A three-dimensional radio frequency tomography method based on multi-layer electromagnetic fence

Abstract

This invention discloses a three-dimensional radio frequency tomography method based on a multi-layer electromagnetic fence, comprising the following steps: establishing a multi-layer electromagnetic fence measurement system, which forms a semi-enclosed three-dimensional measurement space; measuring the received signal strength (RSS) of the wireless link when there is no target and when there is a target, respectively, and calculating the change vector of the RSS of the wireless link when there is no target and when there is a target; establishing a three-dimensional shadow weight model based on the spatial relationship of the target's influence on the wireless link; establishing a three-dimensional radio frequency tomography model represented by tensors and converting it into vector form; performing imaging sequentially on the XOY, XOZ, and YOZ planes of the monitored area based on the three-directional projection method, and obtaining the three-dimensional imaging result of the target through extension and superposition, and finally through threshold filtering. This invention overcomes the shortcomings of existing two-dimensional radio frequency tomography, solves the problem of radio frequency tomography imaging of three-dimensional target shape information, and reduces the required computing and storage resources.

Description

Technical Field

[0001] This invention belongs to the field of radio frequency tomography technology, and specifically relates to a three-dimensional radio frequency tomography method based on a multi-layer electromagnetic fence. Background Technology

[0002] In a highly information-driven society, imaging technology plays an increasingly important role in many non-contact human-computer interaction, behavior analysis, warehousing and logistics, and personnel search and rescue fields, providing diverse and convenient services for people's daily lives, work, and entertainment. However, when the monitored target is unsuitable or unwilling to carry equipment, traditional imaging methods struggle to achieve satisfactory results. Therefore, device-free target imaging technology, which requires no equipment on the target, has emerged.

[0003] Currently, mainstream unmanned target imaging can be achieved through millimeter-wave radar or visible light methods. Although these methods have been widely applied, radar-based unmanned target imaging technology requires expensive specialized equipment and is mostly used in military applications, making it difficult to apply on a large scale in commercial or civilian fields. For a long time, people have used vision-based methods based on optical cameras to obtain quantitative images of targets. Especially with the penetration of intelligent technology into all aspects of people's daily lives and work, imaging targets with digital cameras and using artificial intelligence algorithms to identify them has become a common method of target detection. However, ordinary monocular cameras are easily affected by lighting conditions and cannot work in obstructed or dark conditions. This not only may involve privacy issues but also has drawbacks such as high algorithm complexity and inability to work in non-line-of-sight environments.

[0004] In recent years, with the development of wireless communication and embedded technologies, deviceless target imaging technology based on Wireless Sensor Networks (WSNs) has been used to compensate for the shortcomings of vision-based methods. Radio Tomographic Imaging (RTI), as a mainstream method, is a technique that inverts the target's position or geometry by monitoring changes in the Received Signal Strength Indication (RSSI) of radio frequency signals on wireless links. This technique typically deploys several wireless transceiver nodes at the boundary of the target area to form a closed monitoring zone. The wireless nodes can communicate with each other, forming many intersecting wireless links within the monitoring zone. When a target is located within the monitoring zone, it causes changes in the RSSI values ​​of some wireless links. By monitoring which links exhibit RSSI fluctuations, the target's position and even its geometry can be reconstructed. Due to the advantages of radio frequency signals, such as penetrability, non-invasiveness, and immunity to light, this technology has become another solution for target detection, with very broad application prospects in target recognition, positioning and navigation, and non-destructive testing. RTI has been around for over a decade since it was first proposed. Initially, it was mainly used in the field of localization, using the light spot of its imaging to map the location of the target. Therefore, some researchers call it Device-Free Localization (DFL).

[0005] While traditional RTI methods are relatively mature in position-aware applications, they can only provide light spots with brightness information; obtaining more detailed shape information remains a challenge in terms of modeling and implementation. Furthermore, current RTI imaging capabilities are primarily focused on the two-dimensional plane, with few reports on imaging three-dimensional targets in space. Summary of the Invention

[0006] The purpose of this invention is to provide a three-dimensional radio frequency tomography method based on a multi-layer electromagnetic fence. By constructing RSSI information of cross-layer links and a three-dimensional shadow weight model in a multi-layer electromagnetic fence system, a three-dimensional target imaging method based on three-directional projection solves the shortcomings of existing two-dimensional radio frequency tomography and the RTI imaging problem of three-dimensional target shape information. Moreover, it can reduce the required computing and storage resources.

[0007] To achieve the above objectives, the solution of the present invention is:

[0008] A three-dimensional radio frequency tomography method based on a multi-layer electromagnetic fence includes the following steps:

[0009] Step 1: Establish a multi-layer electromagnetic fence measurement system. The multi-layer electromagnetic fence measurement system consists of K layers, L nodes per layer, for a total of K×L wireless transceiver nodes. The K×L wireless transceiver nodes communicate with each other in pairs to form M wireless links. The multi-layer electromagnetic fence measurement system forms a semi-enclosed three-dimensional measurement space.

[0010] Step 2: Measure the received signal strength (RSS) of the wireless link when there is no target and when there is a target, and calculate the change vector of RSS of the wireless link when there is no target and when there is a target.

[0011] Step 3: Establish a three-dimensional shadow weight model based on the spatial relationship of the target's influence on the wireless link;

[0012] Step 4: Establish a three-dimensional radiofrequency tomography model in tensor representation and convert it into vector form;

[0013] Step 5: Based on the three-directional projection method, images are sequentially formed on the XOY, XOZ, and YOZ planes of the target area. After extension and superposition, the three-dimensional imaging result of the target is finally obtained through threshold filtering.

[0014] In step 1 above, the multi-layer electromagnetic fence measurement system consists of K layers of wireless transceiver nodes. Each layer has L nodes, which are evenly distributed around the perimeter of the area enclosed by the electromagnetic fence system. The layers are spaced at appropriate intervals as needed. The network is based on a round-robin token passing protocol. Each network node is assigned an independent and fixed ID number. At any given time, only one node is in the transmitting state, while the other nodes are in the receiving state. Signals are transmitted sequentially according to the ID order. After transmission, the node immediately switches to the receiving state, and the next ID node switches to the transmitting state. This process is repeated sequentially. There are wireless links between nodes in a single layer and between nodes in different layers. The three-dimensional measurement space is evenly divided into N1, N2, and N3 intervals in the X, Y, and Z directions, respectively, forming N = N1N2N3 three-dimensional voxels.

[0015] In step 2 above, according to communication theory, the formula for the RSS value of the received signal strength at the receiver in the l-th wireless link is as follows:

[0016] y l (t)=P l -L l -S l (t)-F l (t)-v l (t)

[0017] Where l = 1, 2, 3, ..., M, P l L represents the transmit power of the transmitting end. l S represents static loss. l (t) represents the shadow loss, F l(t) represents the fading loss, v l (t) represents noise;

[0018] Measure the RSS values ​​of the l-th link when there is no target and when there is a target, respectively, and measure the RSS change Δy of the l-th link at time t. l The formula for Δy is shown below: l (t)=y l (t)-y l (0), approximately equal to The value;

[0019] Where y l (0)=P l -L l -F l (0)-v l (0) represents the background RSS measurement value of the l-th link when no target exists.

[0020] Since noise and shadow fading are much smaller, Δy l (t) is mainly determined by the shadow fading at time t. Using the same measurement method, the formulas for the measurement vectors of all M wireless links are as follows:

[0021] y(t)=[y1(t)y2(t)…y M (t)] T

[0022] in,[] T Indicates the transpose operation;

[0023] The formula for the background measurement vector is as follows:

[0024] y(0)=[y1(0)y2(0)…y M (0)] T

[0025] The formula for calculating the difference between the measured vector y(t) and the background measured vector y(0) at time t is shown below:

[0026] Δy(t)=abs[y(t)-y(0)]=[Δy1(t)Δy2(t)…Δy M (t)]

[0027] Here, abs[] represents the absolute value operation.

[0028] In step 3 above, the formula for the 3D shadow weight model corresponding to the l-th (l = 1, 2, ..., M) link in the 3D shadow weight is as follows:

[0029]

[0030] Among them, w ln V represents the weight value corresponding to the influence of the target at the nth pixel on the lth link. l =4πa l b l c l / 3 represents the volume of the l-th ellipsoid, which has its foci at the locations of the two wireless nodes constituting the l-th link, where c l Let a be the polar radius of the ellipsoid corresponding to the l-th link. l and b l The equatorial radius of the ellipsoid corresponding to the l-th link is h, respectively. ln This indicates the position p of the nth voxel. n The vertical distance from the current location to the l-th link; Let d represent the difference in distances from the normalized projection point of the nth voxel on the l-th link to the transmitting and receiving nodes. l Let l be the length of the l-th link. p represents the position of the projection point of the nth voxel on the l-th link. i and p j These represent the positions of the sending and receiving nodes that constitute the l-th link, respectively; abs[] represents the absolute value operation.

[0031] The specific content of step 4 above is as follows:

[0032] The target region is divided into several small voxels of the same size, resulting in a tensor of D=3. Its mode-n expansion is as follows Where N1, N2, and N3 represent the number of voxels along the x, y, and z dimensions, respectively;

[0033] The formula for the inverse problem of 3D target reconstruction is shown below:

[0034] Δy=A(X)+n

[0035] Where, Δy∈R M×1 Let A(·) represent the measurement vector, n represent the measurement noise, and A(·) represent the tensor to be reconstructed. X Mapping between the measurement vector Δy and the target vector;

[0036] tensor X The vectorization process is performed to obtain the vector to be reconstructed, as shown in the following formula:

[0037]

[0038] Where vec(·) represents the operation of converting a tensor to a vector;

[0039] tensorX The elements (k1, k2, k3) in the vector are mapped to the l-th element of the vector x, as shown in the following formula:

[0040]

[0041] The formula for vector representation of radiofrequency tomography is shown below:

[0042] Δy=WΔx+n

[0043] Where Δy∈R M×1 For measurement vectors, For vectorized tensors to be reconstructed X , n∈R M×1 To measure noise, Represents the weight matrix;

[0044] In step 5 above, the specific content of the three-direction projection method includes:

[0045] Calculate the difference between the RSS values ​​when there is no target and when there is a target, and denote the result as Δy∈R. M×1 Based on the principle of radiofrequency tomography, the formula is as follows:

[0046] Δy=WΔx+n

[0047] where n∈R M×1 It measures noise. It is a vector representation of the signal attenuation at the coordinate positions of each voxel in the space to be estimated. It is a weight matrix;

[0048] Introducing sparse representation, we obtain the objective function, as shown in the following formula:

[0049]

[0050] Among them, matrix Let Δx be a sparse transformation basis for Δx, then Δx = φs, α represents the regularization coefficient, and ||X| (i) || * Represents matrix X (i) The nuclear norm of is given by ||·||, which represents the 2-norm. The solutions on the three projection planes XOY, XOZ, and YOZ are calculated sequentially to obtain:

[0051]

[0052]

[0053]

[0054] The top-view projection in the XOY direction is obtained as follows The principal projection in the XOZ direction is The side view projection in the YOZ direction is Finally, the vector Δx XOY Δx XOz and Δx YOZ Reshape them into two-dimensional matrices X according to the XOY plane dimensions N1×N2, the XOZ plane dimensions N1×N3, and the YOZ plane dimensions N2×N3 respectively. XOY X XOz and X YOZ ;

[0055] The obtained top view of the imaging space X XoY Main view X Xoz and side view X YOz The projection plane is copied and extended along the perpendicular direction of the plane, and the extension results in the three directions are added together to restore the tensor form of the three-dimensional imaging result, as shown in the following formula:

[0056]

[0057] Depending on the target and imaging space range, 70%-90% of the size of the brightest voxel is selected as the threshold. Pixels with values ​​less than the threshold are set to zero to filter out possible noise, and the remaining pixels form a three-dimensional target image.

[0058] The three-dimensional radio frequency tomography method based on multi-layer electromagnetic fences of this invention has the following beneficial effects:

[0059] (1) The three-dimensional radio frequency tomography method of the present invention utilizes the RSSI information of the cross-layer link in the multi-layer electromagnetic fence, enabling the system to have the ability to perceive three-dimensional targets in space, and obtains the refined imaging results of the target by constructing an imaging model that can describe the fine-grained shape features of the target, thus overcoming the shortcomings of the existing two-dimensional radio frequency tomography.

[0060] (2) The three-dimensional radio frequency tomography method of the present invention proposes a three-dimensional target imaging method based on three-directional projection. The method is used to sequentially image the XOY, XOZ and YOZ planes of the monitored area by acquiring the measurement vector. After extension and superposition, the three-dimensional imaging result of the target is finally obtained by image filtering. This not only solves the problem of radio frequency tomography imaging of three-dimensional target shape information, but also reduces the required computing and storage resources. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the multi-layer electromagnetic fence measurement system of the present invention.

[0062] Figure 2 This is a schematic diagram of the three-dimensional shadow weight model of the present invention.

[0063] Figure 3 This is a schematic diagram of the tensor of the present invention;

[0064] Wherein, (a) mode-1 represents column fiber, (b) represents row fiber, and (c) represents tube fiber.

[0065] Figure 4 These are photos of experimental scenarios from embodiments of the present invention.

[0066] Figure 5 This is an experimental result diagram of the three-dimensional target three-directional projection in an embodiment of the present invention;

[0067] Wherein, (a) represents XOY plane projection imaging in the 915MHz band, (b) represents XOZ plane projection imaging in the 915MHz band, and (c) represents YOZ plane projection imaging in the 915MHz band.

[0068] Figure 6 This is a diagram showing the experimental results of three-dimensional target imaging in an embodiment of the present invention. Detailed Implementation

[0069] The technical solution and beneficial effects of the present invention will be described in detail below with reference to the accompanying drawings.

[0070] like Figure 1 The diagram shows the structure of the multi-layer electromagnetic fence measurement system of the present invention. The system consists of K layers of wireless transceiver nodes, with L nodes in each layer, evenly distributed around the perimeter of the area enclosed by the electromagnetic fence system. Appropriate intervals are set between layers as needed. The network is based on a round-robin token passing protocol, assigning each node an independent and fixed ID number. At any given time, only one node is in the transmitting state, while the others are in the receiving state. Signals are transmitted sequentially according to ID order, and immediately after transmission, the system switches to the receiving state, with the next ID node taking over the transmitting state, and so on. There are wireless links between nodes in a single layer and between nodes in different layers. The three-dimensional space is evenly divided into N1, N2, and N3 intervals along the X, Y, and Z directions, respectively, forming N = N1N2N3 three-dimensional voxels. The KL wireless transceiver nodes communicate with each other in pairs, forming M wireless links.

[0071] like Figure 2 The diagram shown is a schematic of the three-dimensional shadow weight model of the present invention, which is constructed based on the spatial relationship of the target's influence on the wireless link.

[0072] Based on wireless signal transmission theory and the self-developed results summarized from experiments, the formula for the three-dimensional shadow weight model corresponding to the l-th (l=1,2,...,M) link in the three-dimensional shadow weight is as follows:

[0073]

[0074] Among them, w ln V represents the weight value corresponding to the influence of the target at the nth pixel on the lth link. l =4πa l b l c l / 3 represents the volume of the l-th ellipsoid, which has its foci at the locations of the two wireless nodes constituting the l-th link, where c l Let a be the polar radius of the ellipsoid corresponding to the l-th link. l and b l The equatorial radius of the ellipsoid corresponding to the l-th link is h, respectively. ln This indicates the position p of the nth voxel. n The vertical distance from the current location to the l-th link; Let d represent the difference in distances from the normalized projection point of the nth voxel on the l-th link to the transmitting and receiving nodes. l Let l be the length of the l-th link. p represents the position of the projection point of the nth voxel on the l-th link. i and p j These represent the positions of the sending and receiving nodes that constitute the l-th link, respectively; abs[] represents the absolute value operation.

[0075] like Figure 3 The diagram shows a fiber representation of the tensor of this invention, where (a) mode-1 represents column fibers, (b) mode-2 represents row fibers, and (c) mode-3 represents tube fibers. This invention divides the target region into several small voxels of equal size to obtain a tensor of D=3. Its mode-n expansion is as follows Where N1, N2, and N3 represent the number of voxels along the x, y, and z dimensions, respectively;

[0076] tensor X The elements (k1, k2, k3) in the vector are mapped to the l-th element of the vector x, as shown in the following formula:

[0077]

[0078] For the three-dimensional attenuation field reconstruction in this invention, there is the following inverse problem, as shown in the formula below:

[0079] Δy=A( X )+n

[0080] Where, Δy∈R M×1 Let A(·) represent the measurement vector, n represent the measurement noise, and A(·) represent the tensor to be reconstructed. XMapping between the measurement vector Δy and the target vector;

[0081] Since the tensor is a tuple of the rank of the n-rank tensor expanded from mode-n, the formula is as follows:

[0082] n-rank( X )=(rank(X (1) ), rank(X (2) ), rank(X (3) ))

[0083] make The tensor is obtained by minimizing the model as follows. X The inversion reconstruction formula is shown below:

[0084]

[0085] Since the tensor rank is discrete and non-convex, the above equation can be transformed by convex relaxation, as shown below:

[0086]

[0087] Where ||X (i) || * Represents matrix X (i) The nuclear norm of a matrix is ​​defined as the sum of its singular values ​​σ, as shown in the following formula:

[0088]

[0089] Due to the unavoidable noise in actual measurement processes, the above model was ultimately rewritten as an unconstrained tensor optimization problem:

[0090]

[0091] Where α > 0 represents the regularization parameter, which adjusts the weight ratio between the RTI process and the optimization term during the optimization process.

[0092] In summary, the optimization model framework for the inverse problem is shown in the above equation. The remaining problem is to find the solution derived from this model. The linear mapping relationship when the tensor X When there is a high spatial correlation, let F represent the... X Transform it into a linear mapping of the corresponding transform domain, at which point we have Most of the energy is concentrated in S Among the several low-frequency elements, namely S It is an approximately sparse tensor, which, after vectorization, yields s = vec( S If ), then there exists a matrix. For a sparse transformation of x, we have x = φs, such that:

[0093] Δy=WΦs+n

[0094] According to the definition of a tensor, a tensor is a fiber of a high-dimensional signal whose indices are fixed except for the d-dimensional dimension. For example, the tensor signal mentioned above... X i，j， = [ X (i, j, 1)... X [i, j, N3] is called a tensor. X The mode-3 fiber extends to a subset of the signal dimension, X i，.，. =[ X (i, 1, 1)... X [i, N2, N3] is called a tensor. X A horizontal slice of the tensor can be used to obtain a single sparse basis of the entire high-dimensional signal (the tensor to be reconstructed) through the Kronecker product of the sparse basis of each part, enabling encoding of all dimensions using a single transformation. Assuming that each fiber of any dimension of the tensor is sparse or compressible, the vectorized tensor sparse basis is constructed with a Kronecker structure, let F... N ={F k，n |1≤k,n≤N} (N≥1) represents an N×N dimensional unitary matrix, as shown in the following equation: The discrete cosine transformation matrix is:

[0095]

[0096] Right now in Let Kronecker product be represented, assuming but It has the following definition:

[0097]

[0098] In summary, the three-dimensional sparse optimization model is shown in the following equation:

[0099]

[0100] By combining a 3D RTI model with compressed sensing, this invention addresses the ill-posed problem caused by the number of equations (number of measurements) being much smaller than the number of unknowns (number of voxels to be reconstructed), enabling the reconstruction of attenuated fields in 3D space. However, because the number of grids used in 3D field reconstruction is typically multiplied geometrically compared to 2D field reconstruction, direct computation can easily lead to memory overflow. To ensure computation can proceed, this invention performs projection reconstruction in three directions: front view, side view, and top view. Then, signal processing methods such as extension superposition and interpolation filtering are used to achieve high-resolution reconstruction of the human body in real-world conditions.

[0101] By sequentially suppressing different individual dimensions of the tensor, the structure of the tensor elements resembles fibers of equal length, such as... Figure 3 (a) Taking the mode-1 fiber as an example, a three-dimensional exponential shadow weight model is used to model the communication link between any two nodes in a multi-layer electromagnetic fence, and then the tensor is... X The third dimension, the height direction, is treated as a whole, and corresponding weights are assigned to different column fibers to obtain a new weight matrix W. XOY This step is equivalent to a camera taking an image of the area to be monitored and adding a "filter" in the XOY direction, resulting in the corresponding sparse base φ XOY According to the Kronecker structure, it transforms into:

[0102]

[0103] At this point, the regularization term based on tensor recovery can degenerate into matrix recovery, and the objective function becomes:

[0104]

[0105] The final top-view projection in the XOY direction is: Similarly, the main view projection and side view projection in the XOZ and YOZ directions can also be obtained:

[0106]

[0107]

[0108] And the main view projection in the XOZ direction is obtained as follows The side projection in the YOZ direction is

[0109] To convert multiple projected two-dimensional images back into a three-dimensional image, the following needs to be considered: for a solid three-dimensional object, a single projected view can only reflect one side of the object's shape and cannot fully reflect the object's true geometric features. By projecting the same object from three different directions using three views, the object's structure can be described in a basically complete manner. Therefore, this invention copies and extends the projection planes of the target area's front view, side view, and top view along the vertical direction of the plane, restoring them back into tensors.

[0110] Since the shape and size of the XOY projection image are (N1, N2), the number of times it is extended depends on the tensor to be recovered. X The third dimension N3 can be obtained The shape and size of the XOZ projection image are (N1, N3), and the number of times it is extended depends on the tensor to be recovered. X The second dimension N2 can be obtained The shape and size of the YOZ projection image are (N2, N3), and the number of times it is extended depends on the tensor to be recovered. X The first dimension N1 can be obtained The above extended tensors are superimposed according to element positions to obtain This is called the initial tensor or reconstructed tensor, and its formula is shown below:

[0111]

[0112] Because the area where the target exists has a significant attenuation image, this attenuation effect will become more prominent after being superimposed. At the same time, three views are used to depict the edge part of the 3D target, so the edge contour of the target can be well preserved and distinguished from the background and target entity. The attenuation intensity shows characteristics between the background and the target entity.

[0113] like Figure 4The image shown is an experimental scene photograph of an embodiment of the present invention. Based on the 915MHz band SI4463 wireless transceiver chip, a wireless transceiver node was independently developed. The imaging area is a 3m × 3m × 2m rectangular area. Eight supports, each 2.2m high, are placed every 1.5m along the perimeter of the area. Starting 1m above the ground, a wireless transceiver node is placed every 20cm on each support, for a total of six layers, resulting in a total of 48 wireless transceiver nodes forming the radio frequency imaging network. One additional wireless node serves as a data acquisition node, responsible for transmitting measurement data to the computer. Each positioning node is placed on a 1m high support. Regarding the software protocol, this embodiment uses a polling-based token passing protocol for networking. Each network node is assigned an independent and fixed ID number. Only one node is in the transmitting state at any given time, while the others are in the receiving state. Signals are transmitted sequentially according to ID order. After transmission, the node immediately switches to the receiving state, and the next node with the next ID takes over the transmitting state, and so on. The program code for polling measurement and reading the received signal strength value was independently developed. Forty-eight positioning nodes are sequentially IDed from 1 to 48, with each ID distinguishing different modules. When a node sends positioning data, the data packet carries the ID number of the sending module. When the next node receives this ID number, it triggers the transmission of positioning data, thus establishing a round-robin measurement. After a sending node transmits positioning data, other wireless nodes receive the data and generate an RSSI value, immediately saving this data and then sequentially sending it to the data acquisition node, which then transmits it to the computer. Once the data is acquired, after processing, it is calculated using a three-dimensional shadow weight model and a three-directional projection method to obtain the three-dimensional radio frequency tomography positioning result. In this embodiment, the imaging target is a 1.8-meter-tall male, positioned at the center of the region of interest (1.5m, 1.5m) in a standing posture. Figure 5 As shown in the figure, the experimental results of three-dimensional target projection in the embodiment of the present invention are shown, where (a) represents the XOY plane projection imaging in the 915MHz band, (b) represents the XOZ plane projection imaging in the 915MHz band, and (c) represents the YOZ plane projection imaging in the 915MHz band. Figure 6 The figure shown is a diagram of the experimental results of three-dimensional target imaging in an embodiment of the present invention.

[0114] In summary, this invention provides a three-dimensional radio frequency tomography method based on a multi-layer electromagnetic fence, comprising the following steps: establishing a multi-layer electromagnetic fence measurement system, wherein the multi-layer electromagnetic fence measurement system consists of K layers, L nodes per layer, for a total of K×L wireless transceiver nodes, and the K×L wireless transceiver nodes communicate with each other to form M wireless links, and the multi-layer electromagnetic fence measurement system forms a semi-enclosed three-dimensional measurement space; measuring the received signal strength (RSS) values ​​of the wireless links when there is no target and when there is a target, respectively, and calculating the change vector of the RSS of the wireless links when there is no target and when there is a target; establishing a three-dimensional shadow weight model based on the spatial relationship of the target's influence on the wireless links; establishing a three-dimensional radio frequency tomography model in tensor representation and converting it into vector form; imaging the monitored area sequentially on the XOY, XOZ, and YOZ planes based on the three-directional projection method, and after extension and superposition, finally obtaining the three-dimensional imaging result of the target through threshold filtering. The three-dimensional radio frequency tomography method of this invention utilizes RSSI information from cross-layer links in a multi-layer electromagnetic fence, enabling the system to perceive three-dimensional targets in space. By constructing an imaging model that can describe the fine-grained shape features of the target, a refined imaging result is obtained, overcoming the shortcomings of existing two-dimensional radio frequency tomography. The three-dimensional target imaging method based on three-directional projection uses the acquired measurement vectors to sequentially image on the XOY, XOZ, and YOZ planes of the monitored area. After extension and superposition, the three-dimensional imaging result of the target is finally obtained through image filtering. This not only solves the problem of radio frequency tomography imaging of three-dimensional target shape information, but also reduces the required computing and storage resources.

[0115] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A method of three-dimensional radio frequency tomography based on a multi-layer electromagnetic fence, characterized in that, Comprising the following steps: Step 1, establish a multi-layer electromagnetic fence measurement system, the multi-layer electromagnetic fence measurement system is composed of K layers, each layer L K ×L wireless transceiver nodes, a total of KxL wireless transceiver nodes communicate with each other, forming M wireless links, the multi-layer electromagnetic fence measurement system forms a semi-closed three-dimensional measurement space;​ Step 2, respectively measure the receiving signal strength RSS value of the wireless link when there is no target and when there is a target, and calculate the change vector of RSS when there is no target and when there is a target; Step 3, according to the spatial relationship of the target affecting the wireless link, a three-dimensional shadow weight model is established; Step 4, a three-dimensional radio frequency tomography model represented by a tensor is established, and is converted into a vector form; Step 5, based on the three-direction projection method, imaging is sequentially performed on the XOY, XOZ and YOZ planes of the target area, after extension and superposition, finally the three-dimensional imaging result of the target is obtained through threshold filtering; The specific content of step 4 is: The target region is divided into a plurality of small voxels of the same size, to obtain a tensor , mode-n expansion of which is represented as , wherein respectively represent the number of voxels along x, y, z dimensions; The inverse problem formula of three-dimensional target reconstruction is as follows: ， wherein, denotes a measurement vector, denotes a measurement noise, denotes a mapping from the tensor to be reconstructed to the measurement vector between the tensor tensor The vectorization processing is performed on the tensor to obtain a to-be-reconstructed vector, and a formula is as follows: ， wherein represents a conversion of a tensor to a vector operation; tensor elements in Mapped to a vector The first l The formula for the given elements is as follows: ， The formula of radio frequency tomography represented by a vector is as follows: ， wherein is the measurement vector, is the vectorized tensor to be reconstructed , is the measurement noise, denotes the weight matrix; In step 5, the specific content of the three-direction projection method includes: The difference between the RSS of the targetless time and the target time is calculated, and the result is recorded as According to the principle of radio tomographic imaging, the formula is as follows: ， wherein is the measurement noise, is a vector representation of the signal attenuation to be estimated for the coordinate position of each voxel in space, is the weight matrix; Sparse representation is introduced to obtain the objective function, and the formula is as follows: ， where the matrix is a sparse transform basis with , , denotes a regularization coefficient, denotes the nuclear norm of the matrix , denotes the 2-norm, the solutions on the XOY, XOZ, YOZ projection planes are calculated in turn, and the following is obtained: ， ， ， The top view projection in XOY direction is , the front view projection in XOZ direction is , and the side view projection in YOZ direction is ; finally, the vectors , and are reshaped into two-dimensional matrices , and in XOY plane dimension N1xN2, XOZ plane dimension N1xN3 and YOZ plane dimension N2xN3 respectively. The obtained imaging space top view , front view and side view The projection surface is copied and extended along the vertical direction of the plane, and the three-direction extension results are added to restore the three-dimensional imaging results in tensor form, as shown in the following formula: ， According to different targets and imaging space ranges, 70%-90% of the size of the brightest voxel is selected as the threshold, the pixel values less than the threshold are taken as zero, the possible noise points are filtered out, and the remaining pixels form a three-dimensional target image.

2. The method of three-dimensional radio frequency tomography according to claim 1, characterized in that: In step 1, the multi-layer electromagnetic fence measurement system is composed of K layers of wireless transceiver nodes, each layer of nodes is L and is uniformly distributed on the periphery of the area surrounded by the electromagnetic fence system, and is networked based on a token passing protocol based on round robin, each networked node is assigned an independent and fixed ID number, only one node is in the transmitting state at each moment, the remaining nodes are in the receiving state, the transmitting signal is sequentially transmitted according to the ID order, and the transmitting state is immediately converted into the receiving state after the transmitting is completed, the next ID node is converted into the transmitting state, and the process is sequentially performed, there are wireless links between single-layer nodes and between layers of nodes; The three-dimensional measurement space is uniformly divided into N1, N2 and N3 intervals in the X, Y and Z directions respectively, to form N=N1N2N3 three-dimensional voxels.

3. The method of three-dimensional radio frequency tomography of claim 1, wherein: In step 2, according to the communication theory, the received signal strength RSS value of the receiving end in the wireless link is shown as follows: The formula of the received signal strength RSS value of the receiving end in the wireless link is shown as follows: ， wherein , denotes the transmit power of the transmitting end, denotes the static loss, denotes the shadowing loss, denotes the fading loss, denotes the noise; Measure the first time when there is no target and when there is a target respectively. l The RSS measurement value of the link at time [time]. t No. l RSS change of each link The formula is shown below: Approximately equal to The value; wherein , represents the background RSS measurement of the l link when no target is present, ; All M The measurement vector for a strip wireless link is given by: ， wherein [] T denotes a transpose operation; The formula of the background measurement vector is as follows: ， The difference between the background measurement vector and the background measurement vector t is obtained The formula for the momentary RSS change vector is as follows: ， wherein represents an absolute value operation.

4. The method of three-dimensional radio frequency tomography of claim 1, wherein: In step 3, the third dimension shadow weight of the first The three-dimensional shadow weight model formula corresponding to the link is shown as follows: ， in, , Indicates when the target is located at the th n When the pixel is... The weight values ​​corresponding to the impact of each link. For the first The volume of the ellipsoid, which forms the volume of the first ellipsoid. l The two wireless nodes on the link are the focal points, where For the first Each link corresponds to the polar radius of the ellipsoid. and The first Each link corresponds to the equatorial radius of the ellipsoid. Indicates from the first n Individual point location to the first l The vertical distance between the links; , indicating the normalized first n Individual prime points in the th l The difference in distance from the projection point on the link to the transmitting and receiving nodes. For the first l Link length, Indicates the first n Individual prime points in the th l The location of the projection point on the link, and They respectively represent the components of the first l The locations of the sending and receiving nodes of the link; [] indicates absolute value operation.

Images