Through-wall rapid human body target detection method based on ultra wide band radar

Abstract

The invention discloses a through-wall rapid human body target detection method based on an ultra wide band radar. Firstly, a radar is preliminarily deployed to collect data at the central position of a wall body; the method comprises the following steps of: firstly, performing preprocessing methods such as singular value decomposition (SVD), adaptive background subtraction (ABS), robust principal component analysis (RPCA) and the like on radar original echo data so as to effectively remove background clutters and noises, and then, performing the preprocessing methods such as the singular value decomposition (SVD), the adaptive background subtraction (ABS) and the RPCA on the radar original echo data so as to effectively remove the background clutters and the noises; then, the variance of each distance bin of the preprocessed signal matrix is calculated, and whether a human body target exists behind the wall or not is judged by comparing the maximum variance value with a preset threshold value; and finally, gradually narrowing the search range through a bisection method in combination with the signal processing technology, so as to realize the preliminary positioning of the human body target. And then, through constructing a position model and based on a geometric constraint relation, accurate positioning of the human body target is further realized. According to the invention, through combination of radar signal processing and a binary search method, rapid and accurate through-wall human body target detection is realized.

Description

Technical Field

[0001] The invention belongs to human target detection technology, and in particular relates to a through-wall rapid human target detection method based on ultra-wideband radar. Background Art

[0002] Human target detection technology is a technology that uses sensors and signal processing to identify, locate and track human targets. It has important application value in many fields. With the continuous improvement of society's demand for safety, health and intelligence, human target detection technology has gradually become a research hotspot and has shown broad application prospects in emergency rescue, security monitoring, smart home, medical health and other fields.

[0003] In traditional application scenarios, human target detection technology mainly relies on optical cameras, infrared sensors, and ultrasonic devices. Optical cameras use computer vision algorithms to identify human targets by capturing image information within the visible light range, but their performance is significantly reduced in low-light or obstructed environments. Infrared sensors detect targets by detecting thermal radiation emitted by the human body, but are easily interfered by other heat sources in complex environments. Although ultrasonic devices have certain penetration capabilities, their resolution and detection distance are limited, making it difficult to meet high-precision requirements. In non-line-of-sight or complex environments such as through walls, smoke, and ruins, the limitations of traditional technologies are more prominent. These challenges have given rise to the development of through-wall human target detection technology.

[0004] At present, through-wall human target detection based on radar has become a hot research field, among which ultra-wideband radar has gradually become the core technology of through-wall detection due to its advantages such as large bandwidth, high resolution, low power consumption and strong penetration. By emitting electromagnetic waves and analyzing their reflected signals, the radar can penetrate non-metallic walls and detect signal changes caused by micro-movements such as human breathing and heartbeat, thereby realizing accurate detection and positioning of human targets behind the wall. The existing method of through-wall human detection based on ultra-wideband radar requires searching every area of ​​the wall, which wastes a lot of time. Especially in post-disaster rescue operations, this inefficient search method may lead to missing the best rescue opportunity. The present invention discloses a through-wall rapid human target detection method based on ultra-wideband radar. First, the radar is initially deployed to collect data at the center of the wall; secondly, the radar raw echo data is preprocessed by singular value decomposition (SVD), adaptive background subtraction (ABS) and robust principal component analysis (RPCA) to effectively remove background clutter and noise; then, the variance of each distance bin of the preprocessed signal matrix is ​​calculated, and the maximum variance value is compared with the preset threshold to determine whether there is a human target behind the wall; finally, the search range is gradually narrowed by combining the above signal processing technology with the bisection method to achieve the preliminary positioning of the human target. Subsequently, by constructing a position model and based on the geometric constraint relationship, the precise positioning of the human target is further achieved. The patent of this invention realizes fast and accurate detection of human targets through walls by combining radar signal processing with the bisection search method. Summary of the invention

[0005] The purpose of the present invention is to provide a fast and accurate method for detecting human targets through walls, which uses a signal processing algorithm to filter out background clutter and noise from radar signals, and uses a variance threshold to determine whether there is a human target behind the wall. Combining the dichotomy method with geometric relationships, the human target can be located with fewer radar deployment times.

[0006] The present invention discloses a method for rapid through-wall human target detection based on ultra-wideband radar, comprising the following steps:

[0007] Step 1: Build an experimental platform based on impulse-radio ultra-wideband (IR-UWB) radar to collect and store radar echo signals from different areas of the wall.

[0008] Step 2: Preprocess the original echo signal collected in step 1 to remove various background clutter and noise interference. Specifically, it removes wall interference through singular value decomposition; removes static clutter through time mean subtraction (TMS) and adaptive background subtraction; estimates and removes linear components in the signal through linear least squares fitting (LTS); and effectively suppresses residual wall noise and other clutter signals through a robust principal component analysis algorithm.

[0009] Step 3: Calculate the variance of each distance bin of the preprocessed signal, and compare the maximum variance value with the preset threshold. When the maximum variance value exceeds the threshold, it indicates that there is a human target in the area.

[0010] Step 4: Based on the signal processing technology of steps 2 and 3, combined with the iterative regional division mechanism of the binary method, the search range is gradually narrowed to achieve the preliminary positioning of the human target. Due to the potential influence of orientation deviation, the position model is further constructed based on the geometric constraint relationship to achieve accurate positioning of the human target. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flow chart of the present invention;

[0012] Figure 2 Visualization of the data matrix before and after preprocessing when a human target is detected.

[0013] Figure 3 This is the variance curve before and after preprocessing.

[0014] Figure 4 Schematic diagram of human target detection based on dichotomy.

[0015] Figure 5 Schematic diagram of human target positioning based on geometric relationships. Specific implementation plan

[0016] The present invention will be further described below in conjunction with the accompanying drawings:

[0017] like Figure 1 The method for rapid through-wall human target detection based on ultra-wideband radar shown in the figure specifically comprises the following steps:

[0018] Step 1: Build an experimental platform based on impulse ultra-wideband (IR-UWB) radar to collect and store radar echo signals from different areas of the wall.

[0019] Step 2: Preprocess the original echo signal collected in step 1 to remove various background clutter and noise interference to obtain a relatively pure signal matrix. The preprocessing mainly includes 4 steps:

[0020] Step 1: Remove wall interference and perform singular value decomposition (SVD) on the input echo matrix R to obtain the three matrices shown in the following formula:

[0021] SVD[R] (M×N) =U (M×M) *S (M×N) *V T (N×N) (1)

[0022] where S is a diagonal matrix, U is an M×M unitary matrix, and V is an N×N unitary matrix. The elements in S are the singular values ​​σ i , arranged from high to low σ1≥σ2≥σ3≥···≥σ r ≥0. In order to remove the wall effect, it is necessary to remove the first singular value and reconstruct the radar matrix. Singular value decomposition can be used to remove wall interference and obtain

[0023]

[0024] Step 2: Remove static clutter, which includes background clutter and DC components. Use time mean subtraction (TMS) to process the DC component:

[0025]

[0026] Modify the radar receiving echo matrix to:

[0027] R 2(m,n) =R 1(m,n) -DC (4)

[0028] Adaptive Background Subtraction (ABS) is used to remove background clutter caused by static objects in the environment. This method uses a weighting coefficient λ to give:

[0029] B(m,n)=λ×B(m,n-1)+(1-λ)×R2(m,n) (5)

[0030] Where λ = 0.95. After obtaining the background estimate, the radar echo matrix is ​​modified using equation (6):

[0031] R3(m,n)=R2(m,n)-B(m,n) (6)

[0032] Step 3: Remove the linear trend, estimate and remove the linear component in the signal through linear least squares fitting (LinearleasT-Squares, LTS), so that it is no longer affected by the linear trend. The linear least squares fitting results in:

[0033] X T =R3 T -z(z T z) -1 z T R3 T (7)

[0034] Where z = [n,1 N ],n=[0,1,2,...,M] T , 1 N is an N*1 vector of unit values.

[0036] Step 4: Remove residual noise, and further introduce the Robust Principal Component Analysis (RPCA) algorithm to effectively suppress residual wall noise and other clutter signals. The RPCA algorithm uniquely decomposes the original matrix X into a low-rank matrix A and a sparse matrix E, thereby characterizing the principal components of the signal and the noise and abnormal information respectively.

[0037] X=A+E (8)

[0038] For the objective function, A and E satisfy the following conditions:

[0039] minrank(A)+γ||E||0(9)

[0040] stX=A+E

[0041] Since the optimization problem is non-convex and it is difficult to solve directly, the alternating direction method of multipliers (ADMM) is used for iterative solution. To facilitate optimization, a Lagrangian multiplier matrix Y and slack variables are introduced, the constraints are integrated into the objective function, and the enhanced Lagrangian function is constructed:

[0042]

[0043] By fixing other variables, A, E, and Y are iteratively updated respectively. First, the low-rank matrix A is updated by singular value soft threshold decomposition; secondly, the sparse matrix E is updated by element-wise soft threshold operation; finally, the Lagrange multiplier matrix Y is adjusted to approximate the constraints. Through iterative optimization, the residual background noise and interference are separated into matrix A, and the target signal is extracted into matrix E.

[0044] Step 3: Based on the target subspace matrix obtained in step 2, calculate the variance of each distance bin:

[0045]

[0046] in, is the average value of the mth distance bin, is the nth slow time sample of the mth range bin. Then, the maximum variance value V=max(V m ) will be the focus of attention. Since it is impossible to directly determine whether there is a human target behind the wall through V, an optimal threshold that can minimize the false detection rate (the probability that a non-human target is mistakenly identified as a human target) is introduced. If V exceeds the preset threshold, it means that there is a human target within the radar scanning range behind the wall.

[0047] Step 4: Based on the signal processing technology of steps 2 and 3, a binary search method is first used to achieve the preliminary positioning of the human target. The initial detection is performed at the center of the wall. If the human target is not detected, the wall is evenly divided into two sub-areas, and detection is performed at the center of each sub-area. If the target is still not detected, each sub-area is further divided into two smaller sub-areas and the detection process is repeated. Through this iterative area division and detection mechanism, the search range is gradually narrowed until the human target is detected. Considering the spatial distribution characteristics of the radar radiation field, there may be azimuth deviations in the target detection results. Therefore, by further constructing a position model and based on geometric constraints, the precise positioning of the human target can be achieved. When a human target is detected behind the wall, the radar is moved to the left and right sides by a distance m (this distance is the target distance detected by the radar last time) for re-detection. When a human target is detected only on one side and the distance from the radar is n, the angle α between the radar and the human target and the distance d0 required to move when the radar is facing the human target can be directly calculated; when a human target is detected on both sides at the same time and the distances from the radar are n and p respectively, by comparing the sizes of n and p, the area with a smaller distance is selected as the area where the human target is located, and then α and d0 are calculated.

[0048]

[0049] The position of the human target is obtained by moving the radar from the center of the current area by a distance d0. By detecting this position, the shortest distance L between the radar and the human target can be obtained.

[0050] L=vτ / 2 (14)

[0051] Among them, τ is the time delay between the transmitted signal and the received signal, and v is the propagation speed of the electromagnetic pulse in the air, that is, the speed of light.

Claims

1. A method for rapid through-wall human target detection based on ultra-wideband radar, characterized in that The following steps are involved: Step 1: Build an experimental platform based on impulse-radio ultra-wideband (IR-UWB) radar to collect and store radar echo signals from different areas of the wall. Step 2: Preprocess the original echo signal collected in step 1 to remove various background clutter and noise interference to obtain a relatively pure signal matrix. The preprocessing mainly includes 4 steps: Step 1: Remove wall interference and perform singular value decomposition (SVD) on the input echo matrix R to obtain the three matrices shown in the following formula: SVD[R] (M×N) =U (M×M) *S (M×N) *V T (N×N) (1) where S is a diagonal matrix, U is an M×M unitary matrix, and V is an N×N unitary matrix. The elements in S are the singular values ​​σ i , arranged from high to low σ1≥σ2≥σ3≥···≥σ r ≥0. In order to remove the wall effect, it is necessary to remove the first singular value and reconstruct the radar matrix. Singular value decomposition can be used to remove wall interference and obtain Step 2: Remove static clutter, which includes background clutter and DC components. Use time mean subtraction (TMS) to process the DC component: Modify the radar receiving echo matrix to: R 2(m,n) =R 1(m,n) -DC (4) Adaptive Background Subtraction (ABS) is used to remove background clutter caused by static objects in the environment. This method uses a weighting coefficient λ given by: B(m,n)=λ×B(m,n-1)+(1-λ)×R2(m,n) (5) Where λ = 0.

95. After obtaining the background estimate, the radar echo matrix is ​​modified using equation (6): R3(m,n)=R2(m,n)-B(m,n) (6) Step 3: Remove the linear trend by using linear least squares fitting (LTS) to estimate and remove the linear component in the signal so that it is no longer affected by the linear trend. The linear least squares fitting yields: X T =R3 T -z(z T from) -1 from T R3 T (7) Where z=[n,1 N ],n=[0,1,2,...,M] T , 1 N is an N*1 vector of unit values. Step 4: Remove residual noise, and further introduce the Robust Principal Component Analysis (RPCA) algorithm to effectively suppress residual wall noise and other clutter signals. The RPCA algorithm uniquely decomposes the original matrix X into a low-rank matrix A and a sparse matrix E, thereby characterizing the principal components and noise and abnormal information of the signal respectively. X=A+E (8) For the objective function, A and E satisfy the following conditions: min rank(A)+γ||E||0 (9) stX=A+E Since the optimization problem is non-convex and it is difficult to solve directly, the alternating direction method of multipliers (ADMM) is used for iterative solution. To facilitate optimization, a Lagrangian multiplier matrix Y and slack variables are introduced, the constraints are integrated into the objective function, and the enhanced Lagrangian function is constructed: By fixing other variables, A, E, and Y are iteratively updated respectively. First, the low-rank matrix A is updated by singular value soft threshold decomposition; secondly, the sparse matrix E is updated by element-wise soft threshold operation; finally, the Lagrange multiplier matrix Y is adjusted to approximate the constraints. Through iterative optimization, the residual background noise and interference are separated into matrix A, and the target signal is extracted into matrix E. Step 3: Based on the target subspace matrix obtained in step 2, calculate the variance of each distance bin: in, is the average value of the mth distance bin, is the nth slow time sample of the mth range bin. Then, the maximum variance value V=max(V m ) will be the focus of attention. Since it is impossible to directly determine whether there is a human target behind the wall through V, an optimal threshold that can minimize the false detection rate (the probability that a non-human target is mistakenly identified as a human target) is introduced. If V exceeds the preset threshold, it means that there is a human target within the radar scanning range behind the wall. Step 4: Based on the signal processing technology of steps 2 and 3, a binary search method is first used to achieve the preliminary positioning of the human target. The initial detection is performed at the center of the wall. If the human target is not detected, the wall is evenly divided into two sub-areas, and detection is performed at the center of each sub-area. If the target is still not detected, each sub-area is further divided into two smaller sub-areas and the detection process is repeated. Through this iterative area division and detection mechanism, the search range is gradually narrowed until the human target is detected. Considering the spatial distribution characteristics of the radar radiation field, there may be azimuth deviations in the target detection results. Therefore, by further constructing a position model and based on geometric constraints, the precise positioning of the human target can be achieved. When a human target is detected behind the wall, the radar is moved to the left and right sides by a distance m (this distance is the target distance detected by the radar last time) for re-detection. When a human target is detected only on one side and the distance from the radar is n, the angle α between the radar and the human target and the distance d0 required to move when the radar is facing the human target can be directly calculated; when a human target is detected on both sides at the same time and the distances from the radar are n and p respectively, by comparing the sizes of n and p, the area with a smaller distance is selected as the area where the human target is located, and then α and d0 are calculated. When the radar is moved from the center of the current area by a distance d0, the human target is located. By detecting this position, the shortest distance L between the radar and the human target can be obtained. L=vτ / 2 (14) Among them, τ is the time delay between the transmitted signal and the received signal, and v is the propagation speed of the electromagnetic pulse in the air, that is, the speed of light.

Images