CN104486833B - The indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction - Google Patents

Abstract

The invention discloses a kind of indoor wireless tomography Enhancement Methods for deleting interfering link based on motion prediction, belong to target detection and tracking technical field in wireless network.The intersection point set being made of shadow fading link is found using cluster, estimation of the center of set as initial time target location, the estimation of current target position is obtained to delete the interfering link of distance objective position farther out using the positioning result of target movement model and last moment by Kalman filter, RTI imagings are carried out finally by shadow fading link, update using Kalman to obtain target current time position on the basis of RTI.Non-shadow is excluded based on motion prediction to decline the influence of link, effectively eliminates false target, and attenuation effect caused by more acurrate protrusion exists due to target, and target positioning is more accurate to making under multi-path environment, improves image quality, realizes more excellent dynamic tracking.

Description

Indoor Wireless Tomography Enhancement Method Based on Motion Prediction and Deletion of Interfering Links

technical field

The invention relates to a positioning method of a wireless tomography system, in particular to a more accurate measurement and positioning model of a wireless tomography system based on motion prediction and deletion of interference links, and belongs to the technical field of target detection and tracking in wireless networks.

Background technique

Wireless tomography (RTI) is a technology that uses wireless nodes to locate and track targets. It images the monitoring area according to the link attenuation caused by target occlusion in the monitoring area, and realizes the positioning and tracking of the target. Wireless tomography does not require the target to carry anything, and it only requires the received signal strength (RSS) value of the wireless node, which most wireless devices can provide today. Therefore, this technology can be easily extended to the existing network without adding any hardware. The technology has a wide range of practical applications, such as: indoor monitoring, fire rescue and so on.

At present, the indoor application of wireless tomography technology is not very mature, and the error is relatively large. Some technologies to improve the performance of wireless tomography in the indoor environment have not considered the problem of link interference. In the indoor environment, not all links All path fading is caused by the line-of-sight (LOS) path being blocked, that is to say, not all fading links are shadow fading links. In fact, the blocking of multipaths with high power or fast fading caused by multipaths can cause link attenuation. If these links are directly used for imaging, large errors will occur in imaging quality and positioning accuracy.

When the target is moving in the monitoring area, for a link between two wireless nodes, the impact of the target on the link can be divided into two cases: the case where the LOS path is blocked and the case where the LOS path is not blocked, As shown in Figure 1 and Figure 2. In Figure 1 and Figure 2, there are four transmission paths between the transmitting node and the receiving node, one of which is a LOS path, and the other three are reflection paths. This is just for the convenience of analysis. In fact, the number of indoor multipaths is much greater than 4. We know that the propagation paths of the three paths in this example are fixed, and the signals of these paths may be blocked by the target at different times, but the propagation paths of these signals are constant, that is, each path is at different times The only difference is whether it is occluded by the target. As shown in Figure 1, the LOS path is blocked, that is, the direct path (LOS path) is blocked. As shown in Figure 2, the LOS path is not blocked, the third path is blocked, and the second path is exactly the same in both cases, that is, in Figure 1 and Figure 2. The fourth path is generated due to the reflection and scattering of the target, and its propagation path is related to the position of the target, so this path is time-varying.

Suppose the signal of the LOS path is The 2nd and 3rd reflection paths are not generated by the target, assuming that the signals of these two paths are γ _p indicates whether the p-th path is blocked. If γ _p =1, it means that the p-th path is not blocked by the target; otherwise, γ _p =0 means that the p-th path is blocked by the target. The fourth path signal is generated by the reflection or scattering of the target, because this path is time-varying, assuming that this type of path signal is Here A _p and θ _p (p=1,2,3...) represent the magnitude and phase of the received signal of the pth path, then the received signal of link l at discrete time t=1,2,... Intensity RSS values are:

In the absence of a target, only the 1st, 2nd, and 3rd propagation paths exist, then the received signal does not contain time-varying items, and the RSS value of link l is:

When the noise effect of the circuit is not considered, and there are no other disturbances in the environment, the RSS value r _l expressed in (2) is unchanged.

When a target appears in the monitoring area, the RSS value of link l may change, then the change of the RSS value of link l at time t, that is, the attenuation of the received signal of link l at time t is:

Δr _l,t =r _l,t -r _l ,l=1,2,...,L. (3)

where L is the total number of links in the monitoring area, determined by the number of wireless nodes. The power attenuation of the received signal may occur in the following three situations: 1. The direct path is blocked by the target, 2. A reflection path with strong power is blocked by the target, 3. The time-varying multipath caused by target reflection or scattering RSS fast fading. As shown in Equation 1, the received signal strength is the result of the signal vector superposition of each path, and a change in the signal of a certain path may cause a huge change in RSS.

In wireless tomography, the attenuation of the LOS path signal is mainly used to image the target position, but we do not know in advance which link attenuation is caused by the LOS path occlusion. If we simply think that the received signal attenuation of the link l given by formula 3 at time t is the attenuation of the LOS path signal and apply it to wireless tomography, it will cause great interference to the imaging results. This phenomenon is called multipath interference in wireless tomography.

The most significant impact of multipath interference on wireless tomography is that the existence of multipath interference may make wireless tomographic imaging detect multiple false targets. In wireless tomography, objects appear as bright spots in the imaging map. The multipath interference makes the imaging image have many false bright spots besides the bright spots of the target, thus misestimating the number of targets. In addition, false targets will take away part of the energy of the target, making the brightness of the main target weaker, and may even be lower than the brightness of a false target, resulting in wrong target position estimation.

Generally speaking, the above third situation is that the time-varying multipath caused by target reflection or scattering makes the received signal exhibit the characteristics of RSS fast fading or imaging noise, which can be processed by time-domain methods such as low-pass filtering the signal device to eliminate. However, for the situation that a reflection path with strong power is blocked by the target, because the characteristics in the time domain of this situation are the same as those in the time domain when the LOS path is blocked, the processing method in the time domain is suitable for a certain power reflection path. There is nothing you can do about the situation where the reflection path is blocked.

So far, no one has proposed a problem similar to the case where a certain powerful reflection path is blocked and a solution to the related problem, and no one has done similar research in other fields.

Contents of the invention

Aiming at the problem that multi-path links have a great influence on the positioning of wireless tomography technology, the present invention proposes an indoor wireless tomography enhancement method based on motion prediction to delete interfering links. Before RTI imaging, link intersection points Airspace characterization detects and removes interfering links. Generally, the intersecting point of the interfering link due to multipath interference is far away from the real position of the target. Since we do not know the real position of the target, we can use the Kalman filter to use the target motion model and the positioning results of the previous moment to obtain an estimate of the target position at the current moment, thereby deleting the interference link that is far away from the target position. Since this method requires that the target position at the initial moment is known, the present invention uses clustering to find a set of intersection points composed of shadow fading links, and the center of the set can be used as an estimate of the target position at the initial moment, based on the target position estimate obtained by Kalman, and according to The position estimation deletes the interfering links to obtain the set of shadow fading links, and finally performs RTI imaging through the shadow fading links, and uses Kalman update on the basis of RTI to obtain the current position of the target. Experiments prove that after the interference link is deleted, both the imaging quality and the target tracking accuracy are greatly improved.

The method for enhancing indoor wireless tomography based on motion prediction and deleting interfering links of the present invention comprises the following steps:

Step 1: When the target is located in the monitoring area, measure the received signal strength of each link, that is, the change of RSS value:

Step 1.1: Configure Node

The monitoring area is located on the xoy coordinate plane, and o is the coordinate origin; n wireless nodes are deployed equidistantly around the monitoring area, and all nodes are placed on the same xoy coordinate plane, that is, all nodes are placed at the same height, and each node is placed by Assign a unique ID number as an identification, and the coordinates of each node are known as (α _q , β _q ), q=1,2,...,η, where q is the node number; these wireless nodes constitute L= η(η-1)/2 wireless links;

Each node sends signals according to a preset protocol and timing, and receives and measures the RSS value of wireless signals sent by other nodes, that is, at time t, the node numbered q sends data, and other nodes receive data and measure the received data. Signal strength; at the next moment, the node numbered q+1 sends data, other nodes receive data and measure the received signal strength;

Step 1.2: Measure the RSS variation Δr _{l,t of the lth link at time t} :

First measure the RSS value of each link when there is no target in the monitoring area, where _l is the number of the link; then measure each link at discrete time t=1,2, when the target is in the monitoring area. ... the RSS value r _l,t , so that the variation of the RSS value of the lth link at time t is:

Δr _l,t =r _l,t -r _l , l=1,2,...,L. (4)

A negative value of Δr _l,t indicates attenuation;

Step 2: Process the variation Δr _l,t of the RSS value of each link at time t obtained in step 1 to obtain the fading link set l _t at the current moment;

Step 2.1: At the current moment t, the RSS variation Δr _l,t received by each link first passes through the moving average filter to eliminate the rapidly changing noise, and 2ω+1 is the window length of the sliding filter;

Δr _l,i is the variation of RSS value of link l at time t-ω≤i≤t+ω, then the variation of RSS value of link l at time t after filtering is:

Step 2.2: Set a threshold and remove links with insignificant attenuation; that is, if the link l satisfies the following formula at the discrete time t=1, 2, ..., the link l at this moment is called an attenuated link :

in is the attenuation threshold, then the fading link set at time t is:

Step 3: Obtain the set of intersection points (u _k , v _k ) of all fading links in the fading link set l _t at time t in the monitoring area, that is, the set of intersection points ρ _t ;

As a preference, the method of obtaining the intersection coordinates of any two links is as follows:

At time t, suppose the two node coordinates of a link belonging to the set l _t are (α _i , β _i ) and (α _j , β _j ) respectively, and the two node coordinates of the other link belonging to l _t are respectively (α _m , β _m ) and (α _n , β _n ), then the intersection coordinates (u _k , v _k ) of these two links satisfy:

The solution of this expression can be expressed in matrix form as

Among them, [ ] ^-1 means to find the inverse matrix of the matrix.

Step 4: Obtain the initial position of the target at the initial time t=1

Step 4.1: Use the clustering algorithm to group the set of intersection points ρ _t of the attenuated links at time t obtained in step 3 according to the clustering characteristics, so as to find the clusters with obvious clustering;

Assume that K is the number of intersection point clusters at time t, Φ _j,t is the set of intersection points in cluster j at time t, |Φ _j,t | represents the number of points in cluster j; find each cluster at time t The center position of the class (C _x,j ,C _y,j ) minimizes the following objective function:

Here (C _x,j ,C _y,j ) is the coordinates of the central position of cluster j at time t, that is, the set of intersection points Φ _j,t ;

Step 4.2: Initialize K=1 at time t and assume that cluster j is the cluster containing the most intersection points at this time, and check whether the intersection points in this class satisfy Where (u _k ,v _k )∈Φ _j,t , and R is the preset distance threshold; if it is satisfied, the iteration terminates, otherwise the value of K is increased by 1, and return to step 4.1 to find the objective function shown in formula (11) to minimize (C _x,j ,C _y,j ); finally obtain the number K of clusters at time t;

Step 4.3: Select a cluster as the set of intersections of shadow fading links; generally the number of intersections in the set of intersections of shadowed fading links is the largest, so the set of intersections and initial positions of these links whose LOS paths are blocked can be given by the following formula out:

in Indicates that at time t, the cluster J with the largest number of intersection points among the K clusters is taken as the class with obvious aggregation characteristics, is the coordinate estimation of the current position of the target; when t=1 is the initial time, the target position at the initial time can be obtained as

Step 5: Predict the position of the target at discrete time t based on Kalman filter

When t≥2, estimate the position of the target at the current time, that is, the target at time t according to the target position at the previous time, that is, at time t-1. When t=1, it is the initial time, and the target position at the initial time is obtained in step 4.3 Assuming that the target is moving at a uniform speed in the monitoring area, then the motion equation of the target is

According to the uniform motion model of the target, it can be known that:

X _t is a 4-dimensional state variable, including target coordinates and speed; T is the sampling time interval of the target motion state, and are the x-direction velocity and y-direction velocity of the target in the xoy plane of the monitoring area at time t, and (x _t , y _t ) are the position coordinates of the target in the xoy plane of the monitoring area at time t; The noise in the direction ε _t = [ε _x,t ,ε _y,t ] ^T is a Gaussian distribution, and the covariance matrix of the noise is Its value is determined according to the target motion state;

Next, the estimated value of the target’s position coordinates at time t is obtained through Kalman filtering: Assume that the target position observation obtained at time t, that is, the two-dimensional coordinates of the target is Y _t , where the target position at time t=1 is estimated from the initial position of the target. 4.3 Acquired The target position observation Y _t at time t≥2 is obtained from the imaging results of wireless tomography, and the relationship between Y _t and X _t is:

Y _t =HX _t +w _t (15)

where w _t is assumed to have a mean of zero and a covariance matrix of Observation errors that obey a Gaussian distribution, is the variance of the measurement error; the observation matrix H is:

Then according to the following Kalman filter theory, the position of the target at time t is estimated as

Among them, P _t|t-1 is the prediction of the minimum mean square error matrix at time t-1 to the minimum mean square error matrix at time t, which is called the minimum prediction mean square error matrix, and P _t-1|t-1 is t-1 The minimum mean square error matrix at time; K _t is the Kalman gain at time t; is the prediction of the target state variable at time t-1 to the target state variable at time t; is the state variable of the target at time t-1, as the state variable The first two elements of

Step 6: Estimate based on the obtained target position Delete an attenuated link set The non-shadow fading links in , get the shadow fading link subset ξ _t ; the method is as follows:

If the point (u _k , v _k )∈l _t is the intersection point of the shadow fading link, then the distance between the target and the intersection point must satisfy:

in is the estimated value of the coordinates of the target position at time t according to the target position at time t-1 in the previous step, R _th is the distance threshold, and its value is greater than the distance threshold R in step 4.2; if the intersection point does not satisfy the condition of formula (18) , then it is judged that this intersection point is not the intersection point of the shadow fading link, so the intersection point is removed from the intersection point set of the attenuation link, that is, the intersection point set satisfying formula (18) in the attenuation link is obtained, and the attenuation link where these intersection points are located constitutes t Time shadow fading link set ξ _t ;

Step 7: According to the shadow fading link set ξ _t obtained in the above steps, obtain the position observation of the target at the current moment; the method is as follows:

The xoy plane of the monitoring area is divided into grids, Δυ is the side length of the grid, N _r and N _c are the number of grids contained in each row and column, respectively, and the weight matrix of wireless tomography Where d=1,2,...,N _r ×N _c represents the grid number, |ξ _t | represents the number of shadow fading links in the shadow fading link set ξ _t at time t;

by the formula Find the imaging matrix of the target at time t, where μ is the Tikhonov regularization parameter, Δx _d,t represents the dth element at time t, 1≤d≤N _r N _c , the result from formula (5), I is the identity matrix; since it cannot be guaranteed that all elements in is a positive value, it is necessary to force the negative number in to zero and divide the elements by the maximum value at the same time to perform normalization processing to obtain the RSS attenuation of grid d at discrete time t. The elements in are stacked into a two-dimensional matrix of N _r × N _c to image; the brightest point in the image is regarded as the position value of the target, and the imaging result is given by the following formula to obtain the target position:

Y _t is the position observation of the target described in step five, that is, the two-dimensional coordinates of the target, which is used to update the target position observation of the Kalman filter; symbol Indicates that the data is rounded down, D indicates the serial number of the grid with the largest attenuation, 1≤D≤N _r N _c ;

Step 8: Obtain the current position update of the target:

According to the following Kalman filtering theory, the position update of the target at time t is obtained:

Where P _t|t is the minimum mean square error matrix at time t, is the state variable at time t; The first two elements of That is, the updated value of the target position at time t.

Compared with the prior art, the beneficial effect of the present invention is that the RTI enhancement method proposed by the present invention based on motion prediction to delete interfering links comprehensively judges the relationship between the intersection points of each shadow fading link and the motion characteristics of the target, and excludes some non-shadow links based on motion prediction. The impact of fading links can effectively eliminate false targets, and more accurately highlight the attenuation effect caused by the existence of targets, so that target positioning in multipath environments is more accurate and better dynamic tracking is achieved.

Description of drawings

Figure 1: Illustration of target impact on propagation environment: LOS path blocked;

Figure 2: Illustration of the impact of the target on the propagation environment: the LOS path is not blocked;

Figure 3: Flowchart of the indoor wireless tomography enhancement method based on motion prediction interference link deletion;

Figure 4: Impact of interfering links on RTI: Links attenuated in the monitored area;

Figure 5: The clustering results of the intersection points obtained by using the K-means algorithm;

Figure 6: RTI's weight model;

Figure 7: Illustration of node locations and target motion trajectories.

Detailed ways

The present invention will be described in detail below in conjunction with accompanying drawing and embodiment, also described the technical problem and beneficial effect that the technical solution of the present invention solves simultaneously, it should be pointed out that described embodiment is only intended to facilitate the understanding of the present invention, and It has no limiting effect on it.

The flow chart of the present invention is shown in accompanying drawing 3, and the indoor wireless tomography enhancement method for deleting interference links based on motion prediction specifically includes the following steps:

Step 1: Obtain the change of all link received signal strength (RSS) values when the target is located in the monitoring area;

Step 1.1: Configure Node

The monitoring area is located on the xoy coordinate plane, and o is the coordinate origin. Deploy n wireless nodes supporting the IEEE802.15.4 protocol equidistantly around the monitoring area, all nodes are placed on the same xoy coordinate plane, that is, all nodes are placed at the same height, and each node is assigned a unique ID The number is used as an identification, and the coordinates of each node are known as (α _q , β _q ), q=1,2,...,η, where q is the node number. Each node sends and receives signals sequentially in the way of token ring. At a certain moment, the node whose serial number is q sends data, and other nodes receive data and measure the received signal strength. At the next moment, the node whose serial number is q+1 sends data, and other nodes receive data and measure the received signal strength. These wireless nodes can form L=η(η-1)/2 wireless links, and each node can measure the RSS value of wireless signals sent by other nodes according to the preset protocol and timing.

Step 1.2: Measure the RSS variation Δr _{l,t of the lth link at time t} :

First, when there is no target in the monitoring area, the RSS value of each link is r _l :

Among them, l is the number of the link, P is the number of reflection paths when there is no target in the monitoring area; A _LOS and θ _LOS represent the amplitude and phase of the received signal of the direct path, that is, the LOS path when there is no target in the monitoring area, B _p and θ _p (p=1,2,3...P) represents the magnitude and phase of the signal received by the pth reflection path.

Then measure the RSS value r _l,t of each link at discrete time t=1, 2, ... when the target is in the monitoring area, the sampling interval should be small enough to keep up with the speed of the target:

Among them, γ _LOS indicates whether the direct path is blocked: if γ _LOS = 1, it means that the direct path is not blocked by the target, otherwise γ _LOS = 0, it means that the direct path is blocked by the target;

γ _p indicates whether the p-th reflection path is blocked: if γ _p =1, it means that the p-th reflection path is not blocked by the target; otherwise, γ _p =0 means that the p-th reflection path is blocked by the target.

C _z (t) and θ _z (t) (z=1,2,3.....Z) represent the amplitude and phase of the z-th multipath signal generated by the reflection or scattering of the target, and Z is the The number of multipath signals generated by reflection or scattering.

Thus, the variation of the RSS value of the lth link at time t is obtained as:

Δr _l,t =r _l,t -r _l ,l=1,2,...,L.

Step 2: Process the variation Δr _l,t of the RSS value of each link at time t obtained in step 1 to obtain the fading link set l _t at the current moment;

Step 2.1: At the current moment t, the received RSS variation Δr _l,t must first pass through the moving average filter to eliminate rapidly changing noise, 2ω+1 is the window length of the sliding filter, and the window length should be selected properly , usually the preferred value range of the window length is 3-7.

Δr _l,i is the variation of RSS value of link l at time i (t-ω≤i≤t+ω), then the variation of RSS of link l at time t after filtering is:

Step 2.2: Generally speaking, if the link is blocked, regardless of whether the LOS path or the reflection path with strong energy is blocked, the RSS value will show a strong attenuation. Therefore, a threshold can be set to remove links with insignificant attenuation. If the discrete time t=1, 2, ..., the link l satisfies the following formula, the link l at this moment is called an attenuated link:

in is the attenuation threshold, The selection of the attenuation threshold is related to the given detection probability. When the detection probability meets the requirements, the preferred value range of the attenuation threshold is generally -1dB～-3dB. Then the fading link set at time t is:

As shown in Figure 4, a collection of all attenuated links in the detection area is given.

Step 3: Obtain the intersection set ρ t of each link in the set at time t in the monitoring area according to the fading link set l _t at time _t ;

As a preference, the method of obtaining the intersection coordinates of any two links is as follows:

At time t, suppose the two node coordinates of a link belonging to the set l _t are (α _i , β _i ) and (α _j , β _j ) respectively, and the two node coordinates of the other link belonging to l _t are respectively (α _m , β _m ) and (α _n , β _n ), then the intersection coordinates (u _k , v _k ) of these two links satisfy:

The solution of this expression can be expressed in matrix form as

Among them, [ ] ^-1 means to find the inverse matrix of the matrix. Based on this calculation, the intersection point set formed by the intersection points of all fading links in the fading link set l _t at time t in the monitoring area is:

Step 4: Obtain the initial position of the target at the initial time t=1

Step 4.1: Use a clustering algorithm (K-means clustering algorithm as the preferred method) to group the set of intersection points ρ _{t of attenuated links obtained in step 3 at time t} according to clustering characteristics to find clusters with obvious clustering. Assume that K is the number of clusters of intersection points at time t, Φ _j,t is the set of intersection points in cluster j at time t, and |Φ _j,t | represents the number of points in cluster j. Use the K-means method to find the appropriate cluster center (C _{x, j} , C _{y, j} ) at time t, that is, to find the appropriate cluster to minimize the following objective function:

Here (C _x,j ,C _y,j ) is the center of cluster j at time t, that is, the intersection point set Φ _j,t , that is to find the center position of each cluster at time t.

Step 4.2: Usually we don't know how many clusters these intersections should be divided into. If these intersections are the intersections of the links where the LOS path is blocked, that is, shadow fading links, then the number of clusters should be 1. However, if some intersection points are not shadow fading links, the number of clusters should be increased. Therefore, initialize K=1 at time t and assume that cluster j is the cluster with the largest number of elements at this time, and check whether the intersection points in this class satisfy Where R is the distance threshold, generally the radius of the target contour. If it is satisfied, the iteration terminates, otherwise K=K+1, return to step 4.1 to find (C _x,j ,C _y,j ) that minimizes the objective function (11), and finally obtain the number K of clusters at time t. As shown in Figure 5, the final number of clusters is 3, and each cluster exhibits certain clustering characteristics.

Step 4.3: After the clustering is completed, we need to select a cluster as the set of shadow fading link intersections. Generally, the number of intersections in the intersection set of shadow fading links is the largest, because the intersections of those links whose non-LOS paths are blocked are usually isolated, and it is difficult for them to form a cluster with many intersections. Thus the set of intersections and initial locations of these LOS path-occluded links can be given by:

in Indicates that at time t, the cluster J with the largest number of intersection points among the K clusters is taken as the cluster with obvious clustering, is the coordinate estimate of the target's current position. When t=1 is the initial moment, the target position at the initial moment can be obtained as

Step 5: Predict the position of the target at discrete time t based on Kalman filter

Estimate the position of the target at the current time, that is, the target at time t according to the target position at time t-1 (t≥2) at the previous time, when t=1 is the initial time, and the target position at the initial time is obtained in step 4.3 Assuming that the target is moving at a uniform speed in the monitoring area, then the motion equation of the target is

According to the uniform motion model of the target, it can be known that:

X _t is a 4-dimensional state variable, including target coordinates and speed. T is the sampling time interval of the target motion state, and are respectively the x-direction velocity and y-direction velocity of the target in the xoy plane of the monitoring area at time t, (x _t , y _t ) are the position coordinates of the target in the xoy plane of the monitoring area at time t, and then the target is obtained by Kalman filtering position coordinates at time t. Assume that the noise ε _t in the x direction and y direction of the xoy plane = [ε _x,t ,ε _y,t ] ^T is a Gaussian distribution, and the covariance matrix of the noise is Its value is generally determined according to the target motion state.

Next, the estimated value of the position coordinates of the target at time t is obtained through Kalman filtering: assuming that the target position observation obtained at time t, that is, the two-dimensional coordinates of the target is Y _t , where the target position at time t=1 can be estimated from the initial position of the target, namely The target position observation Yt at time _t >1 can be obtained from the imaging results of RTI. The relationship between _Yt and _Xt is:

Y _t =HX _t +w _t (15)

where w _t is assumed to have a mean of zero and a covariance matrix of Observation errors that obey a Gaussian distribution, is the variance of the measurement error. The observation matrix H is:

Then according to the following Kalman filter theory, the position estimate at time t can be obtained as

Among them, P _t|t-1 is the prediction of the minimum mean square error matrix at time t-1 to the minimum mean square error matrix at time t, which is called the minimum prediction mean square error matrix, and P _t-1|t-1 is t-1 The minimum mean square error matrix at any time; (the subscripts t-1|t-1 and t|t are called the minimum mean square error matrix, and the subscript t|t-1 is the minimum prediction mean square error matrix). K _t is the Kalman gain at time t. is the prediction of state variables at time t-1 to time t. is the state variable of the target at time t-1, as the state variable The first two elements of .

Step 6: Estimate based on the obtained target position Delete the non-shadow fading links in the attenuation link set l _t to obtain the shadow fading link subset ξ _t ;

For a link, only when the target is located on the LOS path or is relatively close to the LOS path, the link can observe RSS attenuation. Therefore, if the approximate location of the target is known It can be judged whether the attenuation of the link is caused by the blocking of the LOS path, so as to delete those interfering links whose non-LOS paths are blocked. If the intersection point (u _k , v _k )∈l _t is the intersection point of the shadow fading link, then the distance between the target and the intersection point must satisfy:

where is the estimated value of the coordinates of the target position at time t obtained according to the target position at time t-1 in the previous step, R _th is the distance threshold, and its value is generally greater than the distance threshold R in step 4.2. If the intersection point does not satisfy the condition of formula (18), it is judged that the intersection point is not the intersection point of the shadow fading link, and the intersection point is removed from the intersection point set of the fading link, so a new intersection point set can be obtained, that is, all intersection points in the intersection point set all satisfy the formula (18), so that the set of shadow fading links at time t can be obtained

Step 7: Obtain the position observation of the target at the current moment according to the shadow fading link set ξ _t obtained in the above steps;

The monitoring area is divided into grids, Δυ is the side length of the grid, N _r and N _c are the number of grids contained in each row and each column respectively, firstly, the weight matrix of RTI is obtained by using the ellipse weight model (see patent one Security monitoring system based on radio frequency node network detection, 201120506303.2), where d=1,2,...,N _r ×N _c represents the grid number, |ξ _t | represents the set of shadow fading links at time t The number of shaded fading links in _ξt . The RTI weight model is shown in Figure 6. by the formula Find the imaging matrix of the target at time t, where μ is the Tikhonov regularization parameter, Δx _d,t represents the dth element at time t, 1≤d≤N _r N _c , comes from the result of formula (5), and I is the identity matrix; since it cannot be guaranteed that all elements in If is a positive value, it is necessary to force the negative number in to zero and divide the elements by the maximum value at the same time to perform normalization processing to obtain the RSS attenuation of grid d at discrete time t. Arrange the elements in N _r ×N _c in a two-dimensional matrix in columns to form an image; the brightest point in the image is regarded as the position value of the target (but the target has a volume occupancy), and the RTI imaging is given by the following formula The resulting target position thus obtained is:

Y _t is the position observation of the target described in step five, that is, the two-dimensional coordinates of the target, which is used to update the target position observation of the Kalman filter; symbol Indicates that the data is rounded down, D indicates the serial number of the grid with the largest attenuation, 1≤D≤N _r N _c ;

Step 8: Obtain the current position update of the target:

According to the following Kalman filtering theory, the position update at time t can be obtained:

Where P _t|t is the minimum mean square error at time t, is the state variable at time t. The first two elements of That is, the updated value of the target position at time t. The present invention will be described in detail below in conjunction with specific signal examples:

In this experiment, we use 14 nodes supporting IEEE802.15.4 protocol. These nodes operate at 2.4GHz. IEEE802.15.4 specifies channels from 11 to 26 in the 2.4GHz frequency band, and we use channel 11 in this laboratory.

These nodes measure RSS and send the data to the base station nodes. We use a communication protocol similar to Token Ring to obtain the RSS value of each link in real time. At time t, the node whose serial number is q sends data, and other nodes receive data and measure the received signal strength. At time t+1, the node whose serial number is q+1 sends data, and other nodes receive it. When q=14, it is possible to measure and update the RSS values of all links once, and at this time, assign q to 1 again, and start a new round of measurement. In order to speed up the measurement, in the measurement, the information sent by each node is the RSS value with other nodes. The base station node only receives the data and transmits the RSS value to the local PC through the serial port.

In this experiment, the nodes are placed in an ordinary office environment. This environment is a typical indoor multipath environment, in which reflectors such as tables, chairs, books, walls, and computers are distributed. The 14 nodes are placed in this way, 10 nodes are placed on the table, and the other 4 nodes are fixed on the support and keep the same height as the nodes on the table. The distribution of these nodes is shown in the figure. The distance between nodes on the table is 1m, so 14 nodes can cover a monitoring area of 5*4=20 square meters. The arrangement of 14 nodes in this experiment is shown in Fig.7.

We first measure the static RSS value of the link when there is no target in the monitoring area, and then measure the RSS value of the link when the target exists and in the monitoring area. In this experiment, the target motion trajectory we tested is a rectangle, as shown in Figure 7. After receiving the signal, the local PC runs the data processing program to obtain the imaging and positioning results of the target.

By analyzing and comparing the experimental simulation results, it can be seen that the imaging results of the traditional RTI method are greatly affected by interference, and the image quality is poor. In addition to the bright spots of the target, there are many other false bright spots in the RTI image, which are distributed in the entire imaging area. There was even a point where the target didn't have a bright spot where it was supposed to appear, but the bright spot appeared elsewhere. On the contrary, the imaging result obtained by the method proposed by the present invention is obviously better than that of the traditional RTI method. The bright spot of the target is very close to the real position, and there are almost no other main bright spots except the bright spot of the target, and the image is cleaner. This is because the method proposed by the invention deletes the interfering link, so that the imaging quality is greatly enhanced.

Many bright spots can be seen in the imaging image obtained by the traditional RTI method, which makes the estimation of the number of targets wrong. In order to detect the number of objects in the graph, of course false objects are also included. The generated image is first binarized. In this experiment, the threshold is selected as 0.6, and the gray value greater than 0.6 is assigned 1, and the gray value smaller than 0.6 is assigned 0. Binarization can divide the image into several disjoint regions, and then count the number of disjoint regions in the binary image to obtain the number of targets. From the experimental results, it can be concluded that in the method proposed by the present invention, only one target is detected at 99% of the time, and multiple targets only appear at some time. However, the probability of detecting a single target by traditional methods is 80%, and in many moments three or even four targets can be detected.

From the target tracking trajectory, it can be seen that the target trajectory obtained by the traditional RTI method is far from the real trajectory of the target in some positions. There is a large error in the estimate. However, for the method proposed by the present invention, since those interfering links are deleted by using the method of motion prediction, the situation that the estimated position seriously deviates from the real position basically does not exist, so it seems that the trajectory estimation obtained by using the method of motion prediction is more accurate from the figure. smooth.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person familiar with the technology can understand the conceivable transformation and replacement within the technical scope disclosed in the present invention. It should be included within the scope of the present invention, therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims (9)

1. deleting the indoor wireless tomography Enhancement Method of interfering link based on motion prediction, which is characterized in that comprising as follows Step：

Step 1：When target is located at monitoring region, the variation of each link received signals intensity (RSS values) is measured：

Step 1.1：Configuration node

Monitoring region is located at xoy coordinate planes, and o is coordinate origin；Equidistant be deployed in of η radio node is monitored into region week It encloses, the height that all nodes are placed at i.e. all nodes placements on the same xoy coordinate planes is identical, and each node quilt The unique ID number of distribution one is as mark, and the coordinate of each node is it is known that be (α_q,β_q), q=1,2 ..., η, wherein q are section Point number；These radio nodes constitute L=η (η -1)/2 wireless links；

Each node sends signal according to preset agreement and sequential, and receive and measure other nodes sent out it is wireless The RSS values of signal, method are as follows：In t moment, the node transmission data that number is q, other nodes receive data and measure reception Signal strength；At the next moment, the node transmission data that number is q+1, other nodes receive data and measure receive signal it is strong Degree；

Step 1.2：Measure the l articles link t moment RSS variation deltas r_l,t：

When measuring monitoring region first does not have target, the RSS values of each of the links are r_l, wherein l is the number of the link；Then it surveys Measuring target, each of the links are in discrete instants t=1 when monitoring in region, and 2 ... the RSS values r of_l,t, to obtain the l articles link It is in the variable quantity of the RSS values of t moment：

Δr_l,t=r_l,t-r_l, l=1,2 ..., L. (4)

Δr_l,tDecay for negative value expression；

Step 2：To step 1 obtain each of the links t moment RSS values variation delta r_l,tIt is handled, is obtained current The decline link set at moment

Step 2.1：In current time t, RSS variation deltas r that each of the links receive_l,tFirst pass around moving average filter To eliminate the noise become soon, 2 ω+1 are that sliding filter window is long；

Δr_l,iFor link l the RSS values at t- ω≤i≤t+ ω moment variable quantity, then filtering after link l in t moment RSS variable quantities are：

Step 2.2：A threshold value is set, the unconspicuous link of decaying is removed；Method is as follows, if discrete instants t=1, 2 ... when link l meet following formula just claim moment link l be decaying link：

WhereinIt is drop threshold, then t moment decline link set is：

Step 3：Obtain t moment decline link setIn it is all decaying links intersection point (u_k,v_k) constituted in monitoring region Set, and it is referred to as intersection point set ρ_t；

Step 4：The initial position of target is obtained when carving t=1 at the beginning

Step 4.1：With clustering algorithm by the t moment obtained in step 3 decaying link intersection point set ρ_tAccording to Clustering features Grouping, to find the class for having and significantly building up；

Assuming that K is the number of t moment intersection point set cluster, Φ_j,tIt is the set of intersection point in t moment cluster j, | Φ_j,t| indicate poly- The number at the midpoints class j；Find the center (C each clustered under t moment_x,j,C_y,j) make following the minimization of object function：

Here (C_x,j,C_y,j) it is intersection point set Φ under t moment_j,tCenter position coordinates；

Step 4.2：T moment initializes K=1 and assumes that it includes the most cluster of intersection point at the moment to cluster j to be, is detected in such Whether intersection point meetsWherein (u_k,v_k)∈Φ_j,t, and R is preset distance threshold；Such as Fruit meets then iteration ends, and otherwise K values add 1, and return to step 4.1, which is found, makes the minimization of object function shown in formula (11) (C_x,j,C_y,j)；It is final to obtain the number K clustered under t moment；

Step 4.3：Select set of the cluster as shadow fading link intersection point；In general shadow fading link intersection point set The number of intersection point is maximum, thus the set of the intersection point of link that is blocked of these LOS paths and initial position can by following formula to Go out：

WhereinIt indicates in t moment using the most cluster J of intersection point number in K cluster as with apparent poly- Collect the class of characteristic,It is the coordinate estimation of target current time position；It is initial time as t=1, can obtains initial Moment target location is

Step 5：Position based on Kalman filter prediction target in discrete instants t

When t >=2, the current time i.e. position of t moment target is estimated according to the target location at previous moment, that is, t-1 moment, works as t It is initial time when=1, the target location of initial time is what step 4.3 obtainedAssuming that target is in monitoring region For uniform motion, then the equation of motion of target is

According to the uniform motion model of target, it is known that：

X_tFor the state variable of 4 dimensions, including coordinates of targets and speed；T is the sampling time interval to target state,WithIt is target respectively in t moment in the directions the x speed and the directions y speed of monitoring region xoy planes, (x_t,y_t) it is target in t The position coordinates being engraved in monitoring region xoy planes；Assuming that the noise ε on the directions x and the directions y of xoy planes_t=[ε_x,t, ε_y,t]^TIt is Gaussian Profile, the covariance matrix of noise isIts value is determined according to target state；

Followed by Kalman filter obtain target t moment position coordinates estimated value：Assuming that the target position that t moment obtains It is Y to set the observed quantity i.e. two-dimensional coordinate of target_t, the wherein moment target locations t=1 obtain i.e. step from target initial position estimation 4.3 acquisitionsThe moment target location observed quantities of t >=2 Y_tIt is obtained from the imaging results of wireless tomography, Y_tAnd X_t Relationship be：

Y_t=HX_t+w_t (15)

Wherein assume w_tBe mean value be zero, covariance matrix isGaussian distributed observation error,It is to measure The variance of error；I is unit matrix；Observing matrix H is：

So Kalman filter theory obtains target and is in the location estimation of t moment according to the following formula

Wherein P_t|t-1It is prediction of the t-1 moment least mean-square error matrixes to t moment least mean-square error matrix, is referred to as minimum Predict Square Error matrix, P_t-1|t-1It is t-1 moment least mean-square error matrixes；K_tIt is t moment Kalman gains；It is t-1 Prediction of the moment to t moment dbjective state variable；It is state variable of the target at the t-1 moment,For state VariableThe first two element；R is covariance matrix；

Step 6：According to obtained target location estimationDelete decaying link setIn non-shadow decline link, Obtain shadow fading link subset ξ_t；Method is as follows：

If pointIt is the intersection point of shadow fading link, then the distance between target and the intersection point must satisfy：

WhereinIt is that the estimation of t moment target location coordinate is obtained according to the target location at t-1 moment in previous step Value, R_thFor distance threshold, value is more than the distance threshold R in step 4.2；If intersection point be unsatisfactory for formula (18) this Part, then it is the intersection point of shadow fading link to judge this intersection point not, to remove the intersection point in the intersection point set of decaying link, The intersection point set for meeting formula (18) in decaying link is obtained, the decaying link where these intersection points constitutes t moment shade and declines Fall link set ξ_t；

Step 7：Shadow fading link set ξ obtained by above-mentioned steps_tObtain the position detection amount at target current time；Side Method is as follows：

The xoy planes in monitoring region are divided into grid, and Δ υ is the length of side of grid, N_rAnd N_cEvery row and each column include respectively Meshes number, the weight matrix of wireless tomographyWherein d=1,2 ..., N_r×N_cIndicate net Lattice number, | ξ_t| indicate t moment shadow fading link set ξ_tThe number of middle shadow fading link；

By formulaThe imaging array of t moment target is found out, whereinμ It is Tikhonov regularization parameters, Δ x_d,tIndicate t momentIn d-th of element, 1≤d≤N_rN_c,From formula (5) as a result, I is unit matrix；Due to cannot be guaranteedMiddle all elements are Positive value, need byIn negative pressure be assigned to zero and to wherein element simultaneously divided byMiddle maximum value, obtains normalized Result afterwardsWhereinIndicate the RSS attenuations of the grid d in discrete instants t；It willIn element By row stack arrangement at N_r×N_cTwo-dimensional matrix can be imaged；Most bright spot is considered as the positional value of target in image, then by following Formula provides imaging results to which the target location obtained is：

Y_tIt is the position detection amount i.e. two-dimensional coordinate of target of the target described in step 5, for updating Kalman filter Target location observed quantity；SymbolIndicate that, to the downward rounding of data, D indicates the serial number of the maximum grid of decaying, 1≤D≤N_rN_c；

Step 8：Obtain target current time location updating：

According to following Kalman filter theory obtain target t moment location updating：

Wherein P_t|tFor t moment least mean-square error matrix,It is the state variable of t moment；The first two elementThe as updated value of t moment target location.

2. the indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction according to claim 1, It is characterized in that, the η radio node supports IEEE802.15.4 agreements.

3. the indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction according to claim 1, It is characterized in that, in step 1.1, each node sequentially sends and receives signal successively in the way of token ring.

4. the indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction according to claim 1, It is characterized in that, in step 2.1, the value range of sliding filter window length is 3~7.

5. the indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction according to claim 1, It is characterized in that, in step 2.2,Value range be -1dB~-3dB.

6. the indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction according to claim 1, It is characterized in that, in step 3, the method for obtaining the intersecting point coordinate of arbitrary both links is as follows：

In t moment, it is assumed that belong to setA link two node coordinates be respectively (α_i,β_i) and (α_j,β_j), belong to Another link two node coordinates be respectively (α_m,β_m) and (α_n,β_n), then the intersecting point coordinate (u of this both links_k, v_k) meet：

The solution matrix form of this formula can be expressed as

Wherein []^-1Inverse of a matrix matrix is sought in expression.

7. the indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction according to claim 1, It is characterized in that, in step 4.1, clustering algorithm uses K-means.

8. the indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction according to claim 1, It is characterized in that, in step 4.1, preset distance threshold R values are the radius of objective contour.

9. the indoor wireless tomography Enhancement Method of interfering link is deleted based on motion prediction according to claim 1, It is characterized in that, the weight matrix of wireless tomography is obtained using oval weight model in step 7.