CN106022210A - Vein profile three-dimensional point cloud matching identity identifying method and device - Google Patents

Abstract

The method provides a vein profile three-dimensional point cloud matching identity identifying method and device. A finger vein image is acquired through a vein contour identifying device. The optimal image of a designated region is acquired according to the image quality evaluation. The interference of the surrounding environment is effectively eliminated. The optimal vein image is selected as original data. An image processing method is used to acquire the finger vein contour line of the designated region, and others are set as the background. According to a binocular principle, the finger vein contour line is transformed into a three-dimensional parallax image. The parallax image is converted into a three-dimensional point cloud. A point cloud matching algorithm based on the nearest point of iteration is used for matching to realize identity authentication and identifying. According to the invention, matching is carried out based on all three-dimensional point cloud of the finger vein contour, which improves the matching accuracy and sensitivity of relevant information based on feature points at the present; two-dimensional image misjudgment caused by finger perturbation and unobvious three-dimensional features in the prior art are avoided; and according to a point cloud matching degree threshold, user-defined identifying sensitivity adjustment is realized.

Description

The personal identification method of a kind of vein contoured three-dimensional point cloud matching and device

Technical field

The present invention relates to a kind of biometrics identification technology, particularly to the identity of a kind of vein contoured three-dimensional point cloud matching Recognition methods and device.

Background technology

Existing intravenous methods and device are mainly based upon two dimension vein image and are identified being main.In actual applications, Due to problems such as change, the finger position of illumination condition are inconsistent, finger disturbances, cause the mistake between template and image to be identified Join, thus have impact on the performance of vein identification method dramatically.Solve the root of identification problem based on two dimension vein image This approach is to use three-dimensional to identify.

Existing three-dimensional vein matching process be based on key point information (such as Application No.: 201010508188.2 special Profit application uses SIFT feature vector etc.) three-dimensional feature that forms carries out mating, and it there may be feature calculation cost, spy Levy the problems such as inconspicuous, characteristic information amount is very few, thus affect the Performance And Reliability of three-dimensional vein identification method；Meanwhile, by There is no in existing three-dimensional vein identification method and matching precision is carried out self-defined adjustment and setting, therefore in use lack Certain motility.

Therefore, how to solve the precision of three-dimensional hand vein recognition and motility has become as of current vein identification technology Problem demanding prompt solution.

Summary of the invention

The present invention solves above-mentioned technical problem, it is proposed that the personal identification method of a kind of vein contoured three-dimensional point cloud matching And device.The method utilizes the principle of binocular vision to build based on the embedded or finger vena three-dimensional identification device of computer, Get the initial pictures of finger vena；Optimum collection image is obtained in conjunction with vein image quality evaluation；Use figure the most again As processing the finger vena contour images obtained in area-of-interest；Binocular Vision Principle is used to obtain finger vena profile diagram again The three-dimensional information of picture, is converted into three-dimensional point cloud；The ICP matching algorithm improved is used to realize the three-dimensional identification of finger vena. The method can be effectively improved finger vena accuracy of identification and actually used in motility.

The technical solution used in the present invention is: the personal identification method of a kind of vein contoured three-dimensional point cloud matching, including:

S1, by identify device obtain finger vena initial pictures；

S2, the finger vena initial pictures obtained for step S1, first combine vein image quality evaluation and obtain optimum Collection image；Then image procossing is used to obtain the finger vena contour images in area-of-interest；Use binocular vision again Principle obtains the three-dimensional information of finger vena contour images, is converted into three-dimensional point cloud to be matched；

S3, the three-dimensional point cloud to be matched obtained according to step S2, realize the three-dimensional knowledge of finger vena by ICP matching algorithm Not.

Further, described identification device includes: infrared light supply, infrared fileter, finger groove, light electric shock source switch, a left side Right two video cameras, power supply, power control circuit and external processing apparatus；Described external processing apparatus is: PC or ARM connect Oralia or DSP process plate；

Infrared light supply is arranged at bottom, and two, left and right video camera is arranged at top, the middle finger groove for placing finger, hands Refer to that groove one end is provided with a touch-switch；Two, left and right camera lens has all added infrared fileter, non-red in order to filter Outer light；

Finger groove, infrared light supply and two, left and right video camera are in same level；

Finger is vertical with infrared light supply and distance is 1cm, video camera and the vertical dimension about 8cm of finger；

Distance about 3cm between two photographic head in left and right.

Further, described step S2 include following step by step:

S21, the structural similarity of employing image content-based describe image quality, particularly as follows: select a width standard drawing As reference, it is right the image gathered and standard picture to be carried out in terms of brightness, contrast and structural similarity three every time Ratio, constructs evaluation model based on these three key element as follows:

S (x, y)=[l (x, y)]^α·[c(x,y)]^β·[s(x,y)]^γ

Wherein, (x, y) is brightness comparison function to l, and а is luminance weights, and (x, y) is contrast weight to c, and β is contrast power Weight, (x, y) is structural similarity comparison function to s, and γ is structural similarity weight, and x represents parametric image, the image that y gathers；

According to S, (x, result of calculation y) compare with first threshold, if (x, y) more than or equal to first threshold, then for S Represent that this collection image can accept and uses, otherwise represent that this picture quality is undesirable, be adjusted PWM ripple, again adopt Collection image, (x, y) value is more than or equal to till first threshold until S；

Two video cameras are demarcated by S22, employing Zhang Zhengyou standardizition, obtain the spin matrix peace between two video cameras The amount of shifting to and their Intrinsic Matrix K；Using Fusiello method to carry out polar curve correction, the optical axis making video camera is parallel, Thus obtain only horizontal displacement difference a pair image and correction after camera projection matrix；

S23, use CLAHE algorithm carry out histogram equalization, use adaptive threshold fuzziness technology, it is thus achieved that vein mesh Mark, then uses Sobel edge edge detective operators to obtain vein profile, then uses 8 neighborhoods to follow the trail of to obtain the monodrome of finger vena Profile, and will to obtain vein profile point be key point；

S24, use SAD algorithm do Stereo matching and generate disparity map, select the translation distance making sad value minimum for sliding The parallax value of window center point, calculates the parallax value of each pixel successively with order the most from top to bottom, final To disparity map；

S25, use principle of triangulation calculate the three-dimensional coordinate of point, obtain three-dimensional point cloud to be matched.

Further, described step S23 uses CLAHE algorithm also to include after carrying out histogram equalization: use intermediate value Filter method filters the noise in the image after equalization, and to the linearity gray scale stretching obtained.

Further, the ICP matching algorithm described in step S3 is the ICP matching algorithm improved, and specifically includes following substep Rapid:

1), calculate masterplate point cloud and the center of gravity of three-dimensional point cloud to be matched, and be moved into the identical bits under same coordinate Put；

2), calculate the minimum rectangle bounding box of two groups of some clouds, point of rotation cloud make minimum bounding box towards unanimously；

3), obtain two points and converge the Cross-covariance of conjunction；

4), utilize antisymmetric matrix to construct column vector, and generate the symmetrical matrix of a 4*4 dimension by this column vector；

5), calculation procedure 4), the eigenvalue of gained symmetrical matrix and unit character vector；

Wherein, the unit character vector that maximum feature valuation is corresponding is optimal spin matrix；

6), optimal translation vector is calculated；

7), according to step 5) spin matrix that obtains and step 6) the optimal translation vector that obtains, obtain registration state Vector；

8), match cognization object function is built:

$f\left(\stackrel{\→}{q}\right)={\mathrm{\Σ}}_{i=1}^{N_p}\left|\right|x_i-\left(R\right({\stackrel{\→}{q}}_R){\stackrel{\→}{p}}_i+{\stackrel{\→}{q}}_T)|{|}^2$

Wherein,Represent masterplate point cloud, x_iRepresenting some cloud to be matched, R represents spin matrix,Represent optimal spin moment Battle array,Represent optimal translation vector,Represent Euclidean distance square；

9), according to step 7) the registration state vector that obtains, calculation procedure 8) lowest mean square of match cognization object function Error；

10), according to repeatedly coupling measuring Second Threshold, by step 9) least mean-square error that obtains and Second Threshold Compare, if least mean-square error is more than this threshold value, then the some cloud number obtained with the translation vector calculated and spin matrix Replace former to be matched some cloud according to battle array, and jump to step S31, otherwise when least mean-square error is less than or equal to Second Threshold, or Person's iterations, more than when presetting maximum iteration time, stops iteration.

A kind of vein outline identification device, including: infrared light supply, infrared fileter, finger groove, light electric shock source switch, a left side Right two video cameras, power supply, power control circuit and external processing apparatus；Described external processing apparatus is: PC or ARM connect Oralia or DSP process plate；

Infrared light supply is arranged at bottom, and two, left and right video camera is arranged at top, the middle finger groove for placing finger, hands Refer to that groove one end is provided with a touch-switch；Two, left and right camera lens has all added infrared fileter, non-red in order to filter Outer light；

Finger groove, infrared light supply and two, left and right video camera are in same level；

Finger is vertical with infrared light supply and distance is 1cm, video camera and the vertical dimension about 8cm of finger；

Distance about 3cm between two photographic head in left and right.

Beneficial effects of the present invention: the personal identification method of a kind of vein contoured three-dimensional point cloud matching and device, by quiet Arteries and veins outline identification device gets finger venous image；Then the optimum image specifying region is obtained according to image quality evaluation, Can effectively get rid of the interference of surrounding, select the vein image of optimum as initial data；Use image processing method, Obtaining the finger vena contour line specifying region, other are then set to background；According to binocular principle, it is converted into three-dimensional parallax figure Picture；Anaglyph is converted to three-dimensional point cloud；Use point cloud matching algorithm based on iterative closest point to mate, coupling is set Threshold value, it is achieved authentication and identification.The present invention achieves three-dimensional coupling based on vein profile point cloud certification from three-dimensional perspective Identifying, the present invention whole three-dimensional point clouds based on finger vena profile mate, and can be effectively improved and be currently based on characteristic point Correlated information match accuracy and sensitivity, and effectively eliminate the two dimensional image erroneous judgement caused because of finger disturbance and existing The unconspicuous problem of three-dimensional feature, can according to point cloud matching degree threshold value come self-defined adjust identify sensitivity.

Accompanying drawing explanation

The identification structure drawing of device that Fig. 1 provides for the present invention.

The protocol procedures figure that Fig. 2 provides for the present invention.

Fig. 3 detects the mask used by profile for what the present invention provided；

Wherein, (a) is finger coboundary, and (b) is finger lower boundary.

The finger vena three-dimensional identification process figure that Fig. 4 provides for the present invention.

Detailed description of the invention

For ease of skilled artisan understands that the technology contents of the present invention, below in conjunction with the accompanying drawings present invention is entered one Step explaination.

The present invention, by vein outline identification device as shown in Figure 1, obtains finger vena initial pictures, and this device is main Including: infrared light supply (1), infrared fileter (2), finger groove (3), light electric shock source switch (4), two, left and right video camera (5), electricity Source (6), power control circuit (7), PC (or ARM interface board and DSP process plate) (8), wherein two camera horizon are placed At device top, in order to prevent non-infrared light, camera lens adds infrared fileter.Finger groove is placed in the middle of device, at finger The top of groove is placed with touch-switch (4), and bottom of device lays infrared light supply (1).

After finger is in place in finger groove, finger tip props up touch-switch (4), then starting device power supply (6), opens infrared light Source (1) and video camera (5), hardware device is started working.By predetermined imaging parameters, it is thus achieved that initial pictures, special by structure Levying quantitative description picture quality, if meeting requirement, then using this image；PWM ripple is otherwise used to adjust light source voltage, again Gather image, until picture quality meets requirement；The optimum image of acquisition is cut to the image of unified size.

Described infrared light supply is to be arranged into the electric filament lamp of array format or xenon lamp or infrarede emitting diode LED.Electric filament lamp Or other light that xenon lamp sends are video camera (5) gathers when, because having added infrared fileter on camera lens, can be by except infrared Light other light unexpected filter, and electric filament lamp or xenon lamp are common light sources, can reduce be manufactured into for manufacturing the device of the application This.

It is illustrated in figure 2 the solution of the present invention flow chart, the technical scheme is that a kind of vein contoured three-dimensional point cloud The personal identification method of coupling, including:

S1, by identify device obtain finger vena initial pictures；

Described identification device includes: infrared light supply, infrared fileter, finger groove, light electric shock source switch, the shooting of two, left and right Mechanical, electrical source, power control circuit and external processing apparatus；Described external processing apparatus is:, PC or ARM interface board or DSP Process plate；

Wherein, infrared light supply is in bottom, and two, left and right video camera is at top, and centre is for placing the finger groove of finger, finger Groove top has a touch-switch, and finger groove, infrared light supply and two, left and right video camera to be in same level；

Infrared light supply, finger groove and two, left and right video camera in same level, finger vertical with infrared light supply and away from From the vertical dimension about 8cm for 1cm, video camera and finger；

Distance about 3cm between two photographic head in left and right；

Two, left and right camera lens all adds infrared fileter, in order to filter non-infrared light；

S2, the finger vena initial pictures obtained for step S1, first combine vein image quality evaluation and obtain optimum Collection image, can effectively get rid of the interference of surrounding；Then image procossing is used to obtain the finger in area-of-interest Vein contour images；Use Binocular Vision Principle to obtain the three-dimensional information of finger vena contour images again, be converted into three-dimensional Point cloud；Specifically include following step by step:

S21, the structural similarity of employing image content-based describe image quality, i.e. select a width standard picture conduct Reference, contrasts the image gathered every time with standard picture in terms of brightness, contrast and structural similarity three, structure Evaluation model based on these three key element is as follows:

S (x, y)=[l (x, y)]^α·[c(x,y)]^β·[s(x,y)]^γ

Wherein, (x, y) is brightness comparison function to l, and а is luminance weights, and (x, y) is contrast weight to c, and β is contrast power Weight, (x, y) is structural similarity comparison function to s, and γ is structural similarity weight, and x represents parametric image, the image that y gathers.For Being easy to the design of later image Processing Algorithm, the weight of three factors of quality evaluation here is 1, i.e. а=β=γ=1.

$l(x,y)=\frac{2u_x{u}_y+c_1}{u_x^2+u_y^2+c_1},c(x,y)=\frac{2{\mathrm{\σ}}_x{\mathrm{\σ}}_y+c_2}{{\mathrm{\σ}}_x^2+{\mathrm{\σ}}_y^2+c_2},s(x,y)=\frac{{\mathrm{\σ}}_{xy}+c_3}{{\mathrm{\σ}}_x{\mathrm{\σ}}_y+c_2}$

Here,Represent reference picture and gather the mean flow rate of image,Represent reference picture and gather the standard deviation of image,Represent reference picture and gather the covariance of image, c₁、c₂And c₃It is to avoid denominator respectively It it is the minimum constant of zero.

According to S (x, result of calculation y) are closer to 1, illustrate and reference picture quality closer to, then it represents that more can connect It is subject to.Here arrange S (x, threshold value y) is 0.8, then S (x, y) more than or equal to 0.8, represents that this collection image can accept and uses, If (x, y) less than 0.8, then it represents that this picture quality is undesirable, then adjusts PWM ripple to S, and Resurvey image, until adopting The S of collection image and reference picture (x, y) value is more than or equal to 0.8, till.

S22, use Zhang Zhengyou standardizition that two video cameras are demarcated, obtain spin matrix R between two video cameras and Translation vector T and their Intrinsic Matrix K；

$K=\left[\begin{array}{ccc}{\mathrm{\α}}_u& \mathrm{\γ}& u_0\\ 0& {\mathrm{\α}}_v& v_0\\ 0& 0& 1\end{array}\right];$

Wherein, α_u=-fk_u, α_v=-fk_vRepresent the focal length in horizontal and vertical pixel, k_uAnd k_vIt is along u axle axle (level Axle) and the valid pixel number of upper every millimeter of v axle (vertical axis), (u₀,v₀) it is the seat of principal point (optical axis with retinal plane intersection point) Mark, γ is warping factor.

Then camera projection matrix is represented by P=K [R | T].

Using Fusiello method to carry out polar curve correction, the optical axis making video camera is parallel, thus obtains only horizontal displacement Camera projection matrix after a pair image of difference and correction.

S23, use CLAHE algorithm carry out histogram equalization, use adaptive threshold fuzziness technology, it is thus achieved that vein mesh Mark, then uses Sobel edge edge detective operators to obtain vein profile, then uses 8 neighborhoods to follow the trail of to obtain the monodrome of finger vena Profile, and will to obtain vein profile point be key point.Particularly as follows:

First, Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm is used Limit the enhancing amplitude of local contrast by limiting the height of local histogram, thus realize limiting noise and local contrast Crossing of degree strengthens.Detailed process is: original image is divided into less window, calculates the accumulation histogram function in each window. Cutting rectangular histogram is carried out uniform to reach to limit the purpose of enlargement range the part that these cropped by predefined threshold value Be distributed to other parts histogrammic.Equalization is done, with the gray scale of equalization rear hatch central point for making based on this rectangular histogram Centered by point gray value.Then moving window, constantly repeats said process, until each pixel is carried out gray scale change Change.Thereby enhance picture contrast.In actual applications, contrast limits limits value is the number in scope [0,1] Value, this experimental selection contrast is limited to 0.01.

Then, the noise in image is filtered with median filtering method.Medium filtering principle is the gray value of any in image Replace with the Mesophyticum of each point value in a neighborhood of this point, make the actual value that the pixel value of surrounding is close, thus eliminate isolated Noise spot.Method is with two dimension sleiding form, pixel in template is ranked up according to the size of grey scale pixel value, generates dullness Rise for 2-D data sequence.Two dimension median filter be output as g (x, y)=med{f (x-k, y-l), (k, l ∈ W) }, wherein, F (x, y), g (x, image after y) being respectively original image and processing.W be size be the template of 7 × 7.

Furthermore, use gray scale stretching to strengthen image quality.Grey Linear conversion makes less gray space pass through line Sexual relationship expands to bigger gray space.After grey linear transformation, add the dynamic range of pixel, enhance the right of image Degree of ratio.Image is made to become more fully apparent, it is easy to identify.Assuming that original image f (x, tonal range y) is [a, b], linear transformation After, (x, gray space y) expands to [c, d] to image f, and formula is as follows.In the method, original image Normalized Grey Level value Pixel in the range of [0.15,0.9] is scaled up to [0,1].

$g(x,y)=\frac{d-c}{b-a}\[f(x,y)-a\]+c$

Finally, it is respectively adopted adaptive threshold fuzziness technology, it is thus achieved that vein target, uses contours extract and tracking the most again To obtain the monodrome image outline of finger vena.The detection of secondary Sobel edge edge obtains vein profile and finger boundary profile and incites somebody to action It is as key point.

First, by mask detection finger contours as shown in Figure 3

By calculating each x mask in y-direction and the convolution of image, the position that maximum convolution value occurs is border Line.Make the grey scale pixel value outside finger areas equal to 0.Use adaptive thresholding algorithm dividing vein from background.Carry out this step After the bianry image that obtains still have noise region to be extracted as finger vena or background mistakenly, it is therefore desirable to image is carried out Corrode and add up the size in each UNICOM region.The black connected region that area is little is filled by white pixel, little white of area Territory, zone is disallowable.Sobel operator extraction is finally used to go out the profile of finger and vein.Sobel edge edge detector uses one 3 Discrete difference between the row and column in × 3 fields calculates gradient, and wherein, the center pixel of every row or column weights with 2, to provide Smooth effect:

$\begin{array}{c}\mathrm{\Δ}f={\[g_x^2+g_y^2\]}^{\frac{1}{2}}\\ ={\{{\[(z_7+2z_8+z_9)-(z_1+z_2+z_3)\]}^2+\[(z_3+z_6+z_9)-{(z_1+z_4+z_7)}^2\]\}}^{1/2}\end{array}$

In formula, z item represents gray scale.Therefore, if (x, y) place's Δ f >=T (T is adaptive threshold), then in this position Pixel be edge pixel.After extracting profile, the key point used as coupling.

S24, use SAD algorithm do Stereo matching and generate disparity map, select the translation distance making sad value minimum for sliding The parallax value of window center point, calculates the parallax value of each pixel successively with order the most from top to bottom, final To disparity map.

SAD algorithm is used to do Stereo matching and generate disparity map.Under computing formula figure:

$SAD(u,v,d)={\mathrm{\Σ}}_{i=-w}^w{\mathrm{\Σ}}_{j=-w}^w|I_l(x+i,y+j)-I_r(x+i+d,y+j)|$

In formula, I_lAnd I_rRepresenting the gray value of left images vegetarian refreshments respectively, d represents parallax distance, and w is sliding window size, x Sliding window center pixel coordinate is represented with y.Select the parallax value that translation distance is sliding window central point making sad value minimum.With Order the most from top to bottom calculates the parallax value of each pixel successively, finally gives disparity map.

S25, use principle of triangulation calculate the three-dimensional coordinate of point.

Image (hereinafter referred to as " left figure ") midpoint and right video camera that left video camera obtains obtain the image (hereinafter referred to as " right side Figure ") in corresponding point coordinate under camera coordinates system be respectively x and x.The projection matrix of two cameras is respectively P and P '.By We have obtained two equatioies these: x=PX and x '=P ' X, combine them into a linear equation AX=0 about X.Pass through Calculate cross product, three equations of available each point, and the homogeneous scale factor that disappeared.For left figure, formula x × (PX)=0 can Be written to for:

$\left\{\begin{array}{c}x\left(P^{3T}X\right)-\left(P^{1T}X\right)=0\\ y\left(P^{3T}X\right)-\left(P^{2T}X\right)=0\\ x\left(P^{2T}X\right)-y\left(P^{1T}X\right)=0\end{array}\right.,$

Wherein, P^iTIt it is the row vector of P.

For right figure, x and y of above-mentioned equation group can be replaced and obtain the equation of right figure for x ', y '.

Matrix A is:

$A=\left[\begin{array}{c}{\mathrm{xP}}^{3T}-P^{1T}\\ {\mathrm{yP}}^{3T}-P^{2T}\\ x^{\′}{P}^{\′3T}-P^{\′1T}\\ y^{\′}{P}^{\′3T}-P^{\′2T}\end{array}\right]$

X=(x, y, z, 1)^TBeing its equation of n th order n, X is the least square solution of AX=0.Thus obtain the three-dimensional of corresponding point X Coordinate system.

S3, it is illustrated in figure 4 the identification process figure of the present invention.First to be matched some cloud and masterplate point cloud are moved to each Center of gravity, then calculates the proximity pair in 2 clouds, obtains error and coordinate transform vector, coordinate transform vector is substituted into former treating Mate the some cloud after cloud data is converted.Whether whether error in judgement be more than more than Second Threshold and existing iterations Big iterations, calculates closest approach if it is not, then return, carries out next iteration.The most then point cloud data is registrated to masterplate point Cloud, more whether relative error less than Second Threshold, the most then the match is successful；If it is not, then it fails to match.

The present invention whole three-dimensional point clouds based on finger vena profile mate, and can be effectively improved and be currently based on feature The correlated information match accuracy and sensitivity put, and effectively eliminate the two dimensional image erroneous judgement caused because of finger disturbance and show The unconspicuous problem of some three-dimensional features, can carry out the self-defined sensitivity adjusting and identifying according to point cloud matching degree threshold value.

The three-dimensional point cloud obtained according to step S2, realizes the three-dimensional identification of finger vena by the ICP matching algorithm improved. Assume p_i(i=1,2 ... N_p) it is masterplate point cloud, x_i(i=1,2 ... N_x) it is some cloud to be matched, i represents a numbering at cloud midpoint. Use unit quaternion method to calculate kinematic parameter, obtain representing spin matrix with Quaternion MethodWherein q₀>=0 and q₀ ²+q₁ ²+q₂ ²+q₃ ²=1.And translation vectorRotated eventually through calculating matrix A Matrix R.

$R=\left[\begin{array}{ccc}{q_0}^2+{q_1}^2-{q_2}^2-{q_3}^2& 2(q_1{q}_2-q_0{q}_3)& 2(q_1{q}_3+q_0{q}_2)\\ 2(q_1{q}_2+q_0{q}_3)& {q_0}^2+{q_2}^2-{q_1}^2-{q_3}^2& 2(q_2{q}_3-q_0{q}_1)\\ 2(q_1{q}_3-q_0{q}_2)& 2(q_2{q}_3+q_0{q}_1)& {q_0}^2+{q_3}^2-{q_1}^2-{q_2}^2\end{array}\right]$

Registration state vector isThus matching problem is converted into searching functionMinima

$f\left(\stackrel{\→}{q}\right)={\mathrm{\Σ}}_{i=1}^{N_p}\left|\right|x_i-\left(R\right({\stackrel{\→}{q}}_R){\stackrel{\→}{p}}_i+{\stackrel{\→}{q}}_T)|{|}^2---\left(5\right)$

Specifically include following step by step:

1), masterplate point cloud and the center of gravity of to be matched some cloud are calculatedAnd moved Move the same position under same coordinate；

2), calculate the minimum rectangle bounding box of two groups of some clouds, point of rotation cloud make minimum bounding box towards unanimously；

3), obtain two points and converge the Cross-covariance of conjunction；

${\mathrm{\Σ}}_{px}=\frac{1}{N_p}{\mathrm{\Σ}}_{i=1}^{N_p}\[({\stackrel{\→}{p}}_i-{\stackrel{\→}{\mathrm{\μ}}}_p){({\stackrel{\→}{x}}_i-{\stackrel{\→}{\mathrm{\μ}}}_x)}^T\]=\frac{1}{N_p}{\mathrm{\Σ}}_{i=1}^{N_p}\[{\stackrel{\→}{p}}_i{{\stackrel{\→}{x}}_i}^T\]-{\mathrm{\μ}}_x{{\mathrm{\μ}}_p}^T$

4) antisymmetric matrix A, is utilized_ij=(∑_px∑_px ^T)_ijStructure column vector Δ=(A₂₃ A₃₁ A₁₂)^T, and with these row to Amount generates the symmetrical matrix Q (∑ of a 4*4_px)；

$Q\left({\mathrm{\Σ}}_{px}\right)=\left[\begin{array}{cc}tr\left({\mathrm{\Σ}}_{px}\right)& {\mathrm{\Δ}}^T\\ \mathrm{\Δ}& {\mathrm{\Σ}}_{px}{{\mathrm{\Σ}}_{px}}^T-tr\left({\mathrm{\Σ}}_{px}\right)I_3\end{array}\right];$

5), calculation procedure 4) gained symmetrical matrix Q (∑_px) eigenvalue and unit character vector；

Wherein, the unit character vector that maximum feature valuation is corresponding is optimal spin matrix:

6), optimal translation vector is calculated

7), according to step 5) spin matrix that obtains and step 6) the optimal translation vector that obtains, obtain registration state Vector

8), match cognization object function is built:

$f\left(\stackrel{\→}{q}\right)={\mathrm{\Σ}}_{i=1}^{N_p}\left|\right|x_i-\left(R\right({\stackrel{\→}{q}}_R){\stackrel{\→}{p}}_i+{\stackrel{\→}{q}}_T)|{|}^2;$

Wherein,Represent masterplate point cloud, x_iRepresenting some cloud to be matched, R represents spin matrix,Represent optimal spin moment Battle array,Represent optimal translation vector, | | | |²Represent Euclidean distance square.

9), according to step 7) the registration state vector that obtains, calculation procedure 8) match cognization object functionMinimum Mean square error

10), according to repeatedly coupling measuring Second Threshold, i.e. matching threshold τ, Second Threshold can be sensitive according to identify Degree adjusts.If the sensitivity requirement identified is high, then this threshold value is the least；Whereas if the sensitivity requirement identified is low, then This threshold value is the biggest, i.e. by the setting of this thresholding, can adjust the sensitivity of the identification of system.By step 9) minimum that obtains Mean square error d_msCompare with τ, if d_msMore than τ, the then cloud data obtained with the translation vector calculated and spin matrix Battle array replaces former to be matched some cloudLoop iteration is until d_msLess than or equal to τ, or iterations is big In default maximum iteration time.If in default maximum iteration time, it fails to match, then recognition failures, i.e. body are described Part authentication failed；If in default maximum iteration time, the match is successful, and d_msLess than or equal to τ, then explanation identifies successfully, I.e. authentication success.Wherein, maximum iteration time is the number according to a cloud and the precision of identification, real-time and susceptiveness Experiment determines.The setting of maximum iteration time is the requirement of real-time according to match cognization and required precision, by repeatedly real Test and determine.If the requirement of real-time identified is high, then maximum iteration time may be designed as less；If the real-time identified Require low, then maximum iteration time may be designed as more.If the required precision identified is high, then maximum iteration time may be designed as More；If the requirement of real-time identified is low, then maximum iteration time may be designed as less.

Those of ordinary skill in the art it will be appreciated that embodiment described here be to aid in reader understanding this Bright principle, it should be understood that protection scope of the present invention is not limited to such special statement and embodiment.For ability For the technical staff in territory, the present invention can have various modifications and variations.All within the spirit and principles in the present invention, made Any modification, equivalent substitution and improvement etc., within should be included in scope of the presently claimed invention.

Claims (6)

1. the personal identification method of a vein contoured three-dimensional point cloud matching, it is characterised in that including:

S1, by identify device obtain finger vena initial pictures；

S2, the finger vena initial pictures obtained for step S1, first combine vein image quality evaluation and obtain optimum adopting Collection image；Then image procossing is used to obtain the finger vena contour images in area-of-interest；Use Binocular Vision Principle again Obtain the three-dimensional information of finger vena contour images, be converted into three-dimensional point cloud to be matched；

S3, the three-dimensional point cloud to be matched obtained according to step S2, realize the three-dimensional identification of finger vena by ICP matching algorithm.

The personal identification method of a kind of vein contoured three-dimensional point cloud matching the most according to claim 1, it is characterised in that institute State identification device to include: infrared light supply, infrared fileter, finger groove, light electric shock source switch, two, left and right video camera, power supply, electricity Source control circuit and external processing apparatus；Described external processing apparatus is: PC or ARM interface board or DSP process plate；

Infrared light supply is arranged at bottom, and two, left and right video camera is arranged at top, the middle finger groove for placing finger, finger groove One end is provided with a touch-switch；Two, left and right camera lens all adds infrared fileter, in order to filter non-infrared light；

Finger groove, infrared light supply and two, left and right video camera are in same level；

Finger is vertical with infrared light supply and distance is 1cm, video camera and the vertical dimension about 8cm of finger；

Distance about 3cm between two photographic head in left and right.

The personal identification method of a kind of vein contoured three-dimensional point cloud matching the most according to claim 1, it is characterised in that institute State step S2 include following step by step:

S21, the structural similarity of employing image content-based describe image quality, particularly as follows: select a width standard picture to make For reference, the image every time gathered is contrasted in terms of brightness, contrast and structural similarity three with standard picture, structure Make evaluation model based on these three key element as follows:

S (x, y)=[l (x, y)]^α·[c(x,y)]^β·[s(x,y)]^γ

Wherein, (x, y) is brightness comparison function to l, and а is luminance weights, and (x, y) is contrast weight to c, and β is contrast weight, s (x, y) is structural similarity comparison function, and γ is structural similarity weight, and x represents parametric image, the image that y gathers；

According to S (x, result of calculation y) compare with first threshold, if S (x, y) more than or equal to first threshold, then it represents that This collection image can accept and uses, and otherwise represents that this picture quality is undesirable, is adjusted PWM ripple, Resurvey figure Picture, (x, y) value is more than or equal to till first threshold until S；

Two video cameras are demarcated by S22, employing Zhang Zhengyou standardizition, obtain the spin matrix between two video cameras and are translated towards Amount and their Intrinsic Matrix K；Using Fusiello method to carry out polar curve correction, the optical axis making video camera is parallel, thus Obtain only horizontal displacement difference a pair image and correction after camera projection matrix；

S23, use CLAHE algorithm carry out histogram equalization, use adaptive threshold fuzziness technology, it is thus achieved that vein target, so Rear employing Sobel edge edge detective operators obtains vein profile, and uses the Contour tracing of 8 neighborhoods to obtain the monodrome of finger vena Image outline, will obtain vein profile point as key point；

S24, use SAD algorithm do Stereo matching and generate disparity map, and selecting the translation distance making sad value minimum is in sliding window The parallax value of heart point, calculates the parallax value of each pixel successively, finally gives and regard with order the most from top to bottom Difference figure；

S25, use principle of triangulation calculate the three-dimensional coordinate of point, obtain three-dimensional point cloud to be matched.

The personal identification method of a kind of vein contoured three-dimensional point cloud matching the most according to claim 3, it is characterised in that institute Stating step S23 uses CLAHE algorithm also to include after carrying out histogram equalization: the figure after using median filtering method to filter equalization Noise in Xiang, and to the linearity gray scale stretching obtained.

The personal identification method of a kind of vein contoured three-dimensional point cloud matching the most according to claim 1, it is characterised in that step Rapid ICP matching algorithm described in S3 is the ICP matching algorithm improved, specifically include following step by step:

1), calculate masterplate point cloud and the center of gravity of three-dimensional point cloud to be matched, and be moved into the same position under same coordinate；

2), calculate the minimum rectangle bounding box of two groups of some clouds, point of rotation cloud make minimum bounding box towards unanimously；

3), obtain two points and converge the Cross-covariance of conjunction；

4), utilize antisymmetric matrix to construct column vector, and generate the symmetrical matrix of a 4*4 dimension by this column vector；

5), calculation procedure 4), the eigenvalue of gained symmetrical matrix and unit character vector；

Wherein, the unit character vector that maximum feature valuation is corresponding is optimal spin matrix；

6), optimal translation vector is calculated；

7), according to step 5) spin matrix that obtains and step 6) the optimal translation vector that obtains, obtain registrating state vector；

8), match cognization object function is built:

$f\left(\stackrel{\→}{q}\right)={\mathrm{\Σ}}_{i=1}^{N_p}\left|\right|x_i-\left(R\right({\stackrel{\→}{q}}_R){\stackrel{\→}{p}}_i+{\stackrel{\→}{q}}_T)|{|}^2$

Wherein,Represent masterplate point cloud, x_iRepresenting some cloud to be matched, R represents spin matrix,Represent optimal spin matrix,Represent optimal translation vector, | | | |²Represent Euclidean distance square；

9), according to step 7) the registration state vector that obtains, calculation procedure 8) least mean-square error of match cognization object function；

10), according to repeatedly coupling measuring Second Threshold, by step 9) least mean-square error that obtains carries out with Second Threshold Relatively, if least mean-square error is more than this threshold value, then the cloud data battle array obtained with the translation vector calculated and spin matrix Replace former to be matched some cloud, and jump to step S31, otherwise when least mean-square error is less than or equal to Second Threshold, or repeatedly Generation number, more than when presetting maximum iteration time, stops iteration.

6. a vein outline identification device, it is characterised in that including: infrared light supply, infrared fileter, finger groove, light electric shock Source switch, two, left and right video camera, power supply, power control circuit and external processing apparatus；Described external processing apparatus is: PC Machine or ARM interface board or DSP process plate；

Infrared light supply is arranged at bottom, and two, left and right video camera is arranged at top, the middle finger groove for placing finger, finger groove One end is provided with a touch-switch；Two, left and right camera lens all adds infrared fileter, in order to filter non-infrared light；

Finger groove, infrared light supply and two, left and right video camera are in same level；

Finger is vertical with infrared light supply and distance is 1cm, video camera and the vertical dimension about 8cm of finger；

Distance about 3cm between two photographic head in left and right.