CN103593651A - Method for identifying identities of underground coal mine workers based on gaits and two-dimensional discriminant analysis - Google Patents

Abstract

The invention discloses a method for identifying identities of underground coal mine workers based on gaits and two-dimensional discriminant analysis. The method includes the steps that firstly, gait images of the underground coal mine workers are preprocessed, then, a two-dimensional discriminant analysis algorithm is utilized to conduct dimensionality reduction on gait energy images, all recognition features of the training gait energy images are extracted and input to a template database, the recognition features of each gait energy image correspond to identity information, prestored in the template database, of the underground coal mine workers, and eventually a nearest neighbor classifier is utilized to conduct identity recognition on the underground coal mine workers. The method that the two-dimensional discriminant analysis algorithm is utilized to conduct dimensionality reduction on the gait energy images is adopted, internal distribution and geometric structures of a gait image data set can be revealed, much identification information can be acquired, high fault-tolerant capability is possessed for illumination change generated when the gait images are shot, displacement rotation between the gait images and the like, the recognition speed is high, the recognition effect is stable, and practicality is high.

Description

Coal mine down-hole personnel authentication identifying method based on gait and two-dimentional discriminatory analysis

Technical field

The invention belongs to human body biological characteristics recognition technology field, be specifically related to gait recognition method, particularly the coal mine down-hole personnel personal identification method based on gait and two-dimentional discriminatory analysis.

Background technology

Mine safety is related to the people's the security of the lives and property, is related to reform and development and social stability overall situation.Although at present most of coal mining enterprise of a lot of provinces and cities has been equipped with video monitoring system and based on personnel in the pit's attendance recorders such as fingerprint, gait, people's face, iris or vena metacarpeas.But because colliery subsurface environment is severe, video light photograph is inhomogeneous etc., make people's face, fingerprint etc. image blurring and lack color contrast information, the reasons such as target is similar to background, make the existing identification system based on people's face, fingerprint and iris etc. substantially all be installed on aboveground or well head place, just play work attendance effect on and off duty, can not really monitor in real time and differentiate personnel in the pit's activity and identity.Gait Recognition is a kind of emerging biometrics identification technology, is intended to from identical walking behavior, find and extract the variation characteristic between individuality, to realize the automatic identification of human body.Because Gait Recognition has the unique advantage that other biological identification technology does not have, i.e. identification potentiality in remote or low video quality situation, less demanding to image resolution ratio, and gait is difficult to hide or camouflage, affected by environment less, and be a untouchable far distance identity identifying method, so the authentication identifying method based on gait is particularly suitable for coal mine down-hole personnel real-time video monitoring.

Gait Recognition is a kind of newer biological identification technology that increasing researcher pays close attention in recent years.In fact the gait sequence of video images that the automatic Gait Recognition system of an intelligent video monitoring is mainly packaged by CCTV camera, computing machine and is processed with the software of identification and is formed.Wherein, the software algorithm of Gait Recognition is most critical.For the existing Coupling Metric learning method based on linear transformation, can run into when the solving practical problems dimension disaster and cannot fine description nonlinear model etc. problem, the people such as Wang Kejun are by introducing kernel method, a kind of core Coupling Metric learning method has been proposed, and be applied to [Wang Kejun in Gait Recognition, Yan Tao. core Coupling Metric learning method and the application in Gait Recognition thereof. pattern-recognition and artificial intelligence, 2013, Vol.26 (2): 169-175]; The people such as L.Lee analyze gait [Lee L with the moment characteristics of gait profile each several part, Grimson W E L.Gait Analysis for Recognition and Classification.Proceedings of the IEEE International Conference on Automatic Face and Gesture Recognition, 2002]; The University of Southampton of Britain has been obtained a lot of achievements in research [Zhang Rong aspect Gait Recognition, Vogler C, Metaxas D.Human gait recognition.Computer Vision and Pattern Recognition, 2004,27 (02): 18-20].Although have at present a lot of gait recognition methods and technology, but complicacy and instability due to gait image, gait recognition method and technology are not also effectively applied in actual coal mine down-hole personnel identity identification system, prior art exist to the pre-service of original gait image have relatively high expectations, operand is large, system gait image data too complicated, that collect are easily subject to the defects such as environmental impact such as illumination, position.

Summary of the invention

In order to overcome the shortcoming of above-mentioned prior art, the object of the present invention is to provide the coal mine down-hole personnel personal identification method based on gait and two-dimentional discriminatory analysis, have and extract that the recognition feature speed of gait energygram is fast, identification accuracy is high, recognition effect stable and the advantage such as practical.

In order to achieve the above object, the technical solution adopted in the present invention is:

Coal mine down-hole personnel personal identification method based on gait and two-dimentional discriminatory analysis, specifically comprises the following steps:

The first step, the gait image pre-service of coal mine down-hole personnel: the gait image being photographed by video camera is directly inputted to computing machine, the image that shooting does not have a gait is model as a setting, the gait image sequence being eliminated after background after the gait image photographing and background model are subtracted each other; Then with the rectangle frame that the ratio of height and width is 120:80, come frame to live gait profile, the height of rectangle is the height of body gait image profile, the center of rectangular horizontal direction is the horizontal coordinate of body gait image profile barycenter, this rectangle frame is intercepted out, and be normalized to 120 * 80 sizes by the scaling of 1:1, carry out again binary conversion treatment, obtain the gait image sequence B of binaryzation _t(x, y), determines that mistiming between continuous 2 minimum altitudes when people walks is as a gait cycle, utilizes following formula (1) that the gait image of one-period is synthesized to a width gait energygram through average method,

$G(x,y)=\frac{1}{N}\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{B}_t(x,y)---\left(1\right)$

In formula, N is the length of complete gait cycle sequence, and t is the time, and x, y are two dimensional image plane coordinate;

Second step, utilize two-dimentional Discrimination Analysis Algorithm to carry out Dimensionality Reduction to gait energygram, extract the recognition feature of all training gait energygrams, and by this recognition feature input template database, the recognition feature of each gait energygram is corresponding with the identity information that pre-deposits the coal mine down-hole personnel in template database, and feature extraction is carried out according to the following steps:

If K class n width gait energygram is { G ₁, G ₂..., G _n, classification numbering C ₁, C ₂..., C _krepresent, and establish C _iclass has n _iindividual width image,

Calculate all gait energygrams and C _ithe mean value of class gait energygram is respectively

$\left\{\begin{array}{c}\stackrel{\‾}{G}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{G}_i\\ \stackrel{\‾}{{G_C}_i}=\frac{1}{n_i}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}{G}_j\end{array}\right.---\left(2\right)$

Scatter Matrix S between compute classes _bwith Scatter Matrix S in class _wbe respectively

$\left\{\begin{array}{c}{S}_B=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}P\left(C_i\right)(\stackrel{\‾}{{G_C}_i}-\stackrel{\‾}{G}){(\stackrel{\‾}{{G_C}_i}-\stackrel{\‾}{G})}^T\\ S_W=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}P\left(C_i\right){(G_j-\stackrel{\‾}{{G_C}_i})(G_j-\stackrel{\‾}{{G_C}_i})}^T/n_i\end{array}\right.---\left(3\right)$

In formula, P (C _i) be pattern C _iprior probability, get

By S _band S _wset up objective optimization function

J(A)＝A ^T(S _B-S _W)A (4)

In formula, A is mapping matrix to be asked, and T is matrix transpose computing,

To the S in formula (4) _b-S _wcarry out Eigenvalues Decomposition, calculate (S _b-S _w) d of a=λ a maximum eigenvalue λ ₀, λ ₁, L, λ _d-1(λ ₀>=λ ₁>=L>=λ _d-1) corresponding yojan proper vector a ₁, a ₂..., a _d, wherein λ is eigenwert, the dimension that d is low dimensional feature vector,

By following formula (5) by [a ₁, a ₂..., a _d] be converted to an orthogonal matrix P=[p ₁, p ₂..., p _d],

$p_k=a_k-\underset{i=1}{\overset{k-1}{\mathrm{\Σ}}}\frac{p_i^T{a}_k}{p_i^T{p}_i}{p}_i---\left(5\right)$

By following formula (6) by all gait energygram { G ₁, G ₂..., G _nbe mapped to respectively low n-dimensional subspace n,

G _i→Y _i＝P ^TG _i (6)

By { Y ₁, Y ₂..., Y _nbuild the feature templates of coal mine down-hole personnel identification, by formula (6) by any width gait energygram G to be tested _newbe mapped as low-dimensional characteristic image, i.e. G _new→ Y _new=P ^tg _new;

The 3rd step, use nearest neighbor classifier to carry out coal mine down-hole personnel identification: utilize euclidean distance metric as the distance between any two matrix A, B, d (A, B)=|| A-B||,

By second step, obtain all training gait energygram { G ₁, G ₂..., G _nlow-dimensional mapping matrix be respectively { Y ₁, Y ₂..., Y _n, for any one gait energygram G to be tested _newlow-dimensional mapping matrix Y _new, calculate respectively Y _newwith { Y ₁, Y ₂..., Y _nbetween distance, if

will test gait energygram G _newbe judged to k class, determine thus the identity of coal mine down-hole personnel to be identified.

The present invention has adopted the gait energygram Dimensionality Reduction method of two-dimentional discriminatory analysis, not only can disclose internal distribution and the geometry of gait image data set, and can access more identifying information, light variation when it is taken for gait image simultaneously, the problems such as displacement rotating between gait image have stronger fault-tolerant ability; And owing to directly processing each gait two dimensional image matrix data in gait image data dimension yojan process, so recognition speed is fast, recognition effect is stable, practical.

Accompanying drawing explanation

Accompanying drawing is FB(flow block) of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in detail.

With reference to accompanying drawing, the coal mine down-hole personnel personal identification method based on gait and two-dimentional discriminatory analysis, comprises gait image pre-service, training and three processes of identification, specifically comprises the following steps:

The first step, the gait image pre-service of coal mine down-hole personnel: the gait image being photographed by video camera is directly inputted to computing machine, takes and there is no the image of gait model as a setting.Gait image pre-service comprises the location of gait image, extraction, denoising, gait image normalization and the gait image enhancing etc. in the gait region, center of gait image; Then, the gait image sequence being eliminated after background after subtracting each other by the body gait image photographing with without the background model of human body.With the rectangle frame that the ratio of height and width is 120:80, come frame to live body gait profile.This rectangle frame is intercepted out, and be normalized to 120 * 80 sizes by the scaling of 1:1; Carry out again binary conversion treatment, obtain the gait image sequence B of binaryzation _t(x, y); Determine that mistiming between continuous 2 minimum altitudes when people walks is as a gait cycle.The gait image of one-period is averaged and synthesizes gait energygram, and computing formula is as follows

$G(x,y)=\frac{1}{N}\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{B}_t(x,y)---\left(1\right)$

In formula, N is the length of complete gait cycle sequence, and t is the time, and x, y are two dimensional image plane coordinate.

Gait energygram has comprised many gait information such as phase place, frequency, profile, can reflect gait motion feature, again data volume is reduced to original 1/tens simultaneously.Again all gait energygrams are divided into training set and test set.Before carrying out identification mission, pretreated training sample need to be deposited in to database.In this process, generally need to check the quality of the training sample that deposits database in, guarantee the feature of the high training sample of quality to deposit in database, simultaneously by personal information corresponding to personnel in the pit, as name, No. ID etc. deposit in recognition system.

Second step, utilizes two-dimentional Discrimination Analysis Algorithm to carry out Dimensionality Reduction to gait energygram, extracts mapping matrix and low-dimensional feature masterplate, deposits in masterplate database.The recognition feature of each gait energygram is corresponding with the personal information that pre-deposits the coal mine down-hole personnel in template database.Recognition feature is extracted according to the following steps:

If K class n width gait energygram is { G ₁, G ₂..., G _n, classification numbering C ₁, C ₂..., C _krepresent, and establish C _iclass has n _iwidth energygram,

Calculate all gait energygrams and C _ithe mean value of class gait energygram is respectively

$\left\{\begin{array}{c}\stackrel{\‾}{G}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{G}_i\\ \stackrel{\‾}{{G_C}_i}=\frac{1}{n_i}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}{G}_j\end{array}\right.---\left(2\right)$

Scatter Matrix S between compute classes _bwith Scatter Matrix S in class _wbe respectively

$\left\{\begin{array}{c}{S}_B=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}P\left(C_i\right)(\stackrel{\‾}{{G_C}_i}-\stackrel{\‾}{G}){(\stackrel{\‾}{{G_C}_i}-\stackrel{\‾}{G})}^T\\ S_W=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}P\left(C_i\right){(G_j-\stackrel{\‾}{{G_C}_i})(G_j-\stackrel{\‾}{{G_C}_i})}^T/n_i\end{array}\right.---\left(3\right)$

In formula, P (C _i) be pattern C _iprior probability, get

By S _band S _wsetting up objective optimization function is

J(A)＝A ^T(S _B-S _W)A (4)

In formula, A is mapping matrix to be tried to achieve, and T is matrix transpose computing.

Calculate (S _b-S _w) d of a=λ a maximum eigenvalue λ ₀, λ ₁, L, λ _d-1(λ ₀>=λ ₁>=L>=λ _d-1) characteristic of correspondence vector is a ₁, a ₂..., a _d, wherein λ is eigenwert, the dimension that d is low dimensional feature vector.By [a ₁, a ₂..., a _d] be converted to an orthogonal matrix P=[p ₁, p ₂..., p _d], conversion formula is as follows,

$p_k=a_k-\underset{i=1}{\overset{k-1}{\mathrm{\Σ}}}\frac{p_i^T{a}_k}{p_i^T{p}_i}{p}_i---\left(5\right)$

By P=[p ₁, p ₂..., p _d] by all gait energygram { G in training set ₁, G ₂..., G _nbe mapped to respectively low n-dimensional subspace n, obtain { Y ₁, Y ₂..., Y _n, i.e. G _i→ Y _i=P ^tg _i.By { Y ₁, Y ₂..., Y _nbuild the feature templates of coal mine down-hole personnel identification.The low-dimensional feature masterplate of each gait training gait image collection and the personally identifiable information of the coal mine down-hole personnel previously having deposited in are as corresponding in name, No. ID etc.

By concentrated any width gait energygram G to be tested _newbe mapped as low-dimensional characteristic image,

G _new→Y _new＝P ^TG _new (6)

The 3rd step: use nearest neighbor classifier to carry out coal mine down-hole personnel identification: utilize euclidean distance metric as the distance between any two matrix A, B, d (A, B)=|| A-B||.By the low-dimensional characteristic Y of personnel's to be identified gait energygram _newwith the gait feature template { Y in database ₁, Y ₂..., Y _ncarry out the search coupling of " one-to-many ", find out Y _newwith { Y ₁, Y ₂..., Y _nbetween the label of distance minimum, draw thus the conclusion of " whom this person is ".

Claims (1)

1. the coal mine down-hole personnel personal identification method based on gait and two-dimentional discriminatory analysis, is characterized in that, comprises the following steps:

The first step, the gait image pre-service of coal mine down-hole personnel: the gait image being photographed by video camera is directly inputted to computing machine, the image that shooting does not have a gait is model as a setting, the gait image sequence being eliminated after background after the gait image photographing and background model are subtracted each other; Then with the rectangle frame that the ratio of height and width is 120:80, come frame to live gait profile, the height of rectangle is the height of body gait image profile, the center of rectangular horizontal direction is the horizontal coordinate of body gait image profile barycenter, this rectangle frame is intercepted out, and be normalized to 120 * 80 sizes by the scaling of 1:1, carry out again binary conversion treatment, obtain the gait image sequence B of binaryzation _t(x, y), determines that mistiming between continuous 2 minimum altitudes when people walks is as a gait cycle, utilizes following formula (1) that the gait image of one-period is synthesized to a width gait energygram through average method,

$G(x,y)=\frac{1}{N}\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{B}_t(x,y)---\left(1\right)$

In formula, N is the length of complete gait cycle sequence, and t is the time, and x, y are two dimensional image plane coordinate;

Second step, utilize two-dimentional Discrimination Analysis Algorithm to carry out Dimensionality Reduction to gait energygram, extract the recognition feature of all training gait energygrams, and by this recognition feature input template database, the recognition feature of each gait energygram is corresponding with the identity information that pre-deposits the coal mine down-hole personnel in template database, and feature extraction is carried out according to the following steps:

If K class n width gait energygram is { G ₁, G ₂..., G _n, classification numbering C ₁, C ₂..., C _krepresent, and establish C _iclass has n _iindividual width image,

Calculate all gait energygrams and C _ithe mean value of class gait energygram is respectively

$\left\{\begin{array}{c}\stackrel{\‾}{G}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{G}_i\\ \stackrel{\‾}{{G_C}_i}=\frac{1}{n_i}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}{G}_j\end{array}\right.---\left(2\right)$

Scatter Matrix S between compute classes _bwith Scatter Matrix S in class _wbe respectively

$\left\{\begin{array}{c}{S}_B=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}P\left(C_i\right)(\stackrel{\‾}{{G_C}_i}-\stackrel{\‾}{G}){(\stackrel{\‾}{{G_C}_i}-\stackrel{\‾}{G})}^T\\ S_W=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}P\left(C_i\right){(G_j-\stackrel{\‾}{{G_C}_i})(G_j-\stackrel{\‾}{{G_C}_i})}^T/n_i\end{array}\right.---\left(3\right)$

In formula, P (C _i) be pattern C _iprior probability, get

By S _band S _wset up objective optimization function

J(A)＝A ^T(S _B-S _W)A (4)

In formula, A is mapping matrix to be asked, and T is matrix transpose computing,

To the S in formula (4) _b-S _wcarry out Eigenvalues Decomposition, calculate (S _b-S _w) d of a=λ a maximum eigenvalue λ ₀, λ ₁, L, λ _d-1(λ ₀>=λ ₁>=L>=λ _d-1) corresponding yojan proper vector a ₁, a ₂..., a _d, wherein λ is eigenwert, the dimension that d is low dimensional feature vector,

By following formula (5) by [a ₁, a ₂..., a _d] be converted to an orthogonal matrix P=[p ₁, p ₂..., p _d],

$p_k=a_k-\underset{i=1}{\overset{k-1}{\mathrm{\Σ}}}\frac{p_i^T{a}_k}{p_i^T{p}_i}{p}_i---\left(5\right)$

By following formula (6) by all gait energygram { G ₁, G ₂..., G _nbe mapped to respectively low n-dimensional subspace n,

G _i→Y _i＝P ^TG _i (6)

By { Y ₁, Y ₂..., Y _nbuild the feature templates of coal mine down-hole personnel identification, by formula (6) by any width gait energygram G to be tested _newbe mapped as low-dimensional characteristic image, i.e. G _new→ Y _new=P ^tg _new;

The 3rd step, use nearest neighbor classifier to carry out coal mine down-hole personnel identification: utilize euclidean distance metric as the distance between any two matrix A, B, d (A, B)=|| A-B||,

By second step, obtain all training gait energygram { G ₁, G ₂..., G _nlow-dimensional mapping matrix be respectively { Y ₁, Y ₂..., Y _n, for any one gait energygram G to be tested _newlow-dimensional mapping matrix Y _new, calculate respectively Y _newwith { Y ₁, Y ₂..., Y _nbetween distance, if will test gait energygram G _newbe judged to k class, determine thus the identity of coal mine down-hole personnel to be identified.

Images