CN102663357A - Color characteristic-based detection algorithm for stall at parking lot - Google Patents

Abstract

The invention discloses a color characteristic-based detection algorithm for a stall at a parking lot. A CCD camera is set up at a large-scale parking lot and is used to collect stall image information in real time; the collected stall image data are read in by a computer system and then preprocessing including to-be-detected stall interception and color image smooth filtering and the like is carried out on the image data successively; and an established background mixed Gaussian model is utilized to carry out color characteristic extraction. In order to determine whether there is a car parked at a stall, salient characteristics like a variance, an edge and an angular point and the like are extracted and are used to form a characteristic space of each image; and statistics is carried out on the characteristics and the processed characteristics are input into a classifier that has been trained by utilizing a boundary sample, so that an occupation situation of the stall is determined. The provided algorithm, which has high versatility and a wide application range, can be applied to various indoor and outdoor parking lot environments. Besides, the algorithm has advantages of convenient installation, low cost, good real-time performance and high detection precision and the like.

Description

Parking position detection algorithm based on color characteristic

Technical field

The present invention relates to pattern-recognition, Statistical Learning Theory, image processing field, designed and Implemented and a kind of the indoor and outdoor parking position has been taken the universal method that situation is monitored in real time and detected.

Background technology

In recent years, along with the high speed development of economic society, China's urban automobile quantity increases sharply, and the parking lot construction is slow relatively, and the parking difficulty problem becomes increasingly conspicuous.The research method for detecting parking stalls can effectively solve the problem of parking stall resource-constrained; Improve the utilization rate of parking position; Satisfied the parking lot in efficient, safety and managerial requirement, this will be to the research and development of China's present stage intelligent transportation, rationally efficient utilization the in parking lot played positive impetus.

At present, the parking position detection method has a lot, mainly can be divided into based on the detection method of physical features with based on the detection method of video monitoring, computer vision and image processing techniques.Detection method based on physical features mainly adopts modes such as buried inductive coil, ultrasound wave, earth magnetism detection to realize.This mode have cost low, receive advantages such as climate effect is little, but the construction trouble is wanted excavated pavement, road pavement damages, and the road surface receives season and vehicle pressure influence, coil damages easily, is difficult to maintenance; Detection mode based on video monitoring, computer vision and image processing techniques has many advantages; At first, the video camera of capture video image is easy for installation, and changing does not influence traffic; The position of adjustment and dollying head need not on carriageway surfacing, to construct easily; Characteristics such as that secondly, video image processing technology can reach is real-time, parking stall measure precision height.

Summary of the invention

The objective of the invention is to propose a kind of method for detecting parking stalls that combines based on color characteristic and pattern-recognition.First color characteristic is incorporated in the detection system of parking lot, in the hope of reaching higher parking stall measure accuracy rate.

The present invention adopts following technological means to realize: analyze parking position image information distribution of color characteristics, statistics empty wagons bit image makes up the background color Gauss model, utilizes model to judge the generic situation of each pixel.But this can only judge this pixel is non-background pixel point; What can't judge whether parking stall place park is car, and some are relevant and have the characteristic of very big differentiation property with the empty wagons position with car so should extract, and select angle point; The edge; Variance is utilized the parking lot image training svm classifier device that collects in a large number, carries out the parking stall and accounts for the detection of pool situation.The concrete steps that the present invention realizes are described below:

(1) adopt the CCD camera to take and obtain the parking stall video data, the shooting area scope of video camera is 1-4 parking stall, and camera relative position and shooting angle remain unchanged;

(2) select a width of cloth not have the background image of car, the frame coordinate of no back frame scape image parking stall to be measured is set, purpose is the view data that intercepting only comprises single parking space information to be measured, and the background image of the parking stall to be measured that obtains is made as I ₀

(3) for every width of cloth testing image, the mode intercepting in (2) goes out the regional extent of concrete parking stall to be measured set by step, carries out following pre-treatment step:

(a) the RGB model is to the HSI model conversion.

$H=\begin{array}{cc}\left\{\begin{array}{cc}\mathrm{\θ}& \\ 360-\mathrm{\θ}& \begin{array}{c}B\≤G\\ B>G\end{array}\end{array}\right.& \mathrm{\θ}=\arccos \left\{\frac{\frac{1}{2}[(R-G)+(R-B)]}{{[{(R-G)}^2+(R-G)(G-B)]}^{1/2}}\right\}\end{array}$

$S=1-\frac{3}{(R+G+B)}\left[\min (R,G,B)\right]$

$I=\frac{1}{3}(R+G+B)$

R wherein, G, B represent the red component under the RGB color model respectively, and blue component, green component, θ are the reference angles of tone; H, what S, I represented respectively is chromatic component under the HSI color model, saturation degree component and strength component.

(b) under the HSI space, because will utilize the colouring information of image, we only carry out filtering to the strength component under the HIS space so, keep the colouring information of image.I component is carried out image enhancement processing.

(c) conversion of HSI color model and RGB model.Image transitions after level and smooth is carried out feature extraction under the RGB model space.Formula is as follows:

G＝I(1-S)

R＝3I-(G+B)

R wherein, G, B represent the red component under the RGB color model, blue component, green component respectively; H, what S, I represented respectively is chromatic component under the HIS color model, saturation degree component and strength component.

(4) feature extraction, analysis image, extract the characteristic that empty wagons position and parking stall is had car to take to have obvious separating capacity: color characteristic, edge feature, angle point characteristic, variance, its method is following:

(a) the parking stall color characteristic extracts: the colouring information of car is profuse, and under different illumination, may demonstrate various colors, is suitable difficulties if remove to portray the colouring information of all cars, and is incomplete.Change thought so, actual more single for the colouring information of empty wagons position, be exactly the cement flooring generally speaking, even the influence of illumination, the colouring information that is demonstrated also is very approaching.Under the YCbCr space; Cb, Cr component component can be good at reflecting the colouring information of image, and irrelevant with illumination variation; Utilize this color component to carry out the tolerance of empty wagons position color characteristic so; Overcome the interference that intensity of illumination is brought, and RGB is linear change to the conversion between the YCbCr, calculates simple.According to empty wagons position training color model, set up based on the mixed Gauss model under the color YCbCr space, CbCr has good cluster characteristic.Concrete steps are following:

1. color space conversion (RGB is to the YCbCr space).Conversion formula

Y＝0.299*R+0.587*G+0.114*B

Cb＝-0.147*R-0.289*G+0.436*B

Cr＝0.615*R-0.515*G-0.1*B

Wherein Y represents monochrome information, and Cb, Cr are chrominance informations, and R, G, B represent red component, green component and the blue component under the rgb space respectively.

2. set up the Multidimensional and Hybrid Gauss model.The M kind mode that statistics empty wagons position pixel possibly occur, each mode of m single Gauss model pixel, each mode is given different weights, forms mixed Gauss model.The formula of Multidimensional and Hybrid Gauss model is as follows:

$p\left(x_i\right)=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{a}_j{N}_j(x_i;u_j,{\mathrm{\Σ}}_j)$

$N_j(x;u_j,{\mathrm{\Σ}}_j)=\frac{1}{{\left(2\mathrm{\π}\right)}^{m/2}\sqrt{\left|{\mathrm{\Σ}}_j\right|}}\exp [-\frac{1}{2}{(x-{\mathrm{\μ}}_j)}^T{\mathrm{\Σ}}_j^{-1}(x-{\mathrm{\μ}}_j)]$

In the formula

a _iBe each mode shared weight in mixed Gauss model, N _j(x _iu _j, ∑ _j) probability density function of expression j single Gauss model. Represent the corresponding Cb of j Gauss model, the pixel average of Cr, x (Cb Cr) is sample,

P (x _i) representative sample the class probability density function, ∑ _jBe the covariance matrix of pixel under each mode.

These parameters are trained according to empty wagons position, parking lot pixel and are obtained, and get different light as much as possible, block, the distribution situation of empty wagons position pixel under the shade.The value of m is 3～5.The mixed Gauss model of setting up has stronger adaptive capacity to environment, and utilizes the unicity of colouring information, can well foreground image be detected.

3. pixel judgement.Input image pixels point x, the match condition of judgement pixel and mixed Gauss model, decision rule is to judge according to the ratio of background Gaussian distribution average in current pixel value and the mixed Gauss model whether currency is non-NULL parking stall pixel.As satisfy formula

|x _j-u _j|＞D|∑ _j|

Wherein D is an empirical value, generally gets 2.5.u _j, | ∑ _j|, be the average of each single Gauss model in the mixed Gauss model and the determinant of covariance matrix.If satisfy following formula, so just think that this pixel is the foreground pixel point.

4. add up the length breadth ratio LW of non-background pixel point number S and definite minimum rectangle.(b) improved Canny edge detection algorithm.The purpose of rim detection is to suppress noise effectively, the 2nd, and must accurate as far as possible density of locating the edge.Can overcome the influence of shade, illumination variation.

1. Canny rim detection.Ultimate principle is at first to pass through the Gaussian filter smoothed image, utilizes the finite difference of single order local derviation the assign to amplitude and the direction of compute gradient then, again gradient magnitude is carried out non-maximum value and suppresses, at last with detection of dual threshold algorithm and adjoining edge.

2. improve algorithm, utilize the meanshift algorithm that image is carried out color cluster.And then carry out the canny detection algorithm, through after the cluster, the information that color is close merges, and has filtered pseudo-edge information, can obtain accurate image border like this.Utilize the finite difference of single order local derviation the assign to amplitude and the direction of compute gradient then, again gradient magnitude is carried out non-maximum value and suppress, at last with detection of dual threshold algorithm and adjoining edge.

3. marginal point density calculation.

$d_E=\frac{\underset{(i,j)=1}{\mathrm{\Σ}}{G}_E}{S}$

Wherein, d _EExpression marginal point density, G _EThe edge pixel value is 1 point in the area image of expression two-value parking stall, and S representes the area of this parking stall.

(c) angle point feature detection.Car has abundant angle point information, and the angle point characteristic of empty wagons position and not obvious.Angle point is defined as two dimensional image brightness and changes the point that has maximum value on violent point or the image border curve.This paper select for use calculate simple, the angle point characteristic extracted evenly rationally, can the dose unique point and the stable Harris algorithm of operator carry out the angle point feature detection.

1. Harris algorithm principle: each point to the gray level image of operation, calculate this in horizontal and vertical first order derivative, and the product of the two; Unique point is the corresponding pixel of very big interest value in the subrange.

2. algorithm steps, use the Harris method and extract the process of angle point in the image and can be divided into following step:

I computed image pixel gradient in the horizontal and vertical directions, and both products obtain the value of 4 elements among the M: $M=\left(\begin{array}{cc}{I}_x^2& I_x{I}_y\\ I_x{I}_x& I_y^2\end{array}\right)$ Wherein $I_x^2=I_x\×I_x;$ $I_y^2{=I}_y\×I_y.$

I wherein _xBe pixel gradient in the horizontal direction, I _yBe the gradient of pixel in vertical direction

II carries out gaussian filtering to image, obtains new M.Discrete two-dimensional zero-mean Gaussian function is:

$\mathrm{Gauss}=\exp (-\frac{x^2+y^2}{2{\mathrm{\σ}}^2})$

III calculates the interest value of each corresponding on original image pixel, i.e. R value.K is a coefficient, generally gets 0.04.

$R=\{I_x^2\×I_y^2-{\left(I_x{I}_y\right)}^2\}-k{(I_x^2+I_y^2)}^2$

IV chooses Local Extremum.The Harris method thinks that unique point is the corresponding pixel of very big interest value in the subrange.

The V setting threshold is chosen a certain amount of angle point.

3. angle point number statistical count.The angle point number of statistics parking stall, if there is automobile storage to exist, the number of angle point is many, the angle point number of empty wagons position is fewer.

(d) the parking stall variance parameter calculates: with parking stall to be measured image I and the no back frame scape image I of choosing ₀It is poor to do, and calculates its absolute value G _s=| I-I ₀|, obtain zone, parking stall error image G _s, G _sThe information that only comprises independent parking stall, calculate the variance of this parking stall according to following formula:

${\mathrm{\σ}}_0=\sqrt{\frac{\underset{(i,j)\∈G_s}{\mathrm{\Σ}}{G}_s(i,j)-{\stackrel{\‾}{G}}_s}{n}}$

σ representes the variance in zone, parking stall here,

Expression parking stall zone error image G _sMean value, n represents G _sInterior pixel sum.

(5) svm classifier device design.Select 500 parking stall to be measured images as training sample image, be used for training classifier this paper to select SVM that five pattern features of separating capacity are relatively arranged as the image according to above extraction: the region area S that confirms according to the background mixed Gauss model and length breadth ratio LW, the marginal point density (d of minimum rectangular area _E), angle point number (count), variance (σ ₀), training svm classifier device.

1. the principle of svm classifier device.Support vector machine method is on the VC dimension theory and structural risk minimization principle basis of Statistical Learning Theory; Between complicacy of the model learning accuracy of specific training sample (promptly to) and learning ability (being the ability of faultless identification arbitrary sample), seek best compromise according to limited sample information, in the hope of obtaining best popularization ability.Through the some nonlinear transformations that choose in advance; Say that input vector is mapped in the high-dimensional feature space; In a feature space; Construct one the classification lineoid is arranged most. ask the optimal classification lineoid to be equivalent to and ask largest interval: employing Lagrange multiplier method is also used KKT (Karush-Kuhn-Tucker) condition, can try to achieve decision function and be:

Linear separability, decision function is: $f\left(x\right)=\mathrm{Sgn}[\underset{i}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i{y}_i<x_i,x>+b]$

When inseparable, the decision function after the introducing kernel function is:

$f\left(x\right)=\mathrm{sgn}[\underset{i}{\overset{l}{\mathrm{\&Sigma;}}}{{\mathrm{\&alpha;}}_i}^{*}{y}_{i}K(x_i,x_j)+b^{*}]$

Schedule of samples is shown (x _i, y _i) x wherein _i=(S, d _E, count, σ ₀), y _i=0 or 1

2. parameter is selected: penalty factor (C) and kernel function parameter (σ orq) are selected.Employed method is the cross validation method.Through testing the error rate of non-training sample on certain preset parameter value, continuous then corrected parameter, it is a kind of special case of leaving-one method (L00) estimated error rate.Its ultimate principle is for to be divided into n part with training sample, and optional wherein n-1 part is got remaining portion as test set, the span of setup parameter C and σ as training set.Choose the highest average accuracy of identification corresponding parameters to C and σ.If the parameter that the highest average accuracy of identification correspondence is different is right, get minimum average support vector number corresponding parameters to C and σ.This is that wherein L represents the support vector number because the computation complexity of SVMs is 0 (L*D), and D represents the proper vector dimension, and the more little required calculated amount of L is few more.The kernel function of this project selection is radially basic kernel function (RBF)

3. the selection of sample classification and boundary sample.

The I sample classification.Sample plays crucial effects to the training of sorter.Sample can be divided into, good sample (being easy to and other types differentiation), boundary sample (with the class sample at a distance of nearer sample), difference sample (sample of obscuring easily).The purpose of this subject study is the sample that is difficult to distinguish (difference sample) to those, can be correct distinguish, the quality of sorter training so is very important for the classification of sample.Discover that good sample can be so that the mode class zone of training place be more compact, but the different mode class is interregional at a distance from bigger.The existence of sample of difference possibly make that the mode class zone that trains is big as far as possible, and is easy to generate overlappingly between the adjacent area, has increased error in classification.Boundary sample is trained, can be so that the class zone that trains be big as far as possible, and that adjacent type of zone do not have as far as possible is overlapping, classification performance is best.

The selection of II boundary sample.Directly utilize the RBF_SVM sorter to carry out boundary sample and select, promptly at first utilize the cross validation method to carry out parameter and select, utilize the parameter training RBF_SVM sorter of selecting then, resulting support vector collection is the boundary sample of being asked.

III utilizes the final svm classifier device model of boundary sample at last.

(6) Target Recognition: target parking stall to be measured is calculated four parking stall characteristic parameters by above-mentioned (1)-(4) steps, and in the sorter of being trained that substitution is confirmed by step (5) respectively, the direct parking stall of must arriving takies situation.

The present invention compared with prior art has following remarkable advantages and beneficial effect:

At first, the present invention has proposed five types and can fully reflect the characteristic parameter information of whether parking the parking stall on the basis of fully researching and analysing the concrete environment of motor pool.Invented based on the background detection algorithm of the mixed Gauss model of color and utilize meanshift to carry out the improvement of edge detection algorithm after the color cluster; It is the innovative point of this paper; And each characteristic all has its singularity, avoids the influence of disturbing factor such as water mark on illumination, weather and the parking stall.Also effectively characteristic parameter is provided accurately for statistical model method for identifying and classifying (SVM).Secondly, in order to construct accurate svm classifier device, the present invention will classify to training sample; Utilize the searching of the advanced row bound sample of SVM; Utilize boundary sample at last, train final sorter, it is big that this method makes to try one's best in the mode class zone; And that adjacent area does not have is overlapping, and it is maximum that classification performance reaches.This invention effectively raises the accuracy rate of parking stall measure.This method of experiment proof had both guaranteed that the accuracy rate of parking stall identification had also improved the speed of parking stall measure simultaneously.

Description of drawings

Fig. 1 calculates four the characteristic parameter process flow diagrams in zone, parking stall;

Fig. 2 sets up the mixed Gauss model based on color characteristic.

Fig. 3 SVM mental model figure;

Fig. 4 optimal classification face synoptic diagram;

Fig. 5 project training svm classifier device process flow diagram;

Fig. 6 method for detecting parking stalls process flow diagram.

Embodiment

Adopt the CCD camera to carry out the collection of parking stall image among the present invention, the antenna height of video camera is generally 2-5 rice, and effective scene domain of every video camera covering comprises 1-4 parking stall, and camera relative position and shooting angle remain unchanged.Adopt a CCD camera in this example, comprise 4 parking stalls in the parking stall image of shooting, as shown in Figure 6.With first parking stall, promptly area maximum parking stall in parking stall is an example in the image at this.In computing machine, accomplish following steps, the practical implementation flow process is as shown in Figure 3:

The first step: select a parking stall not have car image image as a setting, select to require: this parking stall image disruption is less, reads this image and with converting gray level image to behind its smoothing denoising;

Second step: in this parking stall background image, confirm the frame coordinate of a parking stall, four apex coordinates of quadrilateral parking stall are (352,458), (550; 675), (490,715), (320; 512), only comprise the view data of a background parking space information, it is made as I0 according to four coordinate interceptings;

The 3rd step: select 500 width of cloth parking stall images as training sample database, wherein 250 width of cloth are the image of a parking stall when having car to take, and all the other 250 width of cloth are the image of a parking stall when not having car and taking.

The 4th step: read each parking stall to be measured image information in the training sample database, the regional extent that goes out a parking stall to be measured by the mode intercepting in second step is carried out pre-service, and its step is following:

At first the RGB color space conversion is utilized conversion formula under the HSI model, and extracts H component, I component, S component.Under the I space, carry out image enhancement processing, keep the colouring information of image.

Be transformed at last under the rgb space, under this model space, make things convenient for Feature Extraction.

The 5th step: extract and calculate four characteristic ginseng values in parking stall image, the characteristic parameter extraction flow process is as shown in Figure 1, and detailed process is following:

(1) color characteristic extracts.Mixed Gauss model based on background color is confirmed; This modeling statistics empty wagons bit image pixel possibility various case, the mixed Gauss model of foundation has very strong antijamming capability; Can be good at background and foreground image being distinguished strong robustness.Concrete color characteristic extraction step is following.

I extracts R, G, the B component of coloured image.

II is transformed under the YCbCr space.Extract Cb, Cr component.

III is brought into the pixel of each parking stall in the mixed Gauss model, and setting threshold is judged.

The length breadth ratio LW of the number S of non-background pixel point and the minimum rectangular area that obtains in the IV statistics zone, parking stall.These non-background areas are likely the pixel of car, and reaching the length breadth ratio ratio more greatly according to the area of car is these characteristics between 1～2.5, and other chaff interferences of car are distinguished.

(2) calculate parking stall marginal point density feature parameter, the concrete steps of improved canny edge detection algorithm:

A, at first carry out the clustering processing of image, smoothly fall non-pseudo-edge information based on color characteristic with the meanshift algorithm.Be transformed under the gray space.Carry out the canny rim detection.

B, utilize classical derivative operator to find the partial derivative in two directions of gradation of image, and obtain the amplitude and the direction of gradient: $\left|G\right|=\sqrt{G_x^2+G_y^2}$ $\mathrm{\&theta;}=\mathrm{Arctan}\frac{G_y}{G_x}$

C, non-maximal value suppress: the traversing graph picture, if the gray-scale value of former and later two pixels is compared and is not maximum on the gray-scale value of certain pixel and its gradient direction, so just this gray values of pixel points is changed to 0, and promptly not the edge.

D, dual threshold algorithm detect and the link edge.The high threshold T1 that utilizes the cumulative statistics histogram to obtain gets a low threshold value again and comes out (to get boundary=0.4T1) usually.Every must be the edge greater than high threshold; Every is the edge less than what hang down threshold value scarcely; If testing result is greater than threshold value but less than high threshold, that will see the edge pixel that whether surpasses high threshold in the adjacent pixels of this pixel: it has been exactly the edge so if any, otherwise he is not the edge just.

E, edge calculation dot density.

$d_E=\frac{\underset{(i,j)=1}{\mathrm{\&Sigma;}}{G}_E}{S}$

Wherein, d _EExpression marginal point density, G _EThe edge pixel value is 1 point in the area image of expression two-value parking stall, and S representes the area of this parking stall.

(3) angle point feature detection.Car has abundant angle point information, and the angle point characteristic of empty wagons position and not obvious.Angle point is defined as two dimensional image brightness and changes the point that has maximum value on violent point or the image border curve.This paper select for use calculate simple, the angle point characteristic extracted evenly rationally, can the dose unique point and the stable Harris algorithm of operator carry out the angle point feature detection.

1. Harris algorithm principle: each point to the gray level image of operation, calculate this in horizontal and vertical first order derivative, and the product of the two; Unique point is the corresponding pixel of very big interest value in the subrange.

2. algorithm steps, use the Harris method and extract the process of angle point in the image and can be divided into following step:

I computed image pixel gradient in the horizontal and vertical directions, and both products obtain the value of 4 elements among the M:

$M=\left(\begin{array}{cc}{I}_x^2& I_x{I}_y\\ I_x{I}_x& I_y^2\end{array}\right)$ Wherein $I_x^2=I_x\&times;I_x;$ $I_y^2{=I}_y\&times;I_y.$

I wherein _xBe pixel gradient in the horizontal direction, I _yBe the gradient of pixel in vertical direction

II carries out gaussian filtering to image, obtains new M.

III calculates the interest value of each corresponding on original image pixel, i.e. R value.

$R=\{I_x^2\&times;I_y^2-{\left(I_x{I}_y\right)}^2\}-k{(I_x^2+I_y^2)}^2$

IV chooses Local Extremum.The Harris method thinks that unique point is the corresponding pixel of very big interest value in the subrange.

The V setting threshold is chosen a certain amount of angle point.

3. angle point number statistical.The angle point number of statistics parking stall, if there is automobile storage to exist, the number of angle point is many, the angle point number of empty wagons position is fewer.

(4) carry out the parking stall variance parameter and calculate, make G _s=| I-I ₀|, (I is a present image, I ₀Be the background empty wagons bit image of choosing before).Obtain parking stall zone error image G _s,, calculate the variance of this parking stall according to following formula:

$\mathrm{\&sigma;}=\sqrt{\frac{\underset{(i,j)\&Element;G_s}{\mathrm{\&Sigma;}}{G}_s(i,j)-{\stackrel{\&OverBar;}{G}}_s}{n}}$

σ representes the variance in zone, parking stall here,

Expression parking stall zone error image G _sMean value, n represents G _sInterior pixel sum.

The 6th step: svm classifier device training.Schedule of samples is shown (x _i, y _i), and x _i=(S, d _E, count, σ ₀), y _i=0 or 1

1 parameter is selected and the cross validation method: penalty factor (C) and kernel function parameter (σ orq) are selected.Employed method is the cross validation method.Through to the error rate of this 500 width of cloth training sample on certain preset parameter value, continuous corrected parameter then, it is a kind of special case of leaving-one method (L00) estimated error rate.Its ultimate principle is for to be divided into 5 parts with training sample, optional wherein 4 parts as training set, get remaining a as test set, the span of setup parameter C and σ.Choose the highest average accuracy of identification corresponding parameters to C and σ.If the parameter that the highest average accuracy of identification correspondence is different is right, get minimum average support vector number corresponding parameters to C and σ.This is that wherein L represents the support vector number because the computation complexity of SVMs is 0 (L*D), and D represents the proper vector dimension, and the more little required calculated amount of L is few more.The kernel function of selecting to the parking lot sample characteristics is radially basic kernel function (RBF), C=3, σ=0.001

The selection of 2 boundary samples.Directly utilize the RBF_SVM sorter to carry out boundary sample and select, promptly at first utilize the cross validation method to carry out parameter and select, utilize the parameter training RBF_SVM sorter of selecting then, resulting support vector collection is the boundary sample of being asked.

3 utilize the final svm classifier device model of boundary sample at last.

The 7th step: read in a parking stall to be measured image, handle by top first to five step, calculate five characteristic ginseng values of parking stall, in the svm classifier device, the direct parking stall of must arriving took situation during substitution the 6th went on foot successively with them.

The 8th step: a parking stall recognition result of output testing image, represent that with 1 the parking stall has car to take, represent that with 0 the parking stall do not have car and take.Be the accuracy and the versatility of checking the inventive method detection parking stall, 500 width of cloth parking stall images of employing parking lot shooting carry out the experiment test of false declaration rate, loss, rate of false alarm, and experimental result shows that the present invention has good detection effect.

Analysis of simulation result

Table 1 is that four parking stalls shown in Figure 6 are (among its figure; 1 represents the parking stall No. one; 2 represent the parking stall No. two, and 3 represent the parking stall No. three, and 4 represent the parking stall No. four; Black is partly represented the background beyond the zone, parking stall) the experiment statistics data, add up the accuracy rate of Fuzzy Pattern Recognition Method through following three rates for the parking stall image detection:

1. false declaration rate=(will not having the number of image frames that car is judged to car)/(whole no car number of image frames);

2. loss=(will have car to be judged to the number of image frames of no car)/(whole no car number of image frames);

3. rate of false alarm=(will not have the number of image frames that car is judged to car+will have car to be judged to the number of image frames of no car)/(all images frame number);

Table 1 parking stall image measurement result

The car item	The false declaration rate	Loss	Rate of false alarm
				A parking stall	2.11％	0.11％	1.8％
No. two parking stalls	3.51％	0	1.37％
				No. three parking stalls	5.6％	0	5.58％
No. four parking stalls	13.09％	1.2％	10.09％

Claims (6)

1. the parking position detection algorithm based on color characteristic is characterized in that: comprise image processing module, characteristic extracting module, statistical sorter design module; May further comprise the steps:

Obtain the parking stall video data 1.1. adopt the CCD camera to take, the shooting area scope of video camera is several parking stalls, and camera relative position and shooting angle remain unchanged;

1.2. select a width of cloth not have the background image of car, read this image and with converting gray-scale map to behind its smoothing denoising;

1.3. the frame coordinate of no back frame scape image parking stall to be measured is set, and intercepting only comprises the view data of single parking space information to be measured, and the background image of the parking stall to be measured that obtains is made as I ₀

1.4. for every width of cloth testing image, the mode intercepting in 1.3 goes out the regional extent of concrete parking stall to be measured set by step then, and carries out pre-service; Read the information of the colored parking stall to be measured of each width of cloth image-region, carry out color space conversion, RGB is transformed under the HSI space, carry out image filtering and handle;

1.5. feature extraction phases: the color characteristic of parking stall is applied in the parking stall measure system; Foundation is based on the mixed Gauss model of background color; Obtain the area in zone, non-NULL parking stall and the length breadth ratio of minimum rectangular area; Extract marginal point density, angle point characteristic and variance characteristic then, utilize mathematical statistics method to calculate parking stall five characteristic parameter areas S, length breadth ratio LW, marginal point density d _E, angle point number count, parking stall variances sigma;

1.6. select several parking stall to be measured images as training sample image, choice of sample is wanted rationally, selects multiple image to train, wherein 50% width of cloth is the empty wagons bit image, and 50% width of cloth has car parking stall image; Utilize the theory of statistical model identification, training svm classifier device;

1.7. target parking stall to be measured is calculated several parking stall characteristic parameters by above-mentioned steps 1.1-1.5 step, in the sorter that substitution is confirmed by step 1.6 respectively, directly carries out Target Recognition.

2. a kind of parking position detection algorithm based on color characteristic according to claim 1, it is characterized in that: several described in the described step 1.6 are 500 width of cloth.

3. the parking position detection algorithm based on color characteristic according to claim 1 is characterized in that: described shooting area scope is 1-4 parking stall.

4. the parking position detection algorithm based on color characteristic according to claim 1 is characterized in that: described marginal point density d _E, adopt and carry out color cluster through meanshift, more close colouring information is synthesized one type, and then under gray space, carry out the canny rim detection, obtain accurate more marginal information:

$d_E=\frac{\underset{(i,j)=1}{\mathrm{\&Sigma;}}{G}_E}{S}$

Wherein, d _EExpression marginal point density, G _EThe edge pixel value is 1 point in the area image of expression two-value parking stall, and S representes the area of this parking stall.

5. the parking position detection algorithm based on color characteristic according to claim 1 is characterized in that: described angle point number count obtains through following steps: utilization Harris Corner Detection Algorithm counts parking stall regional extent angle point number count.

6. the parking position detection algorithm based on color characteristic according to claim 1 is characterized in that: described parking stall variances sigma is calculated: with parking stall to be measured image I and the no back frame scape image I of choosing ₀It is poor to do, and calculates it

σ representes the variance in zone, parking stall,

Expression parking stall zone error image G _sMean value, n represents G _sInterior pixel sum.

Images