[0036] In order to make the objectives, technical solutions and advantages of the present invention clearer, the following describes in further detail a target tracking method based on distributed processing in a wireless sensor network of the present invention with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.
[0037] The wireless sensor network target tracking algorithm based on distributed processing of the present invention is to overcome the complexity of traditional tracking methods, the large amount of information transmission, the large amount of calculation, and the large energy consumption. It is introducing the extended Kalman filter algorithm At the same time, the principle of minimum state error covariance matrix trace is used to select task nodes, which reduces processor energy consumption, reduces data transmission and information interaction between nodes, and improves tracking accuracy.
[0038] The present invention provides a target tracking method based on a distributed processing wireless sensor network, such as figure 1 As shown, including the following steps:
[0039] Step 100: Using the estimation information of the target state at the last moment and the observation data of the task node at the current moment, perform an extended Kalman filtering algorithm to obtain the target state estimation information at the current moment, the target state estimation information including the current position and movement speed of the moving target , The observation data refers to the distance between the task node and the target at the current moment;
[0040] Step 101: According to the position estimation information in the target state estimation information at the current moment, calculate the prior estimation error covariance matrix trace of the candidate task node at the next moment, compare and select the node corresponding to the smallest trace as the next moment task node ;
[0041] Step 102: When the a priori estimation error covariance matrix trace corresponding to the task node at the next time in step 101 is greater than the set threshold, the target trajectory correction algorithm is used to retrieve the target state estimation information at the current time to achieve target positioning and tracking.
[0042] Repeat steps 100 to 102 to keep track of the target.
[0043] In the step 100, the task node is in an awake state, and the non-task node is in a dormant state, figure 2 It shows the mutual relationship between task nodes, non-task nodes, moving targets and base stations in single target tracking, and the process of continuous switching of task nodes as the target moves. The task node is responsible for obtaining observation data, running the extended Kalman filter algorithm, electing the task node at the next time, saving the target state information and ensuring the stable transmission of key information between different task nodes at adjacent times.
[0044] The following describes in detail the extended Kalman filter algorithm process of the present invention, specifically, including the following steps:
[0045] Step 1001: Obtain the target state estimation information at the current moment and the prior estimation error covariance matrix at the current moment according to the moving target system equation from the target state estimation information at the last moment and the posterior estimation error covariance matrix at the last moment;
[0046] In step 1002, the measurement value of the task node is estimated from the position information in the target state estimation information at the current time in step 1001 through the measurement equation, and the difference between the measurement value and the observation data of the task node at the current time is taken as the residual difference;
[0047] Step 1003: From the Kalman gain value, the Jacobian matrix of the measurement equation and the residual error of the observed variable, correct the state estimation information of the target at the current moment, and obtain the current moment posterior estimation error covariance matrix.
[0048] The following describes in detail the process of using the principle of minimum trace of the covariance matrix to elect task nodes in the present invention, which specifically includes the following processes:
[0049] Step 1011: From the target state estimation information at the current moment and the current moment posterior estimation error covariance matrix, obtain the next moment target state estimation information and the next moment a priori estimation error covariance value according to the moving target system equation;
[0050] Step 1012: From the Kalman gain value and the Jacobian matrix of the measurement equations of each candidate task node, correct the next time a priori estimation error covariance value of each candidate task node, and find the next time after each candidate task node is corrected Priori estimation error covariance matrix trace;
[0051] Step 1013: Compare the corrected next-time a priori estimation error covariance matrix trace of each candidate task node in step 1012, and elect the node corresponding to the smallest trace as the next task node.
[0052] In the step 101, when the task node at the next time is not the task node at the current time, the task node at the current time transmits the current target state estimation information to the task node at the next time.
[0053] In the step 101, the task node at the current time not only transmits the current target state estimation information to the task node at the next time, but also transmits the information to the base station for backup. When the task node at the next time cannot work normally due to a failure At this time, the task node secondary election program is started, that is, the target state estimation information of the task node at the previous time backed up in the base station is used to find the node with the smallest target state error covariance matrix as the next time task node.
[0054] The candidate task node in step 101 is selected from neighbor nodes, and includes the following steps:
[0055] Each node in the wireless sensor network has a neighbor node. The distance value between the current task node and the non-task node is compared with the set value, and then the non-task node whose distance value is within the set value range is defined as the neighbor node ;
[0056] Calculate the distance between the two coordinates according to the coordinate value and the neighbor node coordinate value in the position information in the target state estimation information at the current time, and then compare the distance value with the set value, and the distance is within the set value range The neighbor node of is the candidate task node.
[0057] The main purpose of defining neighbor nodes is to clarify the scheduling set selected by the task node, because as the target moves, the task node is constantly changing, and the corresponding candidate task node set is also changing accordingly. The establishment of neighbor nodes can reduce the time and energy for task node selection using the extended Kalman filter (EKF) minimum trace, improve the real-time performance of target tracking and extend the life cycle of the system.
[0058] In the step 102, when the a priori estimation error covariance matrix trace corresponding to the task node at the next moment is greater than the set threshold, it is considered that the deviation of the predicted target position from the target actual position exceeds the system allowable range, and each candidate task is passed The node executes the extended Kalman filter algorithm to predict the coordinate value of the target, and then calculates the average of these coordinate values as the corrected position of the target.
[0059] In the step 100, if there is currently no historical state information of the moving target, the current observation data is used to perform a trajectory start algorithm to obtain the state estimation information of the target, and to establish tracking of the new target.
[0060] The track start algorithm includes:
[0061] For a local area formed by any node and its neighbor nodes, if the number of nodes perceiving the target is greater than the specified threshold, it is considered that a new target appears near the node location, and the three-sided positioning algorithm is used to calculate the coordinate value of the target in the local area. As the starting position of the new target.
[0062] In the present invention, when the number of nodes perceiving the target is greater than 3, the trilateral positioning algorithm is used to locate the target:
[0063] Such as Figure 4 As shown, in a two-dimensional system, we can pass the coordinates of three nodes (x 1 , Y 1 ), (x 2 , Y 2 ), (x 3 , Y 3 ) And the distance d from these nodes to the target 1 , D 2 , D 3 Find the coordinates (x, y) of the target:
[0064] (x 1 -x) 2 +(y 1 -y) 2 =d 1 2
[0065] (x 2 -x) 2 +(y 2 -y) 2 =d 2 2 (1)
[0066] (x 3 -x) 2 +(y 3 -y) 2 =d 3 2
[0067] Can get from it
[0068] x y = 2 ( x 1 - x 3 ) 2 ( y 1 - y 3 ) 2 ( x 2 - x 3 ) 2 ( y 2 - y 3 ) - 1 x 1 2 - x 3 2 + y 1 2 - y 3 2 + d 3 2 - d 1 2 x 2 2 - x 3 2 + y 2 2 - y 3 2 + d 3 2 - d 2 2 - - - ( 2 )
[0069] The principle of selecting the node with the smallest trace of the error covariance matrix in step 101 includes the following steps:
[0070] In the extended Kalman filter algorithm, theoretically, by observing the covariance of the measurement noise and the change trend of the prior estimation error covariance of the target at the next moment, the credibility of the predicted value can be judged. As the covariance of the measurement noise tends to zero, The weight of the measured variable becomes larger and larger, and the weight of the predicted value becomes smaller and smaller; on the other hand, as the prior estimation error covariance tends to zero, the weight of the measured variable becomes smaller and smaller, and the prediction of the measured variable The weight of the value is getting bigger and bigger. However, because the observed noise covariance is related to the observed noise model and changes dynamically, it cannot be achieved in practical applications. Therefore, the target prediction is generally judged by using the priori estimation error covariance of the target next moment (posterior estimation error covariance of the current moment) Tracking accuracy, select the task node at the next moment, and at the same time, in order to reduce the cost of calculation and communication, perform dimensionality reduction processing to obtain a priori estimation error covariance matrix trace, and indirectly judge the target through the priori estimation error covariance matrix trace Predicted tracking accuracy.
[0071] The following is a detailed description of the distributed processing-based wireless sensor network target tracking method of the present invention: assuming that the target is moving in a two-dimensional plane, the selected state variable is X(k)=(x(k), x v (k), y(k), y v (k)) T , Which means that the target occurs at t in the Kth sampling period k The state at the moment, where x(k) and y(k) are the position coordinates along the X and Y axes respectively, and x v (k), y v (k) is the velocity value along the X and Y axes, assuming the target's motion model: as follows:
[0072] X(k+1)=F(Δt k )X(k)+w(k,Δt k ) (3)
[0073] F ( Δt k ) = 1 Δt k 0 0 0 1 0 0 0 0 1 Δt k 0 0 0 1
[0074] Δt k = T k+1 -t k Represents the Kth sampling interval time, F(Δt k ) Is the transition matrix, by Δt k Decide, w(k,Δt k ) Is the process noise and also depends on Δt k. Suppose task node i is at t k Time is used to obtain the Kth measurement value Z i (k), the measurement model is given as follows:
[0075] Z i (k)=h i (X(k))+v i (k) (4)
[0076] among them h i ( X ( k ) ) = ( x ( k + 1 | k ) - x i ( k ) ) 2 + ( y ( k + 1 | k ) - y i ( k ) ) 2 , Is a non-linear measurement function, (x i (k), y i (k)) is the known position coordinate of task node i in the k-th sampling period. (x(k+1|k), y(k+1|k)) are the estimated position coordinates of the target. vi (k) is the measurement noise of task node i, w(k, Δt k ) And v i (k) are all independent, and are assumed to have a mean value of 0, white noise in line with Gaussian normal distribution, P(k|k) and v i The covariance matrix of (k) is Q(Δt k ) And R i (k).
[0077] Set a person as the target to move in a 240cm×240cm square area, and use ultrasonic to measure the distance, so that the target itself does not need to be equipped with sensor nodes. Passive infrared sensors perform target detection and wake up sleeping nodes. Considering that the target relative monitoring area size cannot be ignored, in order to obtain the target centroid The coordinate value of, needs to be compensated for the distance value obtained during distance measurement, here is 10cm. Target starting position such as Figure 5 Shown, namely (41, 38) coordinate point. Such as image 3 As shown, a i-1j , A i+1j , A ij-1 , A ij+1 Is node a ij Neighbor nodes are determined according to the following rules:
[0078] N ( a ij ) ∈ { a k , l | r r s 2 r , k ≠ i , l ≠ j } - - - ( 5 )
[0079] The detailed process of the extended Kalman filtering algorithm is as follows:
[0080] Initialization of the target state, assuming that the target is at t k The initial state X(k) at time is estimated as , The corresponding error covariance matrix is P(k|k). Suppose sensor j is at t k+1 As the task node for ranging, the sensor j is at t k+1 Estimated state at time It can be calculated by the following formula
[0081] X ^ ( k + 1 | k ) = F ( Δt k ) X ^ ( k | k ) - - - ( 6 )
[0082] Find the one-step prediction of the state estimation error covariance matrix:
[0083] P(k+1|k)=F(Δt k )P(k|k)F(Δt K ) T +Q(Δt K ) (7)
[0084] Q ( Δt K ) = q 1 3 Δt k 3 1 2 Δt k 2 0 0 1 2 Δt k 2 Δt 0 0 0 0 1 3 Δt k 3 1 2 Δt k 2 0 0 1 2 Δt k 2 Δt
[0085] q is a scalar, which determines the intensity of the process noise, and the value is 50 here.
[0086] One-step predicted value of the measurement available from the observation equation
[0087] Z ^ j ( k + 1 | k ) = h j ( X ^ ( k + 1 | k ) ) - - - ( 8 )
[0088] The residual is the ultrasonic measurement value of sensor node j and sensor node j at t k+1 Time estimate The difference between is given by:
[0089] γ j ( k + 1 ) = Z j ( k + 1 ) - Z ^ j ( k + 1 | k ) - - - ( 9 )
[0090] Measurement error covariance matrix S j (k+1) is given by
[0091] S j ( k + 1 ) = H j ( k + 1 ) P ( k + 1 | k ) H j T ( k + 1 ) + R j ( k + 1 ) - - - ( 10 )
[0092] Where H j (k+1) is the measurement function h j At t k+1 Time corresponds to the estimated state Jacobian matrix
[0093] H j ( k + 1 ) = - ( x ( k + 1 | k ) - x j ) [ ( x ( k + 1 | k ) - x j ) 2 + ( y ( k + 1 | k ) - y j ) 2 ] 3 2 0 - ( y ( k + 1 | k ) - y j ) [ ( x ( k + 1 | k ) - x j ) 2 + ( y ( k + 1 | k ) - y j ) 2 ] 3 2 0
[0094] The Kalman gain matrix is
[0095] K ( k + 1 ) = P ( k + 1 | k ) H j ( k + 1 ) T S j - 1 ( k + 1 ) - - - ( 11 )
[0096] Update of status and covariance matrix:
[0097] X ^ ( k + 1 | k + 1 ) = X ^ ( k + 1 | k ) + K ( k + 1 ) γ j ( k + 1 ) - - - ( 12 )
[0098] P(k+1|k+1)=P(k+1|k)-K(k+1)S j (k+1)K T (k+1) (13)
[0099] Further, according to the position estimation information in the target state estimation information at the current moment, the prior estimation error covariance matrix trace of the candidate task node at the next moment is calculated. The specific process is as follows:
[0100] Since it is the prior estimation error covariance matrix of the candidate task node at the next moment, no measurement data is required, and the estimated state at the next moment is obtained according to the coordinates of different candidate task nodes The corresponding Jacobian matrix H i (k+2), execute (6)(7)(10)(11)(13) in the extended Kalman filter algorithm to obtain the prior estimation error covariance matrix P of the candidate task node at the next moment i (k+2|k+2) and its corresponding trace is
[0101] Φ i ( k + 2 ) = σ x 2 + σ x v 2 + σ y 2 + σ y v 2 = 1 1 1 1 P i ( k + 2 | k + 2 ) 1 1 1 1 T - - - ( 14 )
[0102] Select the node corresponding to the smallest value as the next task node.
[0103] When the state a priori estimation error covariance matrix trace corresponding to the task node at the next moment is greater than the threshold, the target trajectory correction algorithm is started. The detailed process is as follows:
[0104] Wake up the candidate task nodes of the task node at the current moment, and pass the target of the task node at the previous moment to these nodes. Within the maximum activity range of the target unit time, the unit time is evenly distributed to these nodes, and the targets are respectively ranged to obtain Observe the data, and execute (6)(7)(8)(9)(10)(11)(12) in the extended Kalman filter algorithm to obtain the respective target status information (including position information and speed information), and transmit For the task node at the current time, the task node at the current time uses the position information and speed information in the target status information to average, obtain the status information of the target at the current time, and realize the correction of the target.
[0105] When the next task node crashes due to insufficient power supply or other reasons, start the alternate task node startup program. The detailed process is as follows:
[0106] Using the task node information at the previous time backed up in the base station, within the range of its neighbor nodes, execute (6)(7)(10)(11)(13) in the extended Kalman filter algorithm to obtain the next time The prior estimation error covariance matrix P of neighbor nodes and its corresponding trace are Φ i , Select the node corresponding to the next smallest value as the task node at the next moment.
[0107] When a single node repeatedly acts as a task node, in order to prevent the accumulation of target deviation trajectory errors, the number of times that a single node can continuously act as a task node is limited, and the threshold of the number of times is set to 3. If a single node acts as a task node for more than 3 times, then The candidate task node with the second smallest target state error covariance matrix trace is selected as the task node at the next moment.
[0108] Tracking results such as Figure 5 As shown, the tracking process is shown in the areas 1, 2, 3, and 4 in the figure. The big black circle in the figure represents the moving target, and the numbers in parentheses around it represent the position information of the moving target at this time, that is, the coordinate value; The circle represents the task node, the small white circle represents the non-task node, and its internal number represents the node ID, and the numbers in parentheses around it represent the position information of the node, that is, the coordinate value; the line formed by the black solid dots represents the historical trajectory of the moving target ; A circle with the task node as the center and the distance between the task node and the moving target as the radius represents that the task node is measuring its own distance value relative to the moving target at this time.
[0109] The present invention uses a distributed extended Kalman filter algorithm for target tracking in a wireless sensor network. Each operation only needs the result of the last operation, adopts the principle of minimum state error covariance matrix trace to select task nodes, and uses state error covariance The matrix trace is used to predict the deviation of the target trajectory, which saves memory overhead, reduces processor energy consumption, reduces data transmission and information interaction between nodes, and improves tracking accuracy. When the task node crashes, use the target candidate task node startup program to re-elect the task node, which improves the robustness of the system
[0110] Through the above description of the specific embodiments of the present invention in conjunction with the accompanying drawings, other aspects and features of the present invention will be apparent to those skilled in the art.
[0111] The specific embodiments of the present invention have been described and illustrated above. These embodiments should be regarded as only exemplary and not used to limit the present invention. The present invention should be interpreted according to the appended claims.
