阅读说明：本技术 一种快速被动传感器网络多目标定位方法 (Rapid passive sensor network multi-target positioning method ) 是由 兰剑 刘爽 安翔 吴生盛 张乐 于 2020-11-10 设计创作，主要内容包括：本发明公开了一种快速被动传感器网络多目标定位方法,步骤为：一：利用探测范围以及相对位置判定不同传感器间的关联可能性,以此对传感器进行分类；二：根据当前时刻各传感器是否观测到目标更新传感器分类,并对量测进行编号；三：初始化量测关联假设,给出所有可能的关联结果；四：根据设定的规则进行假设删减及代价计算；五：建立多维分配模型；六：求解多维分配问题并完成多目标精确定位。本发明能够应用于园区、仓储、博物馆、小区以及工厂等可以使用被动传感器网络作为安防系统的区域,解决现有数据关联算法无法应用或无法很好的应用于多被动传感器观测下的多目标定位,造成计算复杂度高而精度低的问题。(The invention discloses a rapid passive sensor network multi-target positioning method, which comprises the following steps: firstly, the method comprises the following steps: determining the association possibility among different sensors by using the detection range and the relative position so as to classify the sensors; II, secondly: updating sensor classification according to whether each sensor observes a target at the current moment, and numbering measurement; thirdly, the method comprises the following steps: initializing a measurement association hypothesis and giving all possible association results; fourthly, the method comprises the following steps: performing hypothesis deletion and cost calculation according to a set rule; fifthly: establishing a multi-dimensional distribution model; sixthly, the method comprises the following steps: and solving the multi-dimensional distribution problem and completing multi-target accurate positioning. The method can be applied to areas of parks, warehouses, museums, communities, factories and the like, and can use the passive sensor network as a security system, and the problems of high calculation complexity and low precision caused by the fact that the conventional data association algorithm cannot be applied or cannot be well applied to multi-target positioning under observation of multiple passive sensors are solved.)

1. A multi-target positioning method for a rapid passive sensor network is characterized by comprising the following steps:

the method comprises the following steps: calculating the distance between the sensors, judging whether a common observation region exists between the sensors according to the mutual distance and the detection range, and classifying the sensors under the condition that no common observation region exists between every two sensors in each class;

step two: at each sampling moment, updating the classification result by deleting the sensors which do not observe the target in each class, and numbering the measurement of the sensors in each class;

step three: initializing possible measurement association combinations in the passive sensor network according to the classification result of the sensors;

step four: calculating target positions and costs corresponding to all measurement combinations meeting the rule requirements according to a set rule;

step five: constructing a multi-dimensional distribution model according to the principle that one target is observed by the same sensor at most once at the same moment, and one measurement can only come from one target or clutter;

step six: and solving the multi-dimensional distribution problem by using a linear programming method to obtain a position set of multi-target positioning and realize the multi-target positioning.

2. The method of claim 1, wherein the first step comprises the steps of:

11) recording a passive sensor set, calculating the distance between sensors according to the positions of the sensors in the sensor network, judging whether a common detection area exists between any two sensors according to the detection range limitation, and obtaining a judgment matrix D_s×s；

12) Arbitrarily selecting one sensor from a passive sensor setFrom the discriminant matrix D_s×sDetermining remaining sensors and selected sensorsIf there is a common detection area between them, it will not be possible to detect the sensor with the selected sensorAlpha with common detection zone₁Individual sensor and selected sensorClassifying the sensors into one type, then sequentially judging whether a common observation area exists between any two sensors, if so, rejecting the sensors until a common detection area does not exist between any two sensors in the passive sensor set;

13) repeat step 12) for the remaining sensors until all sensors have been classified.

3. The method according to claim 1, wherein the second step comprises the steps of:

21) at the moment t, deleting the sensors which are not measured after the classification in the step one to obtain a new classification result;

22) numbering the measurements of the sensors in the same class in sequence; each time measurements from a sensor are numbered, a virtual measurement is added before the first measurement, which when representing a measurement combination association assumption, indicates that the target should not be observed by the sensor to which it belongs.

4. The method of claim 1, wherein step three comprises the steps of:

31) based on the sensor classification results, the possible measurement association combination hypotheses in the passive sensor network can be expressed asWherein n is_iIndicates the i-th sensor A_iThe number of measurement numbers containing virtual measurement, N is the number of sensor classification,to representOne of the measurement assemblies is selected from the group consisting of,i at position s in the subscript_sRepresenting the ith from class s sensors_sMeasuring, according to step two, the measurement may be a real measurement of a sensor in class s, or a virtual measurement;

32) initializing an N-dimensional tensorIs 0, the tensor is used to record whether the measurement combination is wrong or not and whether the measurement combination is calculated or not; when C (i)₁,i₂,...,i_N) When 1, the measurement combination is expressedIt is likely correct, and the corresponding target location and cost have been calculated; when C (i)₁,i₂,...,i_N) When it is-1, the measured combination is expressedIncorrect, that is, a set of measurement association combinations with errors; when C (i)₁,i₂,...,i_N) When 0, the measurement combination is indicatedCorresponding target location and cost are not countedCalculating;

33) according to the classification result of the sensors, the correlation result of the data in the passive sensor network is expressed asThe number of elements in the set represents the number of possible targets obtained by the metrology correlation.

5. The method of claim 1, wherein said step four comprises the steps of:

selecting m classes from the class B of sensors with measurements, m 1,2Calculating the target position x corresponding to the measurement combination_pAnd positioning errorTest statistic χ using chi-square distribution_αDetermining whether the measurement combination is possible to be established, and calculating the combination cost

6. The method of claim 5, wherein the specific process of determining whether the metrology assemblies can be associated in the fourth step is:

41) when m is 1, the set of measurement combinations is

This indicates a high probability of false alarm, but does not exclude the possibility of being a target, where the cost of any measurement combination in the set isWherein Z is_m＝1Indicating the set of measurement combinations when only one sensor class is selected,indicates that the s-th class is selected the i-th_sSelecting measurement combinations of virtual measurement for the other types;

42) when m is 2, the set of measurement combinations is

Wherein Z is_m＝2Representing the set of measurement combinations when two sensor classes are selected,indicates that the s-th class is selected the i-th_sMeasuring, selecting ith from tth class_tSelecting measurement combinations of virtual measurement for the other types;

the positioning result corresponding to any measurement combination in the set is the intersection point of the straight lines of the directions represented by the two measurements, and when the two angle rays cannot intersect, the cost is

When two angular rays should intersect, the cost is

Wherein u (i)_s) Representing measurement combinationsWhether or not to useTrue measurements in class s, P_DIn order to detect the probability of the occurrence,is the detection range of the sensor or sensors,is the measurement noise variance of the sensor, c_2nRepresenting a wrong associated cost, c_2yRepresents the associated cost that is likely to be correct;

if the cost is c_2nThen, the measurement combination is deleted and the current price is c_2yThen, chi-square test is carried out on the positioning result according to the detection distance limit of the sensor, the impossible measurement combination is further deleted, and the measurement combination possibility matrix C (i) is updated at the same time₁,i₂,...,i_N)；

43) When m is greater than or equal to 3, the set of measurement combinations is

Wherein Z is_m＝kThe measurement combination set when k sensor classes are selected is shown, k is larger than or equal to 3, and association assumptions of the unselected sensor classes are virtual measurement.

7. The method of claim 6, wherein the card side verification step in step 42) is as follows:

according to the measurement combinationThe included measurement, the position information of the sensor to which each measurement belongs, and the positioning result of the measurement combination construct chi-square test statistic:

wherein s is_iFor each measurement in the measurement set, the corresponding sensor number i is 1,2_Z，N_ZFor measuring the combinationNumber of true measurements in, Loc(s)_i1) and Loc(s)_iAnd 2) respectively represent the numbers s_iR is N_ZDimension column vector, s_iDimension is s_iDetection distance of individual sensor, H_rIs a Jacobian matrix, P_pIn order to be the covariance of the positioning error,for the distance vector between the target positioning result and each sensor,a target positioning result is obtained;

if it is notIf so, then a correlation hypothesis is considered possible, whereinRepresenting a degree of freedom M ═ N_ZThe confidence is the chi-square distribution quantile of 1-alpha.

8. Method according to claim 6, characterized in that the calculation of the position in step 43) is as follows:

for the measurement combinationFirst, two measurements are selected and the intersection is calculated as follows:

judging whether the intersection exists or not according to the cost, and reselecting two measurements if the intersection does not exist until the intersection exists; then, the following process is repeated until | diff | cells are not covered₂＜δ_diffWherein δ_diffIs a preset threshold value:

where Loc is a matrix formed by the coordinates of the sensors, and Loc (num,1) and Loc (num,2) represent numbers, respectivelyThe abscissa and ordinate of the sensor for num,andrespectively representing two measurements, i, selected at the beginning_s1And i_s2Representing the measurement number, J is the Jacobian matrix calculated in the iterative process, and the dimension n thereof_ZIs composed ofThe number of the actual measurements in the measurement,it is a one-step prediction of measurement, where R is the measurement noise covariance of the passive sensor and diff is the gain.

9. The method according to claim 6, wherein the cost in step 43) is calculated as follows:

wherein the content of the first and second substances,is class i_sThe measurement is carried out by measuring the number of the measurement,for the purpose of the measurement function,is class s i_sThe coordinates of the sensor to which the measurement belongs are measured.

10. The method according to claim 1, wherein step five is as follows:

51) constructing a binary variable according to the principle that one measurement at most comes from one target or false alarm and one target can be observed by the same sensor at most once at the same timeIndicating that no measurement combination is included in the correlation resultWhileRepresenting combinations of assay measurements in correlation results

The expression for the multidimensional assignment problem is derived as follows:

52) further expressing the multi-dimensional distribution model as solving a linear programming problem and constructing a constraint matrix LP_matrixIt is updated by row and column to update the constraint matrix LP_matrixAfter each column of the LP, the row in which the virtual measurement is located is deleted, and the LP is updated_matrixFor each column of (1), the synchronized pair coefficient vector C_LPIs assigned as a measurement groupCorresponding score valueThe linear programming solution model of the multidimensional assignment problem is thus represented as follows:

min C_LPx

wherein x is a solution vector representing the measured correlation determination result, and b is a unit vector.

Technical Field

The invention relates to a passive sensor network target positioning method, in particular to a rapid passive sensor network multi-target positioning method.

Background

Compared with an active sensor, the passive sensor has better concealment and lower manufacturing cost, can cover the same area in a large range and carry out networking detection, and accurately acquires target information through multiple sensors, so that the advantages of the passive sensor based network has great research value in multi-target positioning.

In areas such as parks, warehouses, museums, communities, factories and the like, passive sensor networks are used as areas of security systems, multi-target passive positioning is a big problem in the field of information fusion, data association is the most important and challenging problem of multi-target passive positioning, and the accuracy of any target positioning algorithm is premised on correct data association. The passive sensor can only obtain the direction of the target, the non-linearity of angle measurement is strong, and the included position information is incomplete, so that data association becomes very difficult. Data association in passive sensor networks also faces the following problems: the number of targets is uncertain, missing detection and false alarm exist in sensor detection, the detection range of the array is limited, and one node may have different common detection areas with a plurality of nodes.

At present, a plurality of data association algorithms exist, and widely applied data association technologies include a nearest neighbor algorithm (NN), a Probability Data Association (PDA), a Joint Probability Data Association (JPDA), multi-hypothesis tracking (MHT) and the like, and the algorithms are not suitable for the multi-target positioning problem under the condition of passive sensor network monitoring with limited detection distance due to respective limitations. When the clutter is more, the nearest neighbor algorithm is easy to associate errors; the problem that the number of targets is unknown or the target distribution is dense is difficult to process by a probability data association algorithm; the joint probability data association algorithm needs to know the number of targets, and the calculated amount of the joint probability data association algorithm increases exponentially along with the increase of the number of the targets; multiple hypothesis tracking can achieve fast and accurate data association, but in a passive sensor network, multiple measurements are performed on the same target at the same time, which is inconsistent with the basic hypothesis.

The invention solves the problem of multi-target positioning of a passive sensor network under the framework of a multi-dimensional (S-D) distribution technology. Because the detection range of the sensors in the passive sensor network is limited, the network topology relationship is complex, and a plurality of problems can be caused by directly using the traditional multidimensional distribution algorithm: the number of the sensors is large, so that the number of the associated hypotheses is increased, and the solving is slow due to overhigh model dimension of the corresponding optimization problem; the Lagrange relaxation algorithm can only obtain suboptimal solutions, errors are accumulated when the dimensionality of the model is too high, and correct correlation results are difficult to obtain. Therefore, further optimization and improvement according to the specificity of the passive sensor network are needed in practical use.

Disclosure of Invention

The invention aims to solve the problems that the existing data association technology cannot be used in a passive sensor network or has high complexity and the like, and provides a novel rapid passive sensor network multi-target positioning method. For the application scene of the invention, passive sensors with limited detection distance are uniformly distributed in a detection area and form a sensor network to position multiple targets in the area.

The invention is realized by the following technical scheme.

The invention provides a rapid passive sensor network multi-target positioning method, which comprises the following steps:

the method comprises the following steps: calculating the distance between the sensors, judging whether a common observation region exists between the sensors according to the mutual distance and the detection range, and classifying the sensors under the condition that no common observation region exists between every two sensors in each class;

step two: at each sampling moment, updating the classification result by deleting the sensors which do not observe the target in each class, and numbering the measurement of the sensors in each class;

step three: initializing possible measurement association combinations in the passive sensor network according to the classification result of the sensors;

step four: calculating target positions and costs corresponding to all measurement combinations meeting the rule requirements according to a set rule;

step five: constructing a multi-dimensional distribution model according to the principle that one target is observed by the same sensor at most once at the same moment, and one measurement can only come from one target or clutter;

step six: and solving the multi-dimensional distribution problem by using a linear programming method to obtain a position set of multi-target positioning and realize the multi-target positioning.

With respect to the above technical solutions, the present invention has a further preferable solution:

preferably, the first step comprises the following steps:

11) recording a passive sensor set, calculating the distance between sensors according to the positions of the sensors in the sensor network, judging whether a common detection area exists between any two sensors according to the detection range limitation, and obtaining a judgment matrix D_S×S；

12) Arbitrarily selecting one sensor from a passive sensor setFrom the discriminant matrix D_S×SDetermining remaining sensors and selected sensorsIf there is a common detection area between them, it will not be possible to detect the sensor with the selected sensorAlpha with common detection zone₁Individual sensor and selected sensorClassifying the sensors into one type, then sequentially judging whether a common observation area exists between any two sensors, if so, rejecting the sensors until a common detection area does not exist between any two sensors in the passive sensor set;

13) repeat step 12) for the remaining sensors until all sensors have been classified.

Preferably, the second step comprises the following steps:

21) at the moment t, deleting the sensors which are not measured in the step A to obtain a classification result;

22) numbering the measurements of the sensors in the same class in sequence; each time measurements from a sensor are numbered, a virtual measurement is added before the first measurement, which when representing a measurement combination association assumption, indicates that the target should not be observed by the sensor to which it belongs.

Preferably, the third step includes the following steps:

31) based on the sensor classification results, the possible measurement association combination hypotheses in the passive sensor network can be expressed asWhereinIt is shown that a measurement combination is represented,i at position s in the subscript_sRepresenting the ith from class s sensors_sMeasuring, according to step two, the measurement may be a real measurement of a sensor in class s, or a virtual measurement;

32) initializing an N-dimensional tensorEach element of (1) is 0 and each dimension is n_iWherein n is_iIs A_iThe measurement number of (1) comprises virtual measurement, and the tensor is used for recording whether a measurement combination is wrong or not and whether the measurement combination is calculated or not; when C (i)₁,i₂,...,i_N) When 1, the measurement combination is expressedIt is likely correct, and the corresponding target location and cost have been calculated; when C (i)₁,i₂,...,i_N) When it is-1, the measured combination is expressedIs incorrect, i.e. a groupError measurement association combination; when C (i)₁,i₂,...,i_N) When 0, the measurement combination is indicatedThe corresponding target position and cost are not calculated;

33) according to the classification result of the sensors, the correlation result of the data in the passive sensor network is expressed asThe number of elements in the set represents the number of possible targets obtained by the metrology correlation.

Preferably, the fourth step includes the following steps:

selecting m classes from the classes of sensors with measurements, m 1,2Calculating the target position x corresponding to the measurement combination_pAnd positioning errorTest statistic χ using chi-square distribution_αDetermining whether the measurement combination is possible to be established, and calculating the combination cost

Further, the specific process of determining whether the measurement combinations can be associated in the fourth step is as follows:

41) from the set of measurement combinations when m is 1, it is determined that there is a high probability of being a false alarm, but the probability of being a target cannot be excluded;

42) according to the set of the measurement combinations when m is 2, the positioning result corresponding to any measurement combination in the set is the intersection point of the straight lines of the directions represented by the two measurements, and the cost when the two angle rays cannot intersect and should intersect is obtained;

if the cost is c_2nThen, the measurement combination is deleted and the current price is c_2yThen, chi-square test is carried out on the positioning result according to the detection distance limit of the sensor, the impossible measurement combination is further deleted, and the measurement combination possibility matrix C (i) is updated at the same time₁,i₂,...,i_N)；

43) Obtaining the set of the measurement combination when m is more than or equal to 3.

Preferably, the card side checking step in step 42) is as follows:

according to the measurement combinationThe contained real measurement, the position information of the sensor to which each measurement belongs, and the positioning result of the measurement combination construct chi-square test statistics to obtain whether the associated hypothesis is possible.

Preferably, the step five is as follows:

51) constructing a binary variable according to the principle that one measurement at most comes from one target or false alarm and one target can be observed by the same sensor at most once at the same time Indicating that the correlation result is not a correct measurement combinationIndicating that the correlation result is a correct measurement combination; further obtaining an expression of the multidimensional distribution problem;

52) further expressing the multi-dimensional distribution model as solving a linear programming problem and constructing a constraint matrix LP_matrixUpdating the rows and columns, deleting the row where the virtual measurement is located after updating each column of the constraint matrix, and synchronously updating the coefficient vector C when updating each column of the constraint matrix_LPThe ith row of (2) is assigned as the score corresponding to the measurement combination, thereby obtaining the multi-dimensional distribution questionThe linear programming of the problem solves the model.

Due to the adoption of the technical scheme, the invention has the following beneficial effects:

the invention provides a novel rapid passive sensor network multi-target positioning method aiming at the problem of multi-target positioning in a sensor network. The technology classifies a plurality of nodes in a sensor network offline according to the topological relation on the position of the sensor and the detection distance of a passive sensor, and updates the classification of the sensor online at each moment according to whether the sensor has measurement. And then calculating possible measurement combinations among the sensors, generating a complete association hypothesis by using a chi-square test to delete the impossible combination mode, and calculating corresponding cost according to the measurement related in the hypothesis. The method comprises the steps of constructing and solving a multi-dimensional (S-D) distribution problem of multi-target positioning for a target by taking the principle that one target is observed by one sensor at most at one time at the same moment, one measurement can only come from one target or clutter as constraint and the whole cost is minimum, rewriting the multi-dimensional distribution problem into a linear programming problem, solving by using a linear programming method to obtain an optimal measurement combination corresponding to the multi-target position at the current moment, and completing the positioning of the multiple targets, wherein the process of calculating the target position by using the measurement combination utilizes an iterative least square method.

Compared with the prior art, the method has the advantages that the sensors are classified by utilizing the characteristics of limited topological relation and detection distance among the sensors, the problem is simplified, the combination measured by each sensor is judged by utilizing chi-square test subsequently, the impossible association is eliminated, the time complexity is further reduced, the requirement of an algorithm on observability is relaxed by adopting least square rather than maximum likelihood estimation in the target positioning process, then the multi-dimensional distribution problem is solved by utilizing a linear programming method instead of a Lagrange relaxation method, and the rapid and effective multi-target positioning in a passive sensor network is realized.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention:

FIG. 1 shows a multi-target location process for a passive sensor network;

FIG. 2 shows the flow of step four in multi-target positioning of a passive sensor network;

FIG. 3 shows the positioning of ten targets at 70 times by 135 sensors with a detection range of 100 meters;

FIG. 4 shows the RMSE of 135 sensors with a detection range of 100 meters for the results of ten target locations over 70 time instants.

Detailed Description

The present invention will now be described in detail with reference to the drawings and specific embodiments, wherein the exemplary embodiments and descriptions of the present invention are provided to explain the present invention without limiting the invention thereto.

As shown in fig. 1, an embodiment of the present invention provides a method for fast and passively positioning multiple targets in a sensor network, including the following steps:

the method comprises the following steps: and calculating the distance between the sensors, judging whether a common observation region exists between the sensors according to the mutual distance and the detection range, and classifying the sensors under the condition that no common observation region exists between every two sensors in each class.

The method specifically comprises the following steps:

11) recording the set of passive sensors as Sen ═ Sen₁,Sen₂,...,Sen_SCalculating the distance between each sensor according to the sensor position in the sensor network; judging whether a common detection area exists between any two sensors according to the limitation of the detection range to obtain a discrimination matrix D_S×S，D_S×SElement d of (1)_ijThe association possibility between the ith sensor and the jth sensor is represented, the association possibility indicates that the two sensors have a common detection area when the association possibility is 1, and the association possibility indicates that the two sensors do not have the common detection area when the association possibility is 0;

12) arbitrarily selecting one sensorS is the total number of sensors and is determined by a decision matrix D_S×SDetermining remaining sensors and selectionsSensor with a sensor elementWhether a common detection area exists between them; will not be possible with the selected sensorAlpha with common detection zone₁A sensor andscratch a same kind, and be denoted asThen judging the sensor A in turn_1iAnd A_1j(i < j) in which i is 2,3₁-1，j＝i+1,i+2,...,α₁If present, will A_1jRemoving A₁Up to A₁There is no common detection zone between any two sensors;

13) divide A from the set of passive sensors Sen₁One of the other sensors is selected arbitrarilyThe operation of step 12) above is repeated for the remaining sensors until all sensors have been classified. The total number of sensor classes is N, and the set of each class is A ═ A₁,A₂,...,A_NRecording the sensor classes as A in sequence according to the forming sequence_i，i＝1,2,...,N。

Step two: at each sampling instant, the classification result is updated by deleting sensors in each class that did not observe the target, and the measurements of the sensors in each class are numbered.

The method specifically comprises the following steps:

21) and at the time t, deleting the sensors which are not measured in the step A, and obtaining a classification result B ═ B₁,B₂,...,B_N}；

22) Of sensors of the same classThe measurement is numbered according to the sequence; considering the existence of missed measurements, each time measurements from one sensor are numbered, a virtual measurement is added before the first measurement, and the virtual measurements indicate that the target should not be observed by the sensor to which the measurement belongs when the measurement combination association assumption is expressed. For example A₁In which there are two sensors A₁₁And A₁₂Wherein A is₁₁There are 5 measurements, A₁₂There are 6 measurements, then A₁₁Is numbered 2 to 6, A₁₂The measurements in (1) are numbered 8 to 13, and the measurements numbered 1 and 7 are virtual measurements.

Step three: and initializing possible measurement association combinations in the passive sensor network according to the sensor classification result.

The method specifically comprises the following steps:

31) based on the sensor classification results, the possible measurement association combination hypotheses in the passive sensor network can be expressed asWhereinIt is shown that a measurement combination is represented,i at position s in the subscript_sRepresenting the ith from class s sensors_sA measurement, according to step two, which may be the actual measurement of a sensor in class s (the measurement is numbered i in class s)_s) Or virtual metrology;

32) initializing an N-dimensional tensorEach element of (1) is 0 and each dimension is n_iWherein n is_iIs A_iThe tensor is used to record whether the metrology assembly is erroneous and is computed. When C (i)₁,i₂,...,i_N) When 1, the measurement combination is expressedIt is likely correct, and the corresponding "target" location and cost have been calculated; when C (i)₁,i₂,...,i_N) When it is-1, the measured combination is expressedIncorrect, that is, a set of measurement association combinations with errors; when C (i)₁,i₂,...,i_N) When 0, the measurement combination is indicatedThe corresponding "target" position and cost are not calculated;

33) according to the classification result of the sensor, the correlation result of the data in the passive sensor network can be expressed asThe number of elements in the set represents the number of possible targets obtained by the metrology correlation.

Step four: and calculating target positions and costs corresponding to all measurement combinations meeting the rule requirements according to the set rule. The specific flow is shown in FIG. 2.

Selecting m classes from the class B of sensors with measurements, m 1,2Calculating the corresponding 'target' position x of the measurement combination_pAnd positioning errorTest statistic χ using chi-square distribution_αDetermining whether the measurement combination is possible to be established, and calculating the combination cost

The specific process of determining whether the measurement combination can be associated is as follows:

41) when m is 1, the set of measurement combinations is

Wherein Z is_m＝1Indicating the set of measurement combinations when only one sensor class is selected,indicates that the s-th class is selected the i-th_sSelecting measurement combinations of virtual measurement for the other types;

this indicates a high probability of false alarm, but does not exclude the possibility of being a target, where the cost of any measurement combination in the set is

42) When m is 2, the set of measurement combinations is

Wherein Z is_m＝2Representing the set of measurement combinations when two sensor classes are selected,indicates that the s-th class is selected the i-th_sMeasuring, selecting ith from tth class_tSelecting measurement combinations of virtual measurement for the other types;

the positioning result corresponding to any measurement combination in the set is the intersection point of the straight lines of the directions represented by the two measurements, and when the two angle rays cannot intersect, the cost is

When two angular rays should intersect, the cost is

Wherein u (i)_s) Representing measurement combinationsWhether or not the true measure in class s, P, is used_DIn order to detect the probability of the occurrence,is the detection range of the sensor or sensors,is the measurement noise variance of the sensor, c_2nRepresenting a wrong associated cost, c_2yRepresents the associated cost that is likely to be correct; if the cost is c_2nThen, the measurement combination is deleted and the current price is c_2yThen, chi-square test is carried out on the positioning result according to the detection distance limit of the sensor, the impossible measurement combination is further deleted, and the measurement combination possibility matrix C (i) is updated at the same time₁,i₂,...,i_N)。

The chi fang test comprises the following steps:

according to the measurement combinationThe included measurement, the position information of the sensor to which each measurement belongs, and the positioning result of the measurement combination construct chi-square test statistic:

wherein s is_iFor each measurement in the measurement set, the corresponding sensor number i is 1,2_Z，N_ZFor measuring the combinationNumber of true measurements in, Loc(s)_i1) and Loc(s)_iAnd 2) respectively represent the numbers s_iR is N_ZDimension column vector, s_iDimension is s_iDetection distance of individual sensor, H_rIs a Jacobian matrix, P_pIn order to be the covariance of the positioning error,the distance between the target location result and the sensor,and the target positioning result is obtained.

If it is notIf so, the association hypothesis is considered possible. WhereinRepresenting a degree of freedom M ═ N_ZThe confidence is the chi-square distribution quantile of 1-alpha.

43) When m is greater than or equal to 3, the set of measurement combinations is

Wherein Z is_m＝kAnd the measurement combination set represents the measurement combination set when k sensor classes are selected, k is more than or equal to 3, and the correlation results of the unselected sensor classes are virtual measurement.

In order to simplify the calculation, the invention adopts a recursion mode to judge whether the measurement combination possibly exists: when m is k, the current measurement combinationOptionally selecting k-1 real measurements, combining them with virtual measurements, and if all such combinations are determined to be associable when m is k-1, then considering that the combinations are associableAre possible correlation hypotheses and then the "target" position is calculated using an iterative least squares method and the correlation cost is calculated. The least square estimation is adopted instead of the commonly used maximum likelihood estimation in the multidimensional allocation, because in the multi-passive sensor network, due to the limitation of the detection distance and the constraint of the network arrangement mode, the observability difference of the target at different positions is large. Under the assumption that the measured noise obeys Gaussian distribution, the least square estimation and the maximum likelihood estimation have equivalent target cost functions, and for the nonlinear problem, the performances of the least square estimation and the maximum likelihood estimation are basically consistent under the condition of strong observability, however, under the condition of weak observability, the numerical calculation of the Hessian matrix cannot ensure the correctness in the process of solving the maximum likelihood estimation, so that the maximum likelihood estimation algorithm has no good performance of the least square algorithm.

The position calculation steps are as follows:

for the measurement combinationFirst, two measurements are selected and the intersection is calculated as follows:

and judging whether the intersection exists or not according to the cost, and reselecting the two measurements if the intersection does not exist until the intersection exists. Then, the following process is repeated until | diff | cells are not covered₂＜δ_diffWherein δ_diffIs a preset threshold value:

where Loc is a matrix formed by the coordinates of the sensors, Loc (num,1) and Loc (num,2) respectively represent the horizontal and vertical coordinates of the sensor numbered num,andrespectively representing two measurements, i, selected at the beginning_s1And i_s2Representing the measurement number, J is the Jacobian matrix calculated in the iterative process, and the dimension n thereof_ZIs composed ofThe number of the actual measurements in the measurement,is a one-step prediction of the measurement, R isThe sensor measures the noise covariance, diff is the gain.

Wherein, the cost calculation steps are as follows:

wherein the content of the first and second substances,is class i_sThe measurement is carried out by measuring the number of the measurement,for the purpose of the measurement function,is class s i_sThe coordinates of the sensor to which the measurement belongs are measured.

Step five: and constructing a multi-dimensional distribution model according to the principle that one target is observed by the same sensor at most once at the same time and one measurement can only come from one target or clutter.

The method specifically comprises the following steps:

51) constructing a binary variable according to the principle that one measurement at most comes from one target or false alarm and one target can be observed by the same sensor at most once at the same time Indicating that no measurement combination is included in the correlation resultWhileRepresenting combinations of assay measurements in correlation resultsThe constraint condition based on the corresponding relationship between the measurement and the target is as follows

This formula represents the ith in the s-th class_sOne measurement is associated with only one target, and one target also has only one measurement in the s-th class, where s is 1,2_s＝1,2,...,n_s，n_sIs n of class s_sMeasuring, taking the minimum overall cost of the correlation result as an optimization target, and constructing a cost function

The expression for the multidimensional assignment problem is thus as follows:

52) further expressing the multi-dimensional distribution model as solving a linear programming problem and constructing a constraint matrix LP_matrixThe number of the matrix columns is the number n of the associated hypotheses calculated in the fourth step_hypNumber of rows n₁+n₂+...+n_NThe ith column of the matrix represents the ith hypothesisThe row represents the measurement of the s +1 th class of sensor, n₀If the ith hypothesis is a measurement combinationThen the first of the columnIs 1, and the rest positions are 0, wherein s is 0,1, and N-1, and LP is updated completely_matrixAfter each row of the virtual metrology site is deletedIn the row of, updating LP_matrixFor each column of (1), the synchronized pair coefficient vector C_LPIs assigned asCorresponding score valueAnd the decision variable x of the linear programming is n_hypThe value range of each dimension of the column vector of the dimension is 0-1, and the resource vector b is n₁+n₂+...+n_NAnd each position is 1, so that the linear programming solution model of the multidimensional allocation problem is represented as follows:

min C_LPx

wherein x is a solution vector representing the measured correlation determination result, and b is a unit vector.

Step six: and solving the multi-dimensional distribution problem by using a linear programming method to obtain the optimal result of multi-target measurement association, and performing multi-target positioning.

Solving the optimization problem obtained in the fifth step by utilizing linear programming, wherein the solved result is a data correlation resultObtaining a multi-target positioning position set { x 'according to the measurement combination in the result'_p}。

The invention is further illustrated by the following specific examples.

The method comprises the following steps: calculating the distance between the sensors according to the positions of 135 sensors in the network, judging whether the sensors have a common observation region according to the distance between the sensors and the detection range, and dividing the sensors into 4 classes, which are marked as A ═ A { A ═ A under the condition that the sensors in each class do not have the common observation region between every two sensors₁,A₂,A₃,A₄}；

Step two: at time t, sensors not measured in class A classified in the step are deleted, and a classification result B is obtained as { B }₁,B₂,B₃,B₄Numbering the measurements of the sensors in the same class according to a sequence, and adding a virtual measurement before the first measurement when numbering the measurement of one sensor each time;

step three: possible association of data in a passive sensor network can be expressed as a function of the results of the sensor classificationWherein the elements areIt is shown that a measurement combination is represented,the s-th bit of the subscript of (1) stores i_sDenotes the ith from the s class_sA measurement, according to step two, which may be the actual measurement of a sensor in class s (the measurement is numbered i in class s)_s) Or the virtual measurement added in the step two;

step four: selecting m classes from B, m 1,2, 4, selecting one sensor from each of the classes, combining the measurements of the sensors, combining the measurements for each measurementCalculating the corresponding 'target' position x of the measurement combination_pAnd positioning errorTest statistic χ using chi-square distribution_αDetermining whether the measurement combination can be correlated, and calculating the cost of the measurement combination

Step five: root of herbaceous plantBased on the principle that one target receives at most one measurement from the same sensor and one measurement can only correspond to one target, the method takes the constraint condition as the constraint condition and establishes a binary variable corresponding to each measurement combinationUsing the cost of the measurement combinationAnd a binary variableConstructing a global cost function

Forming an expression of a multi-dimensional distribution problem with the constraint;

step six: solving the optimization problem obtained in the fifth step by utilizing linear programming, wherein the solved result is a data correlation resultObtaining a target position set { x 'by utilizing an iterative least square algorithm according to the measurement combination in the result'_p}. The 135 sensors with a detection distance of 100 meters can locate ten targets in 70 moments as shown in fig. 3.

FIG. 4 shows the RMSE of 135 sensors with a detection range of 100 meters for the results of ten target locations over 70 time instants. As can be seen from the figure, the positioning errors of a plurality of targets are almost always kept below 1m and fluctuate near 0.6m, and the situation that the errors are large only in the case of very few targets (the errors fluctuate to 1.56m only at the 58 th moment) shows that the method has high data association accuracy and positioning precision, and can realize effective positioning of a plurality of targets.

The present invention is not limited to the above-mentioned embodiments, and based on the technical solutions disclosed in the present invention, those skilled in the art can make some substitutions and modifications to some technical features without creative efforts according to the disclosed technical contents, and these substitutions and modifications are all within the protection scope of the present invention.