CN104267394A - High-resolution human body target motion feature detecting method - Google Patents

Abstract

The invention discloses a high-resolution human body target motion feature detecting method. The method comprises the steps that firstly, a radar demodulation target echo signal model with micro-Doppler features is built, then signals are over-sampled, the time frequency distribution of the signals is calculated through time frequency analysis, binaryzation and other optimizing processing are carried out on the time frequency distribution of the signals, and a time frequency curve matrix is obtained; then, straight line detection is carried out on the matrix, human body main trunk motion parameters are obtained, a parameter space is set, the motion parameters of all the joints are detected, and the human body motion parameters are obtained. Compared with a traditional detecting method, the terahertz frequency band feature is fully utilized, the defect that the binding force of an existing human body target motion feature visual image sequential detection method on detecting conditions and the resolution ratio of a traditional radar detecting technology are low is overcome, and the human body target motion parameters can be detected faster and more accurately within a certain distance through the method.

Description

High resolving power human body Target Motion Character detection method

Technical field

The invention belongs to Radar Signal Processing Technology field, relate to high resolving power human body Target Motion Character detection method.

Background technology

Human body motion feature all possesses some special knowledge in biomedical engineering, physiatrics, medical diagnosis and rehabilitation.Not only there is widespread demand in the place such as sanatorium, hospital to the detection of human body motion feature, also have a lot of application in the field such as security protection, battle reconnaissance.Under the promotion of Analysis of mobility, visual surveillance, biometrics development, the method extracting and analyze different human motions is in widespread attention.

At present, the method for the most general human body motion feature uses visual pattern sequence.But visually-perceptible human motion can be subject to the impact that distance, light change, dress ornament change and partes corporis humani position are blocked in appearance, and detection perform reduces.Radar is as a kind of electromagnetic sensor, and because operating distance is far away, daytime and night can work and have the ability penetrating body of wall and ground, are also often used to the detection to human body motion feature.But the frequency of operation of conventional radar is lower, the micro-Doppler effect impact of human motion is very little, and the feature of human motion in a noisy environment is more difficult to carry out high-resolution human body target motion feature and detects.

Summary of the invention

The object of the present invention is to provide high resolving power human body Target Motion Character detection method, solve the constraint of visual pattern sequence pair motion feature detection and the problem of conventional radar detection resolution deficiency under noise conditions.

The technical solution adopted in the present invention is carried out according to following steps:

Step 1: set up the radar demodulation target echo signal model including micro-Doppler feature, described demodulation target echo model is specially:

$S\left(t\right)=S_{\mathrm{LO}}\left(t\right)\·\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{S}_{\mathrm{Ri}}\left(t\right)=e^{-2\mathrm{\πj\αt}}\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{e}^{2\mathrm{\πj\α}(t-{2R}_i\left(t\right)/c)}=\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{e}^{-2\mathrm{\πj}\left(2{\mathrm{\λ}}^{-1}{R}_i\left(t\right)\right)}$

Wherein, echoed signal S (t), S _lOt () is Radar Local-oscillator signal, S _rit () (i ∈ [0, N]) is trunk and each movable joint echoed signal, during i=0, and S _r0t echoed signal that () is trunk, N is the total number in human motion joint producing echo, and j is imaginary unit, and α is radar carrier frequency, c and λ is the light velocity and radar signal wavelength respectively, R ₀t () is Human torso: R ₀(t)=R ₀+ vt, R _it () is the motion simple model under the ideal conditions of human body each joint: R _i(t)=R ₀(t)+r _isin (2 π f _pt+ θ _i), wherein, R ₀for the distance of human body and radar, v is the radial velocity of the relative radar of human body, r _ifor the amplitude in each joint, θ _ifor the phase place in each joint, f _pfor the frequency in joint each when human body is walked about;

Step 2: to echoed signal S (t) over-sampling, obtains discrete signal S [n], does time frequency analysis to S [n], obtain the time-frequency energy distribution W of discretize _s(n, f), described time-frequency energy distribution W _s(n, f) is right approximate evaluation, wherein, represent the radial velocity of each target of human body relative to radar;

Step 3: to time-frequency energy distribution matrix W _s(n, f) carries out process optimization, obtains the curve point matrix of the corresponding time-frequency spectrum of trunk and each joint;

Step 4, carries out Hough straight-line detection to curve point matrix, obtains slope k and the intercept d of trunk line correspondence, according to formula v=f _dthe velocity information of main body trunk can be tried to achieve in λ/2, wherein f _dfor the point-to-point real-time frequency of trunk homologous thread, λ is radar signal wavelength, then by the slope obtained and intercept to curve point matrix H _{x × Y}revise, the straight line of each point according to trunk is compensated heart position in a matrix, namely according to the intercept d obtained, matrix is often arranged and all move down individual position, the point making matrix first row represent trunk position in a matrix between row, then obtain according to slope k the line number that often row should move down obtain revised curve point matrix;

Step 5, asks weighted sum to revised matrix by row, and does FFT conversion, try to achieve the cycle T of each time-frequency spectrum curve, namely the period of motion in each joint of human body, then set up Amplitude-Phase parameter space (r, θ), wherein r is profile amplitude parameter, and θ is phase parameter, sets up totalizer ∑ g (r, θ), by each point in parameter space, namely each group (r, θ) parameter value bring into before motion simple model R _i(t)=r _isin (2 π ft+ θ _i) in revised matrix H _{x × Y}in each curve point calculate, the curve point meeting this parameter drag is added up, find out local maximum point in totalizer g, the range value that each local maximum point is corresponding in totalizer and phase value are the amplitude of each bar curve and the concrete numerical value of phase place, due to revised matrix H _{x × Y}row represent frequency direction count, the time orientation that list is shown is counted, ordinate and the horizontal ordinate of image is in matrix image, can in the hope of this in frequency corresponding to frequency direction in the position in ordinate direction according to each point on time-frequency spectrum curve, according to v=f _dλ/2 can draw the instantaneous velocity information in each joint, then according to motion simple model R _i(t)=r _isin (2 π ft+ θ _i) the relative trunk maximal rate in known each joint and amplitude relation be v _{i_max}=2 π fr _i, can in the hope of the amplitude information in each joint.

Further, in described step 2 with much larger than Nyquist sampling frequency to echoed signal S (t) over-sampling.

Further, doing time frequency analysis process to S [n] in described step 2 is: adopt Short-time Fourier STFT (short-timeFouriertransform) Time-frequency Analysis.Short time discrete Fourier transform is specially

$W_s(n,f)=\underset{m=-\∞}{\overset{\∞}{\mathrm{\Σ}}}s\left[m\right]w[n-m]e^{-j2\mathrm{\πfm}}$

S [n] is signal, and w [n] is window function, and f is frequency, and n is frequency, and m is the frequency of change.

Further, in described step 3 to time-frequency energy distribution matrix W _s(n, f) carries out process optimizing process:

Setting threshold value by matrix W _s(n, f) is converted into the two values matrix H only containing 0 and 1 _{x × Y}, transfer process is for making H _{x × Y}=W _s(n, f), by matrix H _{x × Y}in be more than or equal to T _gelement put 1, the element being less than element sets to 0, and wherein X is that frequency direction is counted, and Y is that time orientation is counted, according to column major order by H _{x × Y}element put into vectorial p _hin, namely in matrix each element according to it in matrix H _{x × Y}middle line number and columns sort ((columns-1) × X+ line number) according to following formula, put into vectorial p successively according to position _hin, p _hin containing X × Y element, and each element is only 0 or 1,

(i.e. p _h=(0,0 ... 0,1,1 ... 1,0,0 ... 0,1,1 ... 1,0,0 ...)), then to vectorial p _hask difference, obtain vectorial p (in p element should for following form p=(0,0 ... 0,1,0 ... 0 ,-1,0 ...)), in p be the element of 1 position correspond to matrix H _{x × Y}in position be the coboundary of time-frequency spectrum curve, for the position correspondence of the element of-1 is the lower boundary of time-frequency spectrum curve, the Shi Fei border, position of 0 element in p; After finding out the position of in p 1 and-1, boundary element can be found out in matrix H according to counter the pushing away of aforementioned row priority ordering _{x × Y}in position, the position of element non-zero in p is recorded in vectorial d successively _nin, so that next step carries out the merging of edge, concrete is exactly being averaging every a pair up-and-down boundary, and obtain the central value of often pair of up-and-down boundary, this value is exactly the position of time-frequency spectrum curve point; By two values matrix H _{x × Y}reset and assignment again, determine each curve point position in a matrix according to the method for row priority ordering, be i.e. the line number of this point and columns, position corresponding in matrix is put 1, obtains curve point matrix:

H _{x × Y}middle each point is exactly the frequency corresponding to each joint of this time point n, and wherein each bar curve approaches the rate signal in each joint.

Further, straight-line detection carried out to curve point specifically the adopting Hough transform method of described step 4.

Further, curve detection carried out to curve point matrix specifically the adopting parameters space to add up the method for estimated parameter of described step 5.

The invention has the beneficial effects as follows and under Low SNR, high resolution detection can be carried out to human body target motion feature.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of high resolving power human body Target Motion Character detection method of the present invention;

Fig. 2 is the time frequency analysis figure of the echoed signal of the embodiment of the present invention;

Fig. 3 is the curve point schematic diagram of the time-frequency distributions of the embodiment of the present invention;

Fig. 4 is the Hough transform human body trunk kinematic parameter figure of the embodiment of the present invention;

Fig. 5 is the schematic diagram of each joint kinematic parameter of parameter space human body of the embodiment of the present invention;

Fig. 6 is the schematic diagram of the human body sport parameter of the embodiment of the present invention.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described further, and compare with traditional detection method.

Time-frequency spectrum curve extracting method based on rim detection of the present invention, as shown in Figure 1, comprises the following steps:

Step 1, set up the radar demodulation target echo signal model including micro-Doppler feature, described demodulation target echo model is specially:

$S\left(t\right)=S_{\mathrm{LO}}\left(t\right)\·\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{S}_{\mathrm{Ri}}\left(t\right)=e^{-2\mathrm{\πj\αt}}\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{e}^{2\mathrm{\πj\α}(t-{2R}_i\left(t\right)/c)}=\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{e}^{-2\mathrm{\πj}\left(2{\mathrm{\λ}}^{-1}{R}_i\left(t\right)\right)}$

Wherein, echoed signal S (t), S _lOt () is Radar Local-oscillator signal, S _rit () (i ∈ [0, N]) is trunk and each movable joint echoed signal, during i=0, and S _r0t echoed signal that () is trunk, N is the total number in human motion joint producing echo, and j is imaginary unit, and α is radar carrier frequency, c and λ is the light velocity and radar signal wavelength respectively, R ₀t () is Human torso: R ₀(t)=R ₀+ vt, R _it () is the motion simple model under the ideal conditions of human body each joint: R _i(t)=R ₀(t)+r _isin (2 π f _pt+ θ _i), wherein, R ₀for the distance of human body and radar, v is the radial velocity of the relative radar of human body, r _ifor the amplitude in each joint, θ _ifor the phase place in each joint, f _pfor the frequency in joint each when human body is walked about.

Step 2, to echoed signal S (t) over-sampling in step 1, obtains discrete signal S [n] (existing standard: sample to signal with much larger than Nyquist sampling frequency.), time frequency analysis is done to S [n], obtains the time-frequency energy distribution W of discretize _s(n, f), described time-frequency energy distribution W _s(n, f) is right approximate evaluation, wherein, represent the radial velocity of each target of human body relative to radar.

Here time frequency analysis specifically adopts Short-time Fourier STFT (short-time Fourier transform) Time-frequency Analysis.Short time discrete Fourier transform is specially

$W_s(n,f)=\underset{m=-\∞}{\overset{\∞}{\mathrm{\Σ}}}s\left[m\right]w[n-m]e^{-j2\mathrm{\πfm}}.$

S [n] is signal, and w [n] is window function, and f is frequency, and n is frequency, and m is the frequency of change.The time frequency analysis figure of concrete echoed signal as shown in Figure 2.

Step 3, to the time-frequency energy distribution matrix W obtained in step 2 _s(n, f) carries out process optimization, obtains the curve point matrix of the corresponding time-frequency spectrum of trunk and each joint.Owing to there being many frequency bands in time-frequency distributions, i.e. corresponding multiple joint, here to W _s(n, f) is optimized and specifically adopts with the following method:

Setting threshold value by matrix W _s(n, f) is converted into the two values matrix H only containing 0 and 1 _{x × Y}, specifically make H _{x × Y}=W _s(n, f), by matrix H _{x × Y}in be more than or equal to T _gelement put 1, the element being less than element sets to 0, and wherein X is that frequency direction is counted, and Y is that time orientation is counted; According to column major order by H _{x × Y}element put into vectorial p _hin, namely in matrix each element according to it in matrix H _{x × Y}middle line number and columns sort ((columns-1) × X+ line number) according to following formula, put into vectorial p successively according to position _hin, obviously, p _hin containing X × Y element, and each element is only 0 or 1, (i.e. p _h=(0,0 ... 0,1,1 ... 1,0,0 ... 0,1,1 ... 1,0,0 ...)),

Again to vectorial p _hask difference, obtain vectorial p (in p element should for following form p=(0,0 ... 0,1,0 ... 0 ,-1,0 ...)), in p be the element of 1 position correspond to matrix H _{x × Y}in position be the coboundary of time-frequency spectrum curve, for the position correspondence of the element of-1 is the lower boundary of time-frequency spectrum curve, the Shi Fei border, position of 0 element in p; After finding out the position of in p 1 and-1, boundary element can be found out in matrix H according to counter the pushing away of aforementioned row priority ordering _{x × Y}in position, the position of element non-zero in p is recorded in vectorial d successively _nin, so that next step carries out the merging of edge, concrete is exactly being averaging every a pair up-and-down boundary, and obtain the central value of often pair of up-and-down boundary, this value is exactly the position of time-frequency spectrum curve point; By two values matrix H _{x × Y}reset and assignment again, determine each curve point position in a matrix according to the method for row priority ordering, be i.e. the line number of this point and columns, position corresponding in matrix is put 1, obtains curve point matrix:

H _{x × Y}middle each point is exactly the frequency corresponding to each joint of this time point n, and wherein each bar curve approaches the rate signal in each joint.The curve point matrix H of concrete time-frequency distributions _{x × Y}image as shown in Figure 3, matrix H _{x × Y}shown in each bar curve, be the time-frequency spectrum curve having reacted Time-Frequency Information.

Step 4, carries out Hough straight-line detection to the curve point matrix in step 3, obtains slope k and the intercept d of trunk line correspondence, and Hough transform is straight-line detection field prior art, according to formula v=f _dthe velocity information of main body trunk can be tried to achieve in λ/2, wherein f _dfor the point-to-point real-time frequency of trunk homologous thread, λ is radar signal wavelength.Again by the slope obtained and intercept to curve point matrix H _{x × Y}revise, the straight line of each point according to trunk is compensated heart position in a matrix, namely according to the intercept d obtained, matrix is often arranged and all move down individual position, the point making matrix first row represent trunk position in a matrix between row, then obtain according to slope k the line number that often row should move down so just have modified the curve point matrix H in step 3 _{x × Y}, namely present matrix H _{x × Y}the time-frequency spectrum curve of image centered by central row, to determine the cycle of each bar time-frequency spectrum curve according to center of curve, amplitude, the parameters such as phase place.Concrete Hough transform human body trunk kinematic parameter figure as shown in Figure 4.

Step 5, to corrected matrix H after matrix step 4 _{x × Y}ask weighted sum by row, and do FFT conversion, try to achieve the cycle T of each time-frequency spectrum curve, namely the period of motion in each joint of human body; Set up Amplitude-Phase parameter space (r, θ) again, wherein r is profile amplitude parameter, θ is phase parameter, sets up totalizer ∑ g (r, θ), by each point in parameter space, namely each group (r, θ) parameter value bring into before motion simple model R _i(t)=r _isin (2 π ft+ θ _i) in corrected matrix H _{x × Y}in each curve point calculate, the curve point meeting this parameter drag is added up, find out local maximum point in totalizer g, the range value that each local maximum point is corresponding in totalizer and phase value are the amplitude of each bar curve and the concrete numerical value of phase place, due to revised matrix H _{x × Y}row represent frequency direction count, the time orientation that list is shown is counted, ordinate and the horizontal ordinate of image is, so can in the hope of this in frequency corresponding to frequency direction in the position in ordinate direction according to each point on time-frequency spectrum curve, according to v=f in matrix image _dλ/2 can draw the instantaneous velocity information in each joint, then according to motion simple model R _i(t)=r _isin (2 π ft+ θ _i) the relative trunk maximal rate in known each joint and amplitude relation be v _{i_max}=2 π fr _i, can in the hope of the amplitude information in each joint.As shown in Figure 5, the schematic diagram of human body sport parameter as shown in Figure 6 for the schematic diagram of the concrete each joint kinematic parameter of parameter space human body.

Beneficial effect of the present invention: first method of the present invention sets up the radar demodulation target echo signal model including micro-Doppler feature, by to echo signal sample and time frequency analysis obtains signal time-frequency distributions, then the curve point matrix that time-frequency distributions matrix obtains the corresponding time-frequency spectrum of trunk and each joint is processed, parameters space is detected curve point again, obtains the characteristic informations such as the frequency of the speed of human motion and the motion of each trunk and speed and reaches and detect graveyard.The invention solves visual pattern sequence pair motion feature detect constraint and conventional radar under noise conditions detection resolution deficiency etc. shortcoming, the method can carry out high resolution detection to human body target motion feature under Low SNR.

The present invention will be described to enumerate specific embodiment below:

Embodiment 1: establish someone at distance radar R ₀=30m sentences the speed motion of relative radar radial velocity v=0.09m/s, and it is joint A everywhere, and B, C, D are all with f _pthe frequency of=1Hz presses sine swing, and wherein joint A, B amplitude of fluctuation is r ₁=r ₂=0.6m, phase by pi, joint C, D amplitude of fluctuation is r ₃=r ₄=0.36m, phase place is phase difference of pi also, joint A, C and B, D phase place is identical, and namely the initial phase of A, B, C, D is respectively radar carrier frequency α=340GHz, wavelength sample frequency is 4096Hz.Then target echo is:

$S\left(t\right)=\underset{i=0}{\overset{4}{\mathrm{\Σ}}}{e}^{-2\mathrm{\πj}\left({2\mathrm{\λ}}^{-1}{R}_i\left(t\right)\right)}$

Wherein j is imaginary unit, and t is that instantaneous time is long, R ₀(t)=R ₀+ vt is trunk and radar instantaneous distance model, R _it () (i=1,2,3,4) are respectively as joint A, the instantaneous distance of B, C, D and radar, it is expressed as: R _i(t)=R ₀(t)+r _isin (2 π f _pt+ θ _i), above-mentioned various in parameter as previously mentioned.

Carry out over-sampling to the noisy echoed signal S (t) that contains in 0-3s with the sample frequency of 4096Hz, obtain discrete signal S [n], carry out STFT to it, window function length is 256 points, obtains time-frequency energy distribution W _s(n, f), its dimension is 256 × 12288, and image as shown in Figure 2.

Find out time-frequency distributions matrix W again _smaximal value in (n, f), setting threshold value by W _sthe value of (n, f) is assigned to new dimension matrix H _{x × Y}, by matrix H _{x × Y}in be more than or equal to T element put 1, the element being less than element sets to 0, obtain only containing 0 and 1 two values matrix H _{x × Y}.

According to column major order by H _{x × Y}element put into vectorial p _hin, namely in matrix each element according to it in matrix H _{x × Y}middle line number and columns, according to formula ((columns-1) × X+ line number) sequence, put into vectorial p according to position successively _hin, obviously, p _hin containing 256 × 12288 elements, and each element is only 0 or 1, (i.e. p _h=(0,0 ... 0,1,1 ... 1,0,0 ... 0,1,1 ... 1,0,0 ...)), then to vectorial p _hask difference, obtain vectorial p (in p element should for following form p=(0,0 ... 0,1,0 ... 0 ,-1,0 ...)), in p be the element of 1 position correspond to matrix H _{x × Y}in position be the coboundary of time-frequency spectrum curve, for the position correspondence of the element of-1 is the lower boundary of time-frequency spectrum curve, the Shi Fei border, position of 0 element in p; After finding out the position of in p 1 and-1, boundary element can be found out in matrix H according to counter the pushing away of aforementioned row priority ordering _{x × Y}in position, the position of element non-zero in p is recorded in vectorial d successively _nin, so that next step carries out the merging of edge, concrete is exactly being averaging every a pair up-and-down boundary, and obtain the central value of often pair of up-and-down boundary, this value is exactly the position of time-frequency spectrum curve point; By two values matrix H _{x × Y}reset and assignment again, determine each curve point position in a matrix according to the method for row priority ordering, be i.e. the line number of this point and columns, position corresponding in matrix is put 1, obtains curve point matrix:

H _{x × Y}middle each point is exactly the frequency corresponding to each joint of this time point n, and wherein each bar curve approaches the rate signal in each joint.The curve point schematic diagram of concrete time-frequency distributions as shown in Figure 3.

To curve point matrix H _{x × Y}carry out Hough straight-line detection, as shown in Figure 4, slope k=0 of straight line and intercept d=115, reflect acceleration and the initial velocity information of the motion of trunk.K=0 represents that the acceleration of human motion be 0, d=115 is straight line line number in a matrix, is expressed as the point-to-point real-time frequency of trunk line correspondence then the speed of human motion can be tried to achieve according to formula conform to pre-conditioned.

Due to matrix H _{x × Y}middle behavior 128 row, and trunk straight line is at the 115th row, so every column matrix element circular to be moved down 13 row, makes trunk straight line in matrix H _{x × Y}middle row, so that next step is to the extraction of each articulation point parameter of curve.

To corrected matrix H _{x × Y}ask weighted sum by row, and do FFT conversion, try to achieve the cycle T=1s of each curve, namely the period of motion in each joint of human body, its frequency is conform to default.Set up parameter space (r, θ), wherein r ∈ Z [1,127] (this span is determined due to matrix line number, and matrix line number is 256, so sinusoidal curve amplitude is 127 to the maximum) is profile amplitude parameter, θ gets 90 intervals in 0 ° ~ 360 ° and is worth for phase parameter uniformly, set up totalizer ∑ g (r, θ), the curve model R before being brought into by point each in parameter space _i(t)=r _isin (2 π ft+ θ _i) in H _{x × Y}in each curve point calculate, the curve point meeting this parameter drag is added up, finds out local maximum point in totalizer g.As shown in Figure 5,4 local maximum point homography H are had in figure _{x × Y}in 4 sinusoidal curves, i.e. 4 joints of target body, initial phase in a matrix and amplitude are respectively (90 °, 53), (270 °, 53), (90 °, 32), (270 °, 32), namely A is reacted, the time-frequency spectrum curve in B, C, D tetra-joints.According to aforementioned formula v=f _dλ/2 can in the hope of each joint and the concrete kinematic parameter of trunk, as shown in Figure 6.

These are only preferred embodiment of the present invention, be not limited to the present invention, for the person of ordinary skill of the art, the present invention can have various modifications and variations.Therefore all within principle of the present invention, other the various amendment made, equivalent replacement, concrete distortion and combination improve, and all should be included in protection scope of the present invention.

Claims (6)

1. high resolving power human body Target Motion Character detection method, is characterized in that carrying out according to following steps:

Step 1: set up the radar demodulation target echo signal model including micro-Doppler feature, described demodulation target echo model is specially:

$S\left(t\right)=S_{\mathrm{LO}}\left(t\right)\·\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{S}_{\mathrm{Ri}}\left(t\right)=e^{-2\mathrm{\πj\αt}}\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{e}^{2\mathrm{\πj\α}(t-2R_i\left(t\right)/c)}=\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{e}^{-2\mathrm{\πj}\left({2\mathrm{\λ}}^{-1}{R}_1\left(t\right)\right)}$

Wherein, echoed signal S (t), S _lOt () is Radar Local-oscillator signal, S _rit () (i ∈ [0, N]) is trunk and each movable joint echoed signal, during i=0, and S _r0t echoed signal that () is trunk, N is the total number in human motion joint producing echo, and j is imaginary unit, and α is radar carrier frequency, c and λ is the light velocity and radar signal wavelength respectively, R ₀t () is Human torso: R ₀(t)=R ₀+ vt, R _it () is the motion simple model under the ideal conditions of human body each joint: R _i(t)=R ₀(t)+r _isin (2 π f _pt+ θ _i), wherein, R ₀for the distance of human body and radar, v is the radial velocity of the relative radar of human body, r _ifor the amplitude in each joint, θ _ifor the phase place in each joint, f _pfor the frequency in joint each when human body is walked about;

Step 2: to echoed signal S (t) over-sampling, obtains discrete signal S [n], does time frequency analysis to S [n], obtain the time-frequency energy distribution W of discretize _s(n, f), described time-frequency energy distribution W _s(n, f) is right approximate evaluation, wherein, represent the radial velocity of each target of human body relative to radar;

Step 3: to time-frequency energy distribution matrix W _s(n, f) carries out process optimization, obtains the curve point matrix of the corresponding time-frequency spectrum of trunk and each joint;

Step 4, carries out Hough straight-line detection to curve point matrix, obtains slope k and the intercept d of trunk line correspondence, according to formula v=f _dthe velocity information of main body trunk can be tried to achieve in λ/2, wherein f _dfor the point-to-point real-time frequency of trunk homologous thread, λ is radar signal wavelength, then by the slope obtained and intercept to curve point matrix H _{x × Y}revise, the straight line of each point according to trunk is compensated heart position in a matrix, namely according to the intercept d obtained, matrix is often arranged and all move down individual position, the point making matrix first row represent trunk position in a matrix between row, then obtain according to slope k the line number that often row should move down obtain revised curve point matrix;

Step 5, asks weighted sum to revised matrix by row, and does FFT conversion, try to achieve the cycle T of each time-frequency spectrum curve, namely the period of motion in each joint of human body, then set up Amplitude-Phase parameter space (r, θ), wherein r is profile amplitude parameter, and θ is phase parameter, sets up totalizer Σ g (r, θ), by each point in parameter space, namely each group (r, θ) parameter value bring into before motion simple model R _i(t)=r _isin (2 π ft+ θ _i) in revised matrix H _{x × Y}in each curve point calculate, the curve point meeting this parameter drag is added up, find out local maximum point in totalizer g, the range value that each local maximum point is corresponding in totalizer and phase value are the amplitude of each bar curve and the concrete numerical value of phase place, due to revised matrix H _{x × Y}row represent frequency direction count, the time orientation that list is shown is counted, ordinate and the horizontal ordinate of image is in matrix image, can in the hope of this in frequency corresponding to frequency direction in the position in ordinate direction according to each point on time-frequency spectrum curve, according to v=f _dλ/2 can draw the instantaneous velocity information in each joint, then according to motion simple model R _i(t)=r _isin (2 π ft+ θ _i) the relative trunk maximal rate in known each joint and amplitude relation be v _{i_max}=2 π fr _i, can in the hope of the amplitude information in each joint.

2., according to high resolving power human body Target Motion Character detection method described in claim 1, it is characterized in that: in described step 2 with much larger than Nyquist sampling frequency to echoed signal S (t) over-sampling.

3. according to high resolving power human body Target Motion Character detection method described in claim 1, it is characterized in that: doing time frequency analysis process to S [n] in described step 2 is: adopt Short-time Fourier STFT (short-time Fourier transform) Time-frequency Analysis, short time discrete Fourier transform is specially

$W_s(n,f)=\underset{m=-\∞}{\overset{\∞}{\mathrm{\Σ}}}s\left[m\right]w[n-m]e^{-j2\mathrm{\πfm}}$

S [n] is signal, and w [n] is window function, and f is frequency, and n is frequency, and m is the frequency of change.

4., according to high resolving power human body Target Motion Character detection method described in claim 1, it is characterized in that: to time-frequency energy distribution matrix W in described step 3 _s(n, f) carries out process optimizing process:

Setting threshold value by matrix W _s(n, f) is converted into the two values matrix H only containing 0 and 1 _{x × Y}, transfer process is for making H _{x × Y}=W _s(n, f), by matrix H _{x × Y}in be more than or equal to T element put 1, the element being less than element sets to 0, and wherein X is that frequency direction is counted, and Y is that time orientation is counted, according to column major order by H _{x × Y}element put into vector _pHin, namely in matrix each element according to it in matrix H _{x × Y}middle line number and columns sort ((columns-1) × X+ line number) according to following formula, put into vectorial p successively according to position _hin, p _hin containing X × Y element, and each element is only 0 or 1,

(i.e. p _h=(0,0 ... 0,1,1 ... 1,0,0 ... 0,1,1 ... 1,0,0 ...)), then to vectorial p _hask difference, obtain vectorial p (in p element should for following form p=(0,0 ... 0,1,0 ... 0 ,-1,0 ...)), in p be the element of 1 position correspond to matrix H _{x × Y}in position be the coboundary of time-frequency spectrum curve, for the position correspondence of the element of-1 is the lower boundary of time-frequency spectrum curve, the Shi Fei border, position of 0 element in p; After finding out the position of in p 1 and-1, boundary element can be found out in matrix H according to counter the pushing away of aforementioned row priority ordering _{x × Y}in position, the position of element non-zero in p is recorded in vectorial d successively _nin, so that next step carries out the merging of edge, concrete is exactly being averaging every a pair up-and-down boundary, and obtain the central value of often pair of up-and-down boundary, this value is exactly the position of time-frequency spectrum curve point; By two values matrix H _{x × Y}reset and assignment again, determine each curve point position in a matrix according to the method for row priority ordering, be i.e. the line number of this point and columns, position corresponding in matrix is put 1, obtains curve point matrix:

H _{x × Y}middle each point is exactly the frequency corresponding to each joint of this time point n, and wherein each bar curve approaches the rate signal in each joint.

5. according to high resolving power human body Target Motion Character detection method described in claim 1, it is characterized in that: straight-line detection is carried out to curve point specifically the adopting Hough transform method of described step 4.

6. according to high resolving power human body Target Motion Character detection method described in claim 1, it is characterized in that: curve detection is carried out to curve point matrix specifically the adopting parameters space to add up the method for estimated parameter of described step 5.