (T is the transpose);
[0114] is the measurement error;
[0115] is an exponential function;
[0116] is the distribution function that obeys the standard normal distribution;
[0117] and is the boundary of the quantization interval.
[0118] Case II: Target track is not assigned to target measurement value
[0119] For the first If there is a target track, the target track cannot be updated and the target track is marked with the preparation deletion flag D_flg. If the target track has not been updated for T consecutive times, that is, when the D_flg of the target track is equal to T, the target track is deleted from the track file;
[0120] Case III: Target measurement value has no corresponding target track
[0121] For the first target measurements, based on the quantized residual , and the residual before quantization is obtained by inverse quantization , and through the formula (1.4) in step 2-2, we can get the pre-quantization calculation Target measurements , for the dead reckoning that is not assigned to a target track Target measurements , start the track, create a temporary track file, and set the initial state of the target track , the initial covariance matrix, and mark the target track with a confirmation flag C_flg. If the target track is updated for H consecutive times, that is, the C_flg of the target track is equal to H, the target track is confirmed in the track file and it is considered that it is not a temporary track but a confirmed track.
[0122] The fusion center uses the calculated target state estimated mean and target state covariance matrix estimated value to calculate the one-step prediction mean and one-step prediction mean square error of each target track in the next operation cycle, and repeats steps 1 to 4, where the initial parameter settings of the fusion center and the sensor do not need to be performed again. Until the multi-target tracking operation is completed, the final multi-target tracking processing result is formed.
[0123] In this embodiment, the tracking results of the two targets are as follows: Figure 2 As shown, target track ① is a non-maneuvering target, and target track ② is a maneuvering target. The present application realizes tracking of multiple targets of interest based on quantized residuals and forming stable tracks for low transmission bandwidth sensor networks.
[0124] The present invention also provides a multi-target tracking system for a low transmission bandwidth sensor network, the system is used to implement a multi-target tracking method for a low transmission bandwidth sensor network, and the system includes the following parts:
[0125] Fusion center: The fusion center is used to obtain the one-step prediction value and send it to the sensor; receive the quantized residual sent by the sensor to form a quantized residual matrix, and distribute the quantized residual matrix; process each distributed target measurement value and each target track to form the final multi-target tracking processing result. The details are as follows:
[0126] The fusion center is responsible for obtaining the one-step prediction and sending it to the sensor. The one-step prediction includes the one-step prediction mean and the one-step prediction mean square error. When the fusion center obtains the one-step prediction, it is based on the Kalman framework.
[0127] The fusion center is responsible for receiving the quantized residuals sent by the sensor to form a quantized residual matrix and distribute the quantized residual matrix as follows:
[0128] The fusion center decodes the received quantized residuals to obtain quantized residuals. When the sensor obtains multiple target measurement values, the fusion center obtains the quantized residuals of multiple target measurement values and multiple target tracks. At this moment, there are M target measurements and N target tracks. The fusion center will receive M*N quantized residuals and generate an M*N quantized residual matrix. The fusion center performs Hungarian allocation on the M*N quantized residual matrix, and each target measurement value is assigned and paired with each target track.
[0129] The fusion center is responsible for processing each assigned target measurement value and each target track to form the final multi-target tracking processing result, as follows:
[0130] If the target track is assigned to the target measurement, e.g. The target track is assigned to target measurement values, according to the quantized residual , calculate the Target tracks currently The estimated mean value of the target state at time and the target state covariance matrix estimate , that is, complete the Update of target track.
[0131] If the target track is not assigned a target measurement value, If there is a target track, the target track cannot be updated and the track is marked with the preparation deletion flag D_flg. If the target track has not been updated for T consecutive times, that is, when the D_flg of the target track is equal to T, the target track is deleted from the track file;
[0132] Jordi There is no corresponding target track for the target measurement value, for example, target measurements, based on the quantized residual , and the residual before quantization is obtained by inverse quantization , and reversely infer through formula (1.4) to obtain the first Target measurements , for the dead reckoning that is not assigned to a target track Target measurements , start the track, create a temporary track file, and set the initial state of the target track , the initial covariance matrix, and mark the target track with a confirmation flag C_flg. If the target track is updated for H consecutive times, that is, the C_flg of the track is equal to H, the target track is confirmed in the track file and it is considered that it is not a temporary track but a confirmed track.
[0133] Sensor: Receives the one-step prediction value sent by the fusion center, combines it with its own target measurement value to form a quantized residual, and sends the quantized residual to the fusion center. The details are as follows:
[0134] After receiving the one-step prediction mean and one-step prediction mean square error sent by the fusion center, the sensor combines its own measured target value and uses formula (1.4) to calculate the residual, quantizes the residual, and obtains the quantized residual. The sensor encodes the quantized residual and sends it to the fusion center.
[0135] Although the present invention has been disclosed as above with preferred embodiments, the embodiments are not intended to limit the present invention. Any equivalent changes or modifications made without departing from the spirit and scope of the present invention also belong to the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the content defined by the claims of this application.
Claims
1. A multi-target tracking method for low transmission bandwidth sensor networks, characterized in that: The steps include: The fusion center obtains a one-step prediction and sends it to the sensor, where the one-step prediction includes a one-step prediction mean and a one-step prediction mean square error; The sensor forms a quantized residual according to its own target measurement value and the one-step prediction, and sends the quantized residual to the fusion center; The fusion center forms a quantized residual matrix from the quantized residual and distributes it; The fusion center processes each target measurement value and each target track in the assigned quantized residual matrix to form a tracking processing result.
2. The multi-target tracking method for low transmission bandwidth sensor networks according to claim 1, characterized in that: When the fusion center obtains the one-step prediction, the method adopted is as follows: The fusion center sets initial parameters, including initial target state, initial covariance matrix, measurement equation, measurement error, quantization threshold and number of quantization intervals; The fusion center calculates the one-step prediction mean: Calculate the first The one-step prediction mean of the target tracks The method is as follows: in, is the measurement equation used to convert the target state into the target measurement value, where Indicates the moment; is the state transfer matrix, which is used to predict the state at the current moment based on the state at the previous moment; for The predicted mean value of the target state at the moment; for The estimated mean of the target state at time ; The serial number representing the target measurement value; The fusion center calculates the mean square error of the one-step prediction value : in, The serial number representing the target measurement value; is the measurement equation used to convert the target state into the target measurement value, where Indicates the moment; for The target state covariance matrix prediction value at the moment; for The transpose of , T is the transpose; is the measurement error.
3. The multi-target tracking method for low transmission bandwidth sensor networks according to claim 2, characterized in that: When the sensor forms a quantized residual based on its own target measurement value and the one-step prediction, the method adopted is as follows: The sensor sets a quantization threshold and a number of quantization intervals, and the quantization threshold and the number of quantization intervals are consistent with the quantization threshold and the number of quantization intervals set by the fusion center; The residual is calculated as follows: in, is the residual, Indicates the time, The serial number representing the target measurement value; Measured by the sensor The target measurement value at the moment; The sensor receives the data sent by the fusion center. The one-step forecast mean at time; The sensor receives the data sent by the fusion center. The mean square error of the one-step forecast value at time; The sensor quantizes the residual according to the quantization threshold and the number of quantization intervals to obtain a quantized residual.
4. The multi-target tracking method for low transmission bandwidth sensor networks according to claim 3, characterized in that: When the fusion center forms the quantized residual into a quantized residual matrix and distributes it, the quantized residual matrix is distributed using the Hungarian algorithm.
5. The multi-target tracking method for low transmission bandwidth sensor networks according to claim 4, characterized in that: The fusion center processes each target measurement value and each target track in the assigned quantized residual matrix to form a tracking processing result. The method used is as follows: There are three situations for the assigned target measurement values and target tracks: Case I: Target track assigned to target measurement value No. The target track is assigned to target measurement values, according to the quantized residual , calculate the Target tracks currently The estimated mean value of the target state at time and the target state covariance matrix estimate , that is, complete the Update of target track; Case II: Target track is not assigned to target measurement value For the first If there is a target track, the target track cannot be updated and the target track is marked with the preparation deletion flag D_flg. If the target track has not been updated for T consecutive times, that is, when the D_flg of the target track is equal to T, the target track is deleted from the track file; Case III: Target measurement value has no corresponding target track For the first target measurements, based on the quantized residual , and the residual before quantization is obtained by inverse quantization , and then get the first Target measurements ; For the dead reckoning that is not assigned a target track Target measurements , start the track, create a temporary track file, and mark the temporary track with a confirmation flag C_flg. If the temporary track is updated for H consecutive times, that is, when the C_flg of the temporary track is equal to H, the temporary track is confirmed in the track file and is considered to be not a temporary track but a confirmed track.
6. A multi-target tracking system for low transmission bandwidth sensor networks, used to implement the multi-target tracking method for low transmission bandwidth sensor networks as claimed in claim 1, characterized in that: The system comprises: Fusion center: used to obtain one-step prediction and send it to the sensor; receive the quantized residual sent by the sensor to form a quantized residual matrix, distribute the quantized residual matrix, process each distributed target measurement value and each target track, and form a tracking processing result. Sensor: Receives the one-step prediction sent by the fusion center, combines it with its own target measurement value to form a quantized residual, and sends the quantized residual to the fusion center.