CN102750702B - Monocular infrared image depth estimation method based on optimized BP (Back Propagation) neural network model - Google Patents

Abstract

The invention relates to a monocular infrared image depth estimation method based on an optimized BP (Back Propagation) neural network model. The method comprises the following steps of: acquiring a monocular infrared image and a depth map to which the monocular infrared image corresponds; setting at least three feature regions with different scales for pixel points in the monocular infrared image; calculating feature vectors of the feature regions to which the pixel points in the monocular infrared image correspond; screening all the feature vectors by successively using stepwise linear regression and independent component analysis methods to obtain feature vectors conforming to depth information of the infrared image; constructing a depth training sample set by using the obtained feature vectors and the depth map to which the infrared image corresponds, and performing nonlinear fitting on the feature vectors in the set and depth values of the depth map by using a BP neuron network, and optimizing the BP neuron network through a genetic algorithm, and then constructing a depth model; and analyzing the monocular infrared image through the depth model to obtain a depth estimated value. By using the monocular infrared image depth estimation method based on the optimized BP neural network model, the depth information of the infrared image can be relatively accurately estimated.

Description

Monocular infrared image depth estimation method based on Optimized BP Neural Network model

Technical field

The present invention relates to infrared image estimation of Depth technical field, particularly relate to a kind of monocular infrared image depth estimation method based on genetic algorithm optimization BP neural network model.

Background technology

The estimation of Depth of image is from image, to obtain depth distance information, is the problem of a depth perception in essence.The spatial positional information being built by depth perception characterizes is the surperficial relative distance detecting in from observer to scene.Recover the now existing more satisfactory algorithm of depth distance information in coloured image, but for infrared image, because of its reflection be the Temperature Distribution of scene, have the defects such as low signal-to-noise ratio, low contrast, the degree of depth algorithm that recovers this image still belongs to blank.If can recover the depth information of infrared image, the understanding effect of human eye to this image will greatly be improved so.

Picture depth method of estimation is mainly to launch for binocular Depth cue and the estimation of Depth based on image sequence at present, and these two kinds of methods all depend on the feature difference between image.And estimate for monocular depth, comparatively classical in the middle of traditional algorithm is in early days " by shape from shading (shape from shading) ", it is theoretical foundation that this algorithm be take space multistory how much, according to light source, being irradiated to that light and shade that body surface produces changes is that the shade of image recovers Object Depth, but because this algorithm needs priori (as reflection model and light source direction etc.) that the limitation of application is increased.Afterwards, some researchers find the importance of experience gradually, start to utilize the method for machine learning to go to address this problem.The team of the Andrew Ng of Stanford University, by utilizing the model of Markov field training to carry out estimation of Depth to single image, has reached good effect; The Aloysha Efros team of CMU is manual simple classification in Calibration Field scape before training, such as sky, trees, ground and perpendicular line etc., then utilize a large amount of data to learn these classifications, thereby and finally by building Bayesian model, new images is classified and recovered depth information.Although this method is comparatively applicable to the simple picture of a series of scenes, and produces a desired effect, for having in scene, the classification of study is not inaccurate often.

Summary of the invention

Technical matters to be solved by this invention is to provide a kind of monocular infrared image depth estimation method based on Optimized BP Neural Network model, and the method can be estimated the depth information of infrared image comparatively exactly.

The technical solution adopted for the present invention to solve the technical problems is: a kind of monocular infrared image depth estimation method based on Optimized BP Neural Network model is provided, comprises the following steps:

(1) obtain any monocular infrared image I (x, y) and with the corresponding depth map of described monocular infrared image I (x, y);

(2) be described monocular infrared image I (x, y) each pixel in is set the characteristic area at least three different scales, three ascending being respectively of different scale: the first yardstick, the second yardstick and the 3rd yardstick, wherein, the characteristic area of each yardstick at least comprises and is positioned at the image block at center and the image block adjacent with described image block upper and lower, left and right, and i pixel image block that is positioned at center on the first yardstick is i pixel itself; I pixel image block that is positioned at center on the second yardstick comprises all image blocks on the first yardstick; I pixel image block that is positioned at center on the 3rd yardstick comprises all image blocks on the second yardstick; By that analogy;

(3) calculate monocular infrared image I (x, y) proper vector of the corresponding characteristic area of each pixel in, the characteristic component of the proper vector of i pixel at least comprises: the gray-scale value of i pixel all image blocks on the first yardstick, the texture energy of i pixel each image block on other yardsticks except the first yardstick, the gradient energy of different directions and the average of all gradient energy and the variance of i pixel each image block on other yardsticks except the first yardstick, and the sharpness of i pixel each image block on other yardsticks except the first yardstick,

(4) to all proper vectors of obtaining, utilize successively the method for progressively linear regression analysis and independent component analysis to screen, obtain meeting the proper vector of infrared image depth information;

(5) utilize screening in step (4) to obtain proper vector and described monocular infrared image I (x, y) corresponding depth map builds the set of degree of depth training sample, the depth value of the proper vector in the set of degree of depth training sample and depth map is carried out to nonlinear fitting by BP neural network, and by genetic algorithm, the initial weight of BP neural network and threshold value are optimized, and then build depth model;

(6) the depth model analysis new monocular infrared image collecting being obtained by structure obtains estimation of Depth value.

Texture energy in described step (3) calculates by Louth mask, and concrete steps are: adopt N basic two-dimentional Louth mask, be designated as M ₁..., M _n, described monocular infrared image I (x, y) is done to convolution with each two-dimentional Louth mask, the value after monocular infrared image I (x, y) and k two-dimentional Louth mask convolution is: T _k(x, y)=I (x, y) * M _k, k=1 ..., N, i pixel m image block N on j yardstick ⁱ(m) texture energy obtaining after monocular infrared image I (x, y) and k two-dimentional Louth mask convolution is $E{n}_{N^i\left(m\right)}^j,$ $E{n}_{N^i\left(m\right)}^j=\underset{x,y\∈N^i\left(m\right)}{\mathrm{\Σ}}\left|T_k(x,y)\right|.$

In described step (3), the gradient energy of different directions of i pixel each image block on the second yardstick and the 3rd yardstick and the average of all gradient energy and variance calculate by following steps, concrete steps are: monocular infrared image I (x, y) is tried to achieve respectively on x direction of principal axis and y direction of principal axis to x axial gradient figure I _gradx(x, y) and y axial gradient figure I _grady(x, y), monocular infrared image I (x, y) is at angle θ _lgrad G in direction _l(x, y)=I _gradx(x, y) * cos (θ _l)+I _grady(x, y) * sin (θ _l), wherein, l=0 ..., 7, calculate subsequently the gradient energy of the different directions of each image block of each pixel on other yardsticks except the first yardstick, wherein, i pixel m image block N on j yardstick ⁱ(m) at angle θ _lgradient energy in direction is i pixel m image block N on j yardstick ⁱ(m) at angle θ _lthe average of the gradient energy in direction i pixel m image block N on j yardstick ⁱ(m) at angle θ _lgradient energy variance in direction wherein, size is image block N ⁱ(m) number of the pixel comprising,

In described step (3), the sharpness of i pixel each image block on the second yardstick and the 3rd yardstick calculates by following steps, and concrete steps are: i pixel m image block N on j yardstick ⁱ(m) sharpness wherein, I (x, y) is image block N ⁱ(m) contain the gray scale of pixel, for image block N ⁱ(m) average of the gray scale of contained pixel, size is image block N ⁱ(m) number of the pixel comprising.

In described step (4) progressively linear regression analysis comprise following sub-step:

(411) related coefficient of each characteristic component and depth value in calculated characteristics vector, obtains the sequence of each characteristic component to effect of depth degree according to the absolute value of related coefficient is descending;

(412) from the characteristic component of the absolute value maximum of related coefficient, start progressively to introduce regression equation, and do regression equation significance test, if significantly do not think, selected whole characteristic component is not all the principal element that affects depth value, if significantly again from depth value is affected to the descending regression equation of introducing one by one successively;

(413) new characteristic component of every introducing all needs each characteristic component contained in regression equation to carry out significance test, by that in new regression equation not significantly and depth value is affected to minimum characteristic component rejection, repeat this step until each characteristic component in regression equation is remarkable;

(414) introduce again the characteristic component in the characteristic component of not introducing, depth value being had the greatest impact, repeating step (413) and step (414), until cannot reject selected characteristic component, till also cannot introducing new characteristic component.

What independent component analysis was used in described step (4) is Fast ICA algorithm, by Fast ICA algorithm, to analyzing through the proper vector of progressively linear regression analysis, makes between each component independently as much as possible, comprises following sub-step:

(421) specifying the proper vector through M pixel after linear regression analysis is progressively observation data X, and observation data X is carried out to centralization, and making it average is 0;

(422), by the observation data X albefaction of centralization, be about to after observation data X projects to new subspace become albefaction vector Z, Z=W ₀x, wherein, W ₀for albefaction matrix, W ₀=Λ ^-1/2u ^t, Λ is the eigenvalue matrix of the covariance matrix of observation data X, U is the eigenvectors matrix of the covariance matrix of observation data X;

(423) upgrade W ^*make W ^*=E{Zg (W ^tz) }-E{g ' (W ^tz}W, wherein g () is nonlinear function, and then to W ^*standardization: W=W ^*/ || W ^*||, if do not restrain, repeat this step, wherein, select initial random weight vector that a mould is 1 as the initial value of W.

Initial weight and the threshold value of in described step (5), utilizing genetic algorithm to be optimized BP neural network also comprise following sub-step:

(511) according to training sample, build initial BP neural network, and determine the number of input layer, hidden layer and the output layer of this network;

(512) carry out the initialization of Population in Genetic Algorithms, wherein each individuality of population is to consist of the threshold value of the weights between the weights between input layer and hidden layer, hidden layer and output layer, hidden layer and the threshold value of output layer four parts, and all weights and threshold value are listed as into a real number vector;

(513) maximum iteration time is set as stopping criterion for iteration, iterative process each time comprises the operation of selection, crossover and mutation, when reaching stopping criterion for iteration, output the last reign of a dynasty population optimum individual be the approximate optimal solution of initial weight and threshold value.

Beneficial effect

Owing to having adopted above-mentioned technical scheme, the present invention compared with prior art, there is following advantage and good effect: the present invention utilizes infrared image " spatial context " and " multiple dimensioned " information extraction depth characteristic vector, by the method for progressively linear regression analysis and independent component analysis, the feature of extracting is screened successively, be conducive to find the depth characteristic that is more applicable to infrared image, and build degree of depth training set with this; Adopt back propagation learning theory to carry out nonlinear fitting to training set, and be optimized by genetic algorithm, improved the training speed of experiment and the accuracy rate of matching.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention.

Embodiment

Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiment are only not used in and limit the scope of the invention for the present invention is described.In addition should be understood that those skilled in the art can make various changes or modifications the present invention after having read the content of the present invention's instruction, these equivalent form of values fall within the application's appended claims limited range equally.

As shown in Figure 1, the present embodiment discloses a kind of monocular infrared image depth estimation method of the BP neural network model based on genetic algorithm optimization, the steps include:

Step 1, obtain any monocular infrared image I (x, y) and with the corresponding depth map of this monocular infrared image I (x, y);

Step 2, for monocular infrared image I (x, y) each pixel in is set the characteristic area on three different scales, three ascending being respectively of different scale: the first yardstick, the second yardstick and the 3rd yardstick, wherein, the characteristic area of each yardstick comprise be positioned at center image block and with this image block on, under, left, right adjacent image block, i pixel image block that is positioned at center on the first yardstick is i pixel itself, the image block that be positioned at center of this i pixel on the second yardstick comprises all image blocks on the first yardstick, the image block that be positioned at center of this i pixel on the 3rd yardstick comprises all image blocks on the second yardstick,

Step 3, calculate monocular infrared image I (x, y) proper vector of the corresponding characteristic area of each pixel in, the characteristic component of the proper vector of i pixel at least comprises: the gray-scale value of i pixel all image blocks on the first yardstick, the texture energy of i pixel each image block on the second yardstick and the 3rd yardstick, gradient energy in 8 directions of i pixel each image block on the second yardstick and the 3rd yardstick and average and the variance of all gradient energy, and the sharpness of i pixel each image block on the second yardstick and the 3rd yardstick,

The texture energy of i pixel each image block on the second yardstick and the 3rd yardstick calculates by following steps:

Adopt 9 basic two-dimentional Louth masks, be designated as M ₁..., M ₉, monocular infrared image I (x, y) is done to convolution with each two-dimentional Louth mask, the value after monocular infrared image I (x, y) and k two-dimentional Louth mask convolution is: T _k(x, y)=I (x, y) * M _k, k=1 ..., 9, i pixel m image block N on j yardstick ⁱ(m) texture energy obtaining after monocular infrared image I (x, y) and k two-dimentional Louth mask convolution is $E{n}_{N^i\left(m\right)}^j,$ $E{n}_{N^i\left(m\right)}^j=\underset{x,y\∈N^i\left(m\right)}{\mathrm{\Σ}}\left|T_k(x,y)\right|,j=\mathrm{2,3}.$

The gradient energy of different directions of i pixel each image block on the second yardstick and the 3rd yardstick and the average of all gradient energy and variance calculate by following steps:

Monocular infrared image I (x, y) is tried to achieve respectively on x direction of principal axis and y direction of principal axis to x axial gradient figure I _gradx(x, y) and y axial gradient figure I _grady(x, y), monocular infrared image I (x, y) is at angle θ _lgrad G in direction _l(x, y)=I _gradx(x, y) * cos (θ _l)+I _grady(x, y) * sin (θ _l), wherein, l=0 ..., 7, calculate subsequently the gradient energy of the different directions of each image block of each pixel on other yardsticks except the first yardstick, wherein, i pixel m image block N on j yardstick ⁱ(m) at angle θ _lgradient energy in direction is i pixel m image block N on j yardstick ⁱ(m) at angle θ _lthe average of the gradient energy in direction i pixel m image block N on j yardstick ⁱ(m) at angle θ _lgradient energy variance in direction wherein, size is image block N ⁱ(m) number of the pixel comprising,

The sharpness of i pixel each image block on the second yardstick and the 3rd yardstick calculates by following steps:

I pixel m image block N on j yardstick ⁱ(m) sharpness wherein, I (x, y) is image block N ⁱ(m) contain the gray scale of pixel, for image block N ⁱ(m) average of the gray scale of contained pixel,

Step 4, to all proper vectors of having obtained in step 3, utilize successively the method for progressively linear regression analysis and independent component analysis to screen, obtain comparatively meeting the proper vector of infrared image depth information.

In step 4, progressively the concrete steps of linear regression analysis are:

The related coefficient of each characteristic component and depth value in step 4.1.1, calculated characteristics vector, obtains the sequence of each characteristic component to effect of depth degree according to the absolute value of related coefficient is descending;

Step 4.1.2, from the characteristic component of the absolute value maximum of related coefficient, start progressively to introduce regression equation, and do regression equation significance test, if significantly can not think, selected whole characteristic component is not all the principal element that affects depth value, if significantly again from depth value is affected to the descending regression equation of introducing one by one successively;

Step 4.1.3, new characteristic component of every introducing all need each characteristic component contained in regression equation to carry out significance test, by that in new regression equation not significantly and depth value is affected to minimum characteristic component rejection, repeat this step until each characteristic component in regression equation is remarkable;

Step 4.1.4, introduce the characteristic component in the characteristic component of not introducing, depth value being had the greatest impact again, repeating step 4.1.3 and step 4.1.4, until cannot reject selected characteristic component, till also cannot introducing new characteristic component.

What in step 4, independent component analysis was used is Fast ICA algorithm, by Fast ICA algorithm, to analyzing through the proper vector of progressively linear regression analysis, makes between each component independent as much as possible.In the present embodiment, Fast ICA algorithm is usingd negentropy maximum as searching direction, and its concrete steps are as follows:

Step 4.2.1, to specify the proper vector through M pixel after linear regression analysis be progressively observation data X, and observation data X is carried out to centralization, and making it average is 0;

Step 4.2.2, by the observation data X albefaction of centralization, be about to after observation data X will project to new subspace become albefaction vector Z, Z=W ₀x, wherein, W ₀for albefaction matrix, W ₀=Λ ^-1/2u ^t, Λ is the eigenvalue matrix of the covariance matrix of observation data X, U is the eigenvectors matrix of the covariance matrix of observation data X;

Step 4.2.3, renewal W ^*be to solve the separation matrix drawing, the separation matrix W finally drawing ^*and observation data X multiplies each other and can obtain between each component as far as possible independently source data and make W ^*=E{Zg (W ^tz) }-E{g ' (W ^tz) } W, wherein E () is expectation computing, wherein g () is nonlinear function, and then to W ^*standardization: W=W ^*/ || W ^*||, if do not restrain, repeat this step, wherein, select initial random weight vector that a mould is 1 as the initial value of W.

The proper vector of infrared image that step 5, utilization are drawn by step 4 and the depth map of infrared image build degree of depth training sample set { f _i, depth _i, wherein, f _ibe the proper vector of i pixel, i=1 ..., l, f _i∈ χ, depth _ibe i the corresponding depth value of pixel.Then utilize back propagation learning the Theory Construction depth model, and be optimized by genetic algorithm, and then new infrared image is carried out to estimation of Depth.

In step 5, the concrete steps of BP neural metwork training are as follows:

Step 5.1.1, according to proper vector f _iwith depth value depth _idetermine the input layer of BP neural network and the nodes of output layer.Then by reference to formula roughly definite hidden layer node is counted scope, and wherein n is input layer number, and m is output layer nodes, and a is the constant between 0 ~ 10, and l is hidden layer node number.By trial method, determine again the hidden layer node number of square error minimum;

Weights and the threshold value of step 5.1.2, initialization BP neural network, and given maximum iteration time, Study rate parameter η and neuron excitation function wherein a is constant;

The output H of step 5.1.3, calculating hidden layer:

$H_j=f(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{ij}}{x}_i-a_j),j=\mathrm{1,2},...,l$

Wherein l is hidden layer node number, ω _ijfor connecting the weights between input layer and hidden layer, x _ibe the proper vector f mentioning in step 5.1.1 _i, a _jfor the threshold value of each hidden layer node, the neuron excitation function of f (.) for mentioning in step 5.1.2;

Step 5.1.4, calculating prediction output O:

$O_k=\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{H}_j{\mathrm{\ω}}_{\mathrm{jk}}-b_k,k=\mathrm{1,2},...,m$

Wherein m is output layer nodes, ω _jkfor connecting the weights between hidden layer and output layer, b _kthreshold value for each output layer node;

Step 5.1.5, according to prediction output and desired output error of calculation e, and according to error e, upgrade weights and the threshold value of network:

e _k=Y _k-O _k,k=1,...,m

${\mathrm{\ω}}_{\mathrm{ij}}={\mathrm{\ω}}_{\mathrm{ij}}+\mathrm{\η}{H}_j(1-H_j)x\left(i\right)\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{jk}}{e}_k,i=1,...,n;j=1,...,l$

ω _jk=ω _jk+ηH _je _k,j=1,...,l；k=1,...,m

$a_j=a_j+\mathrm{\η}(1-H_j)\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{jk}}{e}_k,j=1,...,l$

b _k=b _k+e _k,k=1,...,m

Wherein n is input layer number, Y _kfor desired output, i.e. depth value in training set;

Step 5.1.6, circulation step 5.1.3 are to step 5.1.5, until finish after reaching maximum iteration time;

Step 5.1.7, new infrared image is obtained to the proper vector of each pixel after by the feature extraction of step 1 and step 2 and screening, each component of proper vector is corresponded to respectively to the input layer of the BP neural network of having trained, by prediction, draw corresponding depth value, just completed the estimation to new infrared image depth value.

The concrete steps that in step 5, BP neural network initial weight and threshold value are optimized by genetic algorithm are as follows:

Step 5.2.1, carry out the initialization of Population in Genetic Algorithms, wherein each individuality of population is to consist of the threshold value of the weights between the weights between input layer and hidden layer, hidden layer and output layer, hidden layer and the threshold value of output layer four parts.All weights and threshold value are listed as into a real number vector;

Step 5.2.2, carry out the selection operation of genetic algorithm, first individuality is acted on to BP neural network, the Error Absolute Value between the prediction output drawing and desired output is as this individual fitness value F.According to each individual adaptive value, select, the selection Probability p of each individual i _ifor:

f _i=k/F _i

$P_i=f_i/\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{f}_i$

F wherein _ifor individual fitness value, the coefficient of k for arranging, is a constant; N is population at individual number, because of F _iwhat reflect is the error between desired output and prediction output, and more the bright individuality of novel is more excellent for error, and selecteed probability is just larger, so by F _iget inverse and become f _iso that the size of probability is proportional with selecting; f _jmeaning and f _iidentical, wherein subscript j represents population individual amount;

Step 5.2.3, carry out the interlace operation of genetic algorithm, need the chromosome intersecting and need the position intersecting to select at random.Suppose k chromosome a _kwith l chromosome a _lj position, intersect, method of operating is as follows:

$\left\{\begin{array}{c}{a}_{\mathrm{kj}}=a_{\mathrm{kj}}(1-b)+a_{\mathrm{lj}}b\\ a_{\mathrm{lj}}=a_{\mathrm{lj}}(1-b)+a_{\mathrm{kj}}b\end{array}\right.$

Wherein b is the random number between [0,1];

Step 5.2.4, carry out the mutation operation of genetic algorithm, need the chromosome of variation and the position of variation to select at random.Suppose the gene a of i j individual position _ijneed to make a variation, method of operating is as follows:

$a_{\mathrm{ij}}=\left\{\begin{array}{c}{a}_{\mathrm{ij}}+(a_{\mathrm{ij}}-a_{\max })*f\left(g\right),r\&GreaterEqual;0.5\\ a_{\mathrm{ij}}+(a_{\min }-a_{\mathrm{ij}})*f\left(g\right),r<0.5\end{array}\right.$

A wherein _maxfor gene a _ijthe upper bound, a _minfor gene a _ijlower bound; r and r ₂be all the random number between [0,1], g is current iteration number of times, G _msxit is maximum iteration time.

Step 5.2.5, circulation step 5.2.2, to step 5.2.4, arrange circulation maximum times, when circulation reaches maximum times, stop, and take out the individuality of fitness value F minimum as initial weight and the threshold value of neural metwork training.

Claims (5)

1. the monocular infrared image depth estimation method based on Optimized BP Neural Network model, is characterized in that, comprises the following steps:

(1) obtain any monocular infrared image I (x, y) and with the corresponding depth map of described monocular infrared image I (x, y);

(2) be described monocular infrared image I (x, y) each pixel in is set the characteristic area on three different scales, three ascending being respectively of different scale: the first yardstick, the second yardstick and the 3rd yardstick, wherein, the characteristic area of each yardstick at least comprises and is positioned at the image block at center and the image block adjacent with described image block upper and lower, left and right, and i pixel image block that is positioned at center on the first yardstick is i pixel itself; I pixel image block that is positioned at center on the second yardstick comprises all image blocks on the first yardstick; I pixel image block that is positioned at center on the 3rd yardstick comprises all image blocks on the second yardstick;

(3) calculate monocular infrared image I (x, y) proper vector of the corresponding characteristic area of each pixel in, the characteristic component of the proper vector of i pixel at least comprises: the gray-scale value of i pixel all image blocks on the first yardstick, the texture energy of i pixel each image block on other yardsticks except the first yardstick, the gradient energy of different directions and the average of all gradient energy and the variance of i pixel each image block on other yardsticks except the first yardstick, and the sharpness of i pixel each image block on other yardsticks except the first yardstick, wherein, the gradient energy of different directions of i pixel each image block on the second yardstick and the 3rd yardstick and the average of all gradient energy and variance calculate by following steps, concrete steps are: monocular infrared image I (x, y) is tried to achieve respectively on x direction of principal axis and y direction of principal axis to x axial gradient figure I _gradx(x, y) and y axial gradient figure I _grady(x, y), monocular infrared image I (x, y) is at angle θ _lgrad G in direction _l(x, y)=I _gradx(x, y) * cos (θ _l)+I _grady(x, y) * sin (θ _l), wherein, calculate subsequently the gradient energy of the different directions of each image block of each pixel on other yardsticks except the first yardstick, wherein, i pixel m image block N on j yardstick ⁱ(m) at angle θ _lgradient energy in direction is i pixel m image block N on j yardstick ⁱ(m) at angle θ _lthe average of the gradient energy in direction i pixel m image block N on j yardstick ⁱ(m) at angle θ _lgradient energy variance in direction wherein, size is image block N ⁱ(m) number of the pixel comprising, the sharpness of i pixel each image block on the second yardstick and the 3rd yardstick calculates by following steps, and concrete steps are: i pixel m image block N on j yardstick ⁱ(m) sharpness wherein, I (x, y) is image block N ⁱ(m) contain the gray scale of pixel, for image block N ⁱ(m) average of the gray scale of contained pixel, size is image block N ⁱ(m) number of the pixel comprising,

(4) to all proper vectors of obtaining, utilize successively the method for progressively linear regression analysis and independent component analysis to screen, obtain meeting the proper vector of infrared image depth information;

(5) utilize screening in step (4) to obtain proper vector and described monocular infrared image I (x, y) corresponding depth map builds the set of degree of depth training sample, the depth value of the proper vector in the set of degree of depth training sample and depth map is carried out to nonlinear fitting by BP neural network, and by genetic algorithm, the initial weight of BP neural network and threshold value are optimized, and then build depth model;

(6) the depth model analysis new monocular infrared image collecting being obtained by structure obtains estimation of Depth value.

2. the monocular infrared image depth estimation method based on Optimized BP Neural Network model according to claim 1, it is characterized in that, texture energy in described step (3) calculates by Louth mask, and concrete steps are: adopt N basic two-dimentional Louth mask, be designated as M ₁..., M _n, described monocular infrared image I (x, y) is done to convolution with each two-dimentional Louth mask, the value after monocular infrared image I (x, y) and k two-dimentional Louth mask convolution is: T _k(x, y)=I (x, y) * M _k, k=1 ..., N, i pixel m image block N on j yardstick ⁱ(m) texture energy obtaining after monocular infrared image I (x, y) and k two-dimentional Louth mask convolution is

${\mathrm{En}}_{N^i\left(m\right)}^j=\underset{x,y\&Element;N^i\left(m\right)}{\mathrm{\&Sigma;}}\left|T_k(x,y)\right|.$

3. the monocular infrared image depth estimation method based on Optimized BP Neural Network model according to claim 1, is characterized in that, in described step (4) progressively linear regression analysis comprise following sub-step:

(411) related coefficient of each characteristic component and depth value in calculated characteristics vector, obtains the sequence of each characteristic component to effect of depth degree according to the absolute value of related coefficient is descending;

(412) from the characteristic component of the absolute value maximum of related coefficient, start progressively to introduce regression equation, and do regression equation significance test, if significantly do not think, selected whole characteristic component is not all the principal element that affects depth value, if significantly again from depth value is affected to the descending regression equation of introducing one by one successively;

(413) new characteristic component of every introducing all needs each characteristic component contained in regression equation to carry out significance test, by that in new regression equation not significantly and depth value is affected to minimum characteristic component rejection, repeat this step until each characteristic component in regression equation is remarkable;

(414) introduce again the characteristic component in the characteristic component of not introducing, depth value being had the greatest impact, repeating step (413) and step (414), until cannot reject selected characteristic component, till also cannot introducing new characteristic component.

4. the monocular infrared image depth estimation method based on Optimized BP Neural Network model according to claim 1, it is characterized in that, what in described step (4), independent component analysis was used is Fast ICA algorithm, by Fast ICA algorithm to analyzing through the proper vector of progressively linear regression analysis, make between each component independently as much as possible, comprise following sub-step:

(421) specifying the proper vector through M pixel after linear regression analysis is progressively observation data X, and observation data X is carried out to centralization, and making it average is 0;

(422), by the observation data X albefaction of centralization, be about to after observation data X projects to new subspace become albefaction vector Z, Z=W ₀x, wherein, W ₀for albefaction matrix, W ₀=Λ ^-1/2u ^t, Λ is the eigenvalue matrix of the covariance matrix of observation data X, U is the eigenvectors matrix of the covariance matrix of observation data X;

(423) upgrade W ^*make W ^*=E{Zg (W ^tz) }-E{g'(W ^tz) } W, wherein g () is nonlinear function, and then to W ^*standardization: W=W ^*/ || W ^*||, if do not restrain, repeat this step, wherein, select initial random weight vector that a mould is 1 as the initial value of W.

5. the monocular infrared image depth estimation method based on Optimized BP Neural Network model according to claim 1, it is characterized in that, initial weight and the threshold value of in described step (5), utilizing genetic algorithm to be optimized BP neural network also comprise following sub-step:

(511) according to training sample, build initial BP neural network, and determine the number of input layer, hidden layer and the output layer of this network;

(512) carry out the initialization of Population in Genetic Algorithms, wherein each individuality of population is to consist of the threshold value of the weights between the weights between input layer and hidden layer, hidden layer and output layer, hidden layer and the threshold value of output layer four parts, and all weights and threshold value are listed as into a real number vector;

(513) maximum iteration time is set as stopping criterion for iteration, iterative process each time comprises the operation of selection, crossover and mutation, when reaching stopping criterion for iteration, output the last reign of a dynasty population optimum individual be the approximate optimal solution of initial weight and threshold value.