CN102880863B - Method for positioning license number and face of driver on basis of deformable part model - Google Patents

Abstract

The invention relates to a method for positioning a license number and the face of a driver on the basis of a deformable part model, and belongs to the field of object detection. The method comprises the following steps: modeling a vehicle in a front view through the deformable part model, taking the license number and the face of the driver as parts of the model and training to acquire parameters of the model; and positioning the license number and the face of the driver precisely on the basis of the model and identifying the vehicle type on the basis of the relative position relationship between the license number and the face of the driver. With the method, the information of the position between the license number and the face of the driver can be utilized completely, the license number and the face of the driver can be positioned accurately and the information of the vehicle type can be acquired.

Description

A kind of car plate based on deformable part model and driver's Face detection method

Technical field

The invention belongs to object detection technical field, particularly the Combination application of vehicle license location technique and human face detection tech.

Background technology

Along with the development of society, the management of Modern Traffic is also increasingly sophisticated, heavy.In order to monitoring and controlling traffic situation accurately and efficiently and cost-saving, a lot of intelligent traffic monitoring technology is widely used.Wherein, automatic license plate identification system is the technology of current comparative maturity, is applicable to various traffic scene.Vehicle License Plate Recognition System is divided into License Plate, Character segmentation and character recognition three part.Vehicle license location technique wherein can obtain higher accuracy rate at present, but has also reached a bottleneck.If can not utilize other information in scene, the accuracy rate of License Plate cannot be further improved substantially.

Along with the development of imaging technique, high-definition camera is widely used.High-definition camera can obtain the picture rich in detail of vehicle in traveling and driver.The research of the Face detection under this type of scene of current correspondence is also few, and because driver's face is generally after front windshield, is easily subject to the impact at light and visual angle, also cannot obtain good positioning result at present.

Therefore, if the face information of driver can be combined effectively with the license board information of vehicle, then the accuracy rate of License Plate can further be improved.Simultaneously because the relative position of car plate and driver's face is more fixing, the positioning result of driver's face also can be improved by the positioning result of car plate.The identification of vehicle can be carried out in addition by the relative position of car plate and driver's face.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, in the image that high-definition camera obtains, carry out driver's Face detection and License Plate, and then carry out vehicle cab recognition, for follow-up traffic monitoring task provides more effective information.

The object of the invention is to be achieved through the following technical solutions.

Based on car plate and the driver's Face detection method of deformable part model, concrete implementation step is as follows:

Step one: the deformable part model setting up front vehicle

Using car plate and driver's face as parts, set up the deformable part model of front vehicle, by training data, obtain the position relationship between car plate and driver's face, as model parameter;

The deformable part model M of the vehicle in front view is defined as follows

M＝{part _plate,part _face,pos _plate,face} (1)

Wherein, part _platerepresent the car plate parts in model, part _facerepresent the face component in model, pos _{plate, face}={ p _{plate, face}, d _{plate, face}represent position relationship between car plate and driver's face.Wherein p _{plate, face}represent the spatial relation of car plate and driver's face.For country variant and area, because the position at driver place is different, this position relationship is also different.For Continental Area, need meet wherein plate.x, face.x represent the x coordinate of car plate and face, and plate.y, face.y represent the y coordinate of car plate and face, and namely car plate is in the lower left of face.D _{plate, face}represent the distance between car plate and face, and

d _plate,face∈N _i(μ _i,δ _i),i∈{big,middle,samll} (2)

N (μ, δ) represents that average is μ, and variance is the Gauss model of δ.

By the distance between the car plate of statistics mark and face, corresponding to the vehicle obtaining each type

All referring to and variance of Gauss model, namely obtain the deformable part model of vehicle.

Step 2: the coarse positioning carrying out car plate, obtains the candidate region of car plate and corresponding confidence level.

Have the method for multiple License Plate can obtain the candidate region of car plate and corresponding confidence level at present, the present invention adopts the license plate locating method based on paired morphological operator to obtain the candidate region of car plate and corresponding confidence level.

If S _{m × n}to be size be the structural elements of m × n and all values is 1, and the local neighborhood of certain pixel is by S _{m × n}determine.I represents gray level image, with represent the corrosion in mathematical morphology and expansive working respectively, with used morphological operation of giving a definition:

Closed operation: $I\·S_{m\×n}=(I\⊕S_{m\×n})\⊗S_{m\×n}$

Opening operation:

High cap computing: $\mathrm{I\Δ}{S}_{m\×n}=I-(I\·S_{m\×n})$

Black cap computing:

High cap computing (top-hat) does difference by source images and opening operation image, can extract the region that local is brighter; Black cap conversion (bot-hat) does difference by closed operation image and source images can extract darker region, local.Because between car plate background luminance and Character Intensity, contrast significantly, utilizes these two operations can isolate character and the background area of car plate, and Background suppression, eliminate uneven illumination.And continent car plate has dark word and the dark end bright word two type of putting one's cards on the table, (cap transformation or the conversion of black cap) cannot successfully extracting character zone carries out License Plate simultaneously only to use single morphological operation.Character information and car plate background information, by paired morphological operation, are carried out explicit combination, can detect the car plate of two types under unified framework by us.

For the bright word car plate in the dark end, in order to extract character zone, the operation of high cap can be carried out and binaryzation to it, as shown in Figure 2.

Now consider the background area of the bright word car plate in the dark end, if choose the linear structure unit S of horizontal direction _{1 × n}, car plate background can be divided into three parts, be respectively intercharacter background parts (redness), character inner background parts (green) and other background parts (blueness), as shown in Fig. 3 (b).

At S _{1 × n}under effect, blue region is local non-significant region of variation (the set of pixels brightness in pixel place linear structure unit neighborhood is consistent), and red area and green area are local marked changes.If carry out the conversion of black cap and binaryzation to bright word car plate of the dark end, then by filtering, green area and background corresponding to red area are retained, as shown in Fig. 3 (c) background that blue region is corresponding.Zones of different is carried out background classification (Fig. 3 (d)), and only consider intercharacter part background (Fig. 3 (e)), can find, the background area of intercharacter part meets characters on license plate height unanimously, equally distributed feature.The region that we claim intercharacter background corresponding is pseudo-character zone.

Carry out paired morphological operation to bright word car plate of the dark end and can extract actual characters and pseudo-character respectively, it is consistent that they all meet characters on license plate height, and the feature be evenly distributed, as shown in Fig. 3 (e).Therefore their union can be used for carrying out License Plate.The same car plate for dark word of putting one's cards on the table carries out paired morphological operation and can extract pseudo-character and actual characters respectively.Paired morphological operator method efficiently solves the restriction of single morphological operator method---needs to know that car plate foreground-background is arranged in pairs or groups in advance.Paired operator is respectively used to extract actual characters region and pseudo-character zone, car plate foreground information and car plate background information is effectively combined and jointly represents car plate, can process unified for two type car plates.It is as follows that character zone extracts flow process:

1) respectively following morphological operation is carried out to gray level image I, I ₁=I △ S _{1 × n}, I ₂=I ▽ S _{1 × n}

2) use Da-Jin algorithm to I ₁and I ₂carry out binaryzation respectively, obtain corresponding binary map I ₃and I ₄;

3) to I ₃and I ₄carry out Connected component mark respectively, obtain two Connected component set C ^topand C ^bot;

4) true character and pseudo-character are merged, and utilize area to carry out filtering to character by the priori size of car plate, obtain final character set: C=C ^top∪ C ^bot

After obtaining character set, the central point of each character can be obtained further.According to the character distribution feature point-blank on car plate, we are by judging whether these central points are located on the same line positioning licence plate.The variation range of usual car plate size is known in scene, and according to this priori, it is wide high 2 times that detection window is set to maximum car plate by us.Mobility detect window carries out License Plate in the picture, and the step-length of each movement is in the x direction 1/2nd of detection window width, and the step-length of each movement is in y-direction 1/2nd of detection window height.Can guarantee that car plate will at least appear in a windows detector like this, and significantly can reduce the number of candidate region.In each detection window, judge whether the central point of obtained character is located on the same line, if the number of the central point be located on the same line is greater than given threshold value, then think that region corresponding to these central points is candidate license plate region.The degree of confidence that simultaneously can calculate this candidate region is

confidence _plate＝|num_c-num_thres| (3)

Wherein num_c is obtained number of characters, and num_thres is the threshold value of setting.

Step 3: the coarse positioning carrying out driver's face, obtains the candidate region of face and corresponding confidence level.

The method having multiple face to monitor at present can obtain the candidate region of driver's face and corresponding confidence level, and the present invention adopts the method for detecting human face based on AdaBoost to obtain the candidate region of driver's face and corresponding confidence level.

Be according to constructing Weak Classifier based on the method for detecting human face of AdaBoost with rectangular characteristic, a small amount of key feature is picked out again by AdaBoost method, summation be weighted to corresponding Weak Classifier thus construct strong classifier, and it can be used as final sorter for Face datection.Wherein, each rectangular characteristic is made up of 2-3 rectangle, and as shown in Figure 4, its value is the pixel value sum that the pixel value sum in white rectangle deducts in black rectangle.

Each Weak Classifier is made up of a rectangular characteristic, strong classifier training flow process as follows:

(1) given training data (x ₁, y ₁) ... (x _n, y _n), wherein y _i=0 represents negative sample, y _i=1 represents positive sample, and n is the number of training sample.

(2) initialization weights, y _iwhen=0 y _iwhen=1 m, l are the number of negative sample and positive sample respectively.

(3) corresponding t=1 ..., T:

A. normalization weights

B. to each feature j, training Weak Classifier h _jif the error of this sorter is

C. the sorter h with least error is selected _t

D. weights are upgraded wherein e _iif=0 x _icorrectly classified, otherwise

e _i＝1， ${\mathrm{\β}}_t=\frac{{\mathrm{\ϵ}}_t}{1-{\mathrm{\ϵ}}_t}$

(4) then final strong classifier is

$h\left(x\right)=\left\{\begin{array}{cc}1& {\mathrm{\Σ}}_{t=1}^T{\mathrm{\α}}_t{h}_t\left(x\right)\≥\frac{1}{2}{\mathrm{\Σ}}_{t=1}^T{\mathrm{\α}}_t\\ 0& \mathrm{otherwise}\end{array}\right.,$ Wherein ${\mathrm{\α}}_t=\log \frac{1}{{\mathrm{\β}}_t}$

Owing to can carry out filtering based on deformable part model to the region that coarse positioning obtains by follow-up, therefore we are when utilizing strong classifier to carry out Face datection, and target makes loss minimum, admits of certain false drop rate.Arranging detected rule is: when detected region is by front K (K < T) individual Weak Classifier, think that this region is face candidate region, continue use K+1 to scan to T Weak Classifier simultaneously, and record the confidence level of number as this candidate region of the Weak Classifier that it passes through.Both

confidence _face＝num _{passed_classifier} (4)

Step 4: based on car plate and the meticulous location of driver's face of deformable part model

Step 2 provides the possible license plate candidate area in image, and step 3 provides the candidate region of the possible driver's face in image.Then we are by the deformable part model of vehicle obtained in step one, carry out the meticulous location of car plate and driver's face.

If L={l _plate, l _faceit is a model M realization in the picture.Wherein l _platerepresent car plate position in the picture, l _facerepresent the position of face in car plate.If m is (l _plate) represent that car plate position is at l _plateconfidence level, by formula (3) calculate, m (l _face) represent that face location is at l _faceconfidence level, calculate by formula (4).M (l _plate, l _face) represent the degree of conformity of position relationship between car plate and face and model, and

$m(l_{\mathrm{plate}},l_{\mathrm{face}})\arg \underset{i}{\max }P(d_{l_{\mathrm{plate}},l_{\mathrm{face}}}\&Element;N_i({\mathrm{\&mu;}}_i,{\mathrm{\&delta;}}_i))---\left(5\right)$

Wherein for the distance between license plate candidate area and face candidate region, N _i(μ _i, δ _i) train the Gauss model obtained, i ∈ { big, middle, samll} for passing through in step one.

Carry out car plate and the meticulous location of driver's face according to deformable part model, both find L ^*make

$L^{*}=\{{l^{*}}_{\mathrm{plate}},{l^{*}}_{\mathrm{face}}\}=\arg \underset{L}{\max }(m\left(l_{\mathrm{plate}}\right)+m\left(l_{\mathrm{face}}\right)+m(l_{\mathrm{platte}},l_{\mathrm{face}}))---\left(6\right)$

Can extract car plate subsequent sections by step 2, can extract face candidate region by step 3, we, according to the confidence level of candidate region, sort to candidate region.If the license plate candidate area after sequence is l _{plate_1}... l _{plate_n}, n is the number of license plate candidate area, and the driver's face candidate region after sequence is l _{face_1}... l _{face_m}, m is driver's face candidate region.The process of then carrying out car plate and driver's Face detection by deformable part model is: corresponding j=1 ..., m, calculates marking value record the car plate of maximum marking value and correspondence thereof and the driver's face location result as final car plate and driver's Face detection.

Step 5: the relative position relation based on car plate and driver's face carries out vehicle cab recognition

Best car plate position l is found by step 4 ^* _plateand driver face location l ^* _faceafter, by calculating

$i^{*}=\arg \underset{i}{\max }P(d({l^{*}}_{\mathrm{plate}},{l^{*}}_{\mathrm{face}})\&Element;N_i({\mathrm{\&mu;}}_i,{\mathrm{\&delta;}}_i))---\left(7\right)$

{ big, middle, samll}, can obtain final vehicle cab recognition result to i ∈, and both having obtained this car is large car, in-between car or compact car.

Main contents of the present invention are: use deformable part model to carry out modeling to the vehicle of front view, using car plate and driver's face as the parts in auto model, carry out the location of car plate and driver's face according to model, the accuracy rate of License Plate and the accuracy rate of driver's Face detection can be improved.Vehicle cab recognition can be carried out according to the relative position relation of car plate and driver's face simultaneously.

Advantage of the present invention

The present invention compared with prior art, has the advantage of the following aspects:

(1) adopt widely used high-definition camera to obtain image, drop into without the need to extra cost.

(2) make full use of the various information in scene, improve the locating accuracy of car plate and driver's face.

(3) can vehicle cab recognition be realized according to the relative position relation of car plate and driver's face, have a extensive future.

Accompanying drawing explanation

Fig. 1 is car plate based on deformable part model and driver's Face detection process flow diagram;

Fig. 2 is that bright character schematic diagram is extracted in the operation of high cap; Wherein, (a) original image, (b) high cap image, (c) high cap binary map;

Fig. 3 is that pseudo-character extracts schematic diagram; Wherein, (a) original image, (b) background divides schematic diagram, (c) black cap binary map, (d) background classification, (e) pseudo-character zone, (f) actual characters and pseudo-character

Fig. 4 is rectangular characteristic.

Embodiment

As shown in Figure 1, concrete implementation step is as follows for the flow process of the car plate based on deformable part model that the present invention proposes and driver's Face detection method:

Step one: the deformable part model setting up front vehicle

Deformable part model (Deformable Part-Based Model) is the model carrying out object detection in the picture popular in recent years, is one of best at present object detection algorithms.Deformable part model represents object by the position relationship described between all parts.In the auto model of the front view of our foundation, comprise two parts: car plate and driver's face.After determining the parts in model, the training of model is exactly by training data, obtains the position relationship between car plate and driver's face.

Our training data comprises large car, in-between car and compact car three class, manually marks the car plate position in the every width image in each class and driver's face location.Use mixed Gauss model represents the position relationship between car plate and driver's face, for the vehicle of each type, has a Gauss model to represent the position relationship between car plate and driver's face.

The deformable part model M of the vehicle in front view is defined as follows

M＝{part _plate,part _face,pos _plate,face} (1)

Wherein, part _platerepresent the car plate parts in model, part _facerepresent the face component in model, pos _{plate, face}={ p _{plate, face}, d _{plate, face}represent position relationship between car plate and driver's face.Wherein p _{plate, face}represent the spatial relation of car plate and driver's face.For country variant and area, because the position at driver place is different, this position relationship is also different.For Continental Area, need meet wherein plate.x, face.x represent the x coordinate of car plate and face, and plate.y, face.y represent the y coordinate of car plate and face, and namely car plate is in the lower left of face.

D _{plate, face}represent the distance between car plate and face, and

d _plate,face∈N _i(μ _i,δ _i),i∈{big,middle,samll} (2)

N (μ, δ) represents that average is μ, and variance is the Gauss model of δ.

We, by adding up the distance between the car plate of mark and face, obtain all referring to and variance of the Gauss model corresponding to vehicle of each type, namely obtain the deformable part model of vehicle.

Step 2: adopt the license plate locating method based on paired morphological operator to carry out the coarse positioning of car plate, obtains the candidate region of car plate and corresponding confidence level.

The license plate locating method based on paired morphological operator is adopted to carry out the coarse positioning of car plate.If S _{m × n}to be size be the structural elements of m × n and all values is 1, and the local neighborhood of certain pixel is by S _{m × n}determine.I represents gray level image, with represent the corrosion in mathematical morphology and expansive working respectively, with used morphological operation of giving a definition:

Closed operation: $I\&CenterDot;S_{m\&times;n}=(I\&CirclePlus;S_{m\&times;n})\&CircleTimes;S_{m\&times;n}$

Opening operation:

High cap computing: $\mathrm{I\&Delta;}{S}_{m\&times;n}=I-(I\&CenterDot;S_{m\&times;n})$

Black cap computing:

High cap computing (top-hat) does difference by source images and opening operation image, can extract the region that local is brighter; Black cap conversion (bot-hat) does difference by closed operation image and source images can extract darker region, local.Because between car plate background luminance and Character Intensity, contrast significantly, utilizes these two operations can isolate character and the background area of car plate, and Background suppression, eliminate uneven illumination.And continent car plate has dark word and the dark end bright word two type of putting one's cards on the table, (cap transformation or the conversion of black cap) cannot successfully extracting character zone carries out License Plate simultaneously only to use single morphological operation.Character information and car plate background information, by paired morphological operation, are carried out explicit combination, can detect the car plate of two types under unified framework by us.

For the bright word car plate in the dark end, in order to extract character zone, the operation of high cap can be carried out and binaryzation to it, as shown in Figure 2.

Now consider the background area of the bright word car plate in the dark end, if choose the linear structure unit S of horizontal direction _{1 × n}, car plate background can be divided into three parts, be respectively intercharacter background parts (redness), character inner background parts (green) and other background parts (blueness), as shown in Fig. 3 (b).

At S _{1 × n}under effect, blue region is local non-significant region of variation (the set of pixels brightness in pixel place linear structure unit neighborhood is consistent), and red area and green area are local marked changes.If carry out the conversion of black cap and binaryzation to bright word car plate of the dark end, then by filtering, green area and background corresponding to red area are retained, as shown in Fig. 3 (c) background that blue region is corresponding.Zones of different is carried out background classification (Fig. 3 (d)), and only consider intercharacter part background (Fig. 3 (e)), can find, the background area of intercharacter part meets characters on license plate height unanimously, equally distributed feature.The region that we claim intercharacter background corresponding is pseudo-character zone.

Carry out paired morphological operation to bright word car plate of the dark end and can extract actual characters and pseudo-character respectively, it is consistent that they all meet characters on license plate height, and the feature be evenly distributed, as shown in Fig. 3 (e).Therefore their union can be used for carrying out License Plate.The same car plate for dark word of putting one's cards on the table carries out paired morphological operation and can extract pseudo-character and actual characters respectively.Paired morphological operator method efficiently solves the restriction of single morphological operator method---needs to know that car plate foreground-background is arranged in pairs or groups in advance.Paired operator is respectively used to extract actual characters region and pseudo-character zone, car plate foreground information and car plate background information is effectively combined and jointly represents car plate, can process unified for two type car plates.It is as follows that character zone extracts flow process:

1) respectively following morphological operation is carried out to gray level image I, I ₁=I △ S _{1 × n}, I ₂=I ▽ S _{1 × n}

2) use Da-Jin algorithm to I ₁and I ₂carry out binaryzation respectively, obtain corresponding binary map I ₃and I ₄;

3) to I ₃and I ₄carry out Connected component mark respectively, obtain two Connected component set C ^topand C ^bot;

4) true character and pseudo-character are merged, and utilize area to carry out filtering to character by the priori size of car plate, obtain final character set: C=C ^top∪ C ^bot

After obtaining character set, the central point of each character can be obtained further.According to the character distribution feature point-blank on car plate, we are by judging whether these central points are located on the same line positioning licence plate.The variation range of usual car plate size is known in scene, and according to this priori, it is wide high 2 times that detection window is set to maximum car plate by us.Mobility detect window carries out License Plate in the picture, and the step-length of each movement is in the x direction 1/2nd of detection window width, and the step-length of each movement is in y-direction 1/2nd of detection window height.Can guarantee that car plate will at least appear in a windows detector like this, and significantly can reduce the number of candidate region.In each detection window, judge whether the central point of obtained character is located on the same line, if the number of the central point be located on the same line is greater than given threshold value, then think that region corresponding to these central points is candidate license plate region.The degree of confidence that simultaneously can calculate this candidate region is

confidence _plate＝|num_c-num_thres| (3)

Wherein num_c is obtained number of characters, and num_thres is the threshold value of setting.

Step 3: adopt the method for detecting human face based on AdaBoost to carry out the coarse positioning of driver's face, obtains the candidate region of face and corresponding confidence level.

We adopt the method for detecting human face based on AdaBoost to carry out the coarse positioning of driver's face.The method is according to constructing Weak Classifier with rectangular characteristic, then picks out a small amount of key feature by AdaBoost method, is weighted summation thus constructs strong classifier, and it can be used as final sorter for Face datection to corresponding Weak Classifier.Wherein, each rectangular characteristic is made up of 2-3 rectangle, and as shown in Figure 4, its value is the pixel value sum that the pixel value sum in white rectangle deducts in black rectangle.

Each Weak Classifier is made up of a rectangular characteristic, strong classifier training flow process as follows:

(1) given training data (x ₁, y ₁) ... (x _n, y _n), wherein y _i=0 represents negative sample, y _i=1 represents positive sample, and n is the number of training sample.

(2) initialization weights, y _iwhen=0 y _iwhen=1 m, l are the number of negative sample and positive sample respectively.

(3) corresponding t=1 ..., T:

A. normalization weights $w_{t,i}\&LeftArrow;\frac{w_{t,i}}{{\mathrm{\&Sigma;}}_{j=1}^n{w}_{t,j}}$

B. to each feature j, training Weak Classifier h _jif the error of this sorter is

C. the sorter h with least error is selected _t

D. weights are upgraded wherein e _iif=0 x _icorrectly classified, otherwise e _i=1, ${\mathrm{\&beta;}}_t=\frac{{\mathrm{\&epsiv;}}_t}{1-{\mathrm{\&epsiv;}}_t}$

(4) then final strong classifier is

$h\left(x\right)=\left\{\begin{array}{cc}1& {\mathrm{\&Sigma;}}_{t=1}^T{\mathrm{\&alpha;}}_t{h}_t\left(x\right)\&GreaterEqual;\frac{1}{2}{\mathrm{\&Sigma;}}_{t=1}^T{\mathrm{\&alpha;}}_t\\ 0& \mathrm{otherwise}\end{array}\right.,$ Wherein ${\mathrm{\&alpha;}}_t=\log \frac{1}{{\mathrm{\&beta;}}_t}$

Owing to can carry out filtering based on deformable part model to the region that coarse positioning obtains by follow-up, therefore we are when utilizing strong classifier to carry out Face datection, and target makes loss minimum, admits of certain false drop rate.We arrange detected rule: when detected region is by front K (K < T) individual Weak Classifier, think that this region is face candidate region, continue use K+1 to scan to T Weak Classifier simultaneously, and record the confidence level of number as this candidate region of the Weak Classifier that it passes through.Both

confidence _face＝num _{passed_classifier} (4)

Step 4: based on car plate and driver's Face detection of deformable part model

Step 2 provides the possible license plate candidate area in image, and step 3 provides the candidate region of the possible driver's face in image.Then we are by the deformable part model of vehicle obtained in step one, carry out the meticulous location of car plate and driver's face.

If L={l _plate, l _faceit is a model M realization in the picture.Wherein l _platerepresent car plate position in the picture, l _facerepresent the position of face in car plate.If m is (l _plate) represent that car plate position is at l _plateconfidence level, by formula (3) calculate, m (l _face) represent that face location is at l _faceconfidence level, calculate by formula (4).M (l _plate, l _face) represent the degree of conformity of position relationship between car plate and face and model, and

$m(l_{\mathrm{plate}},l_{\mathrm{face}})\arg \underset{i}{\max }P(d_{l_{\mathrm{plate}},l_{\mathrm{face}}}\&Element;N_i({\mathrm{\&mu;}}_i,{\mathrm{\&delta;}}_i))---\left(5\right)$

Wherein for the distance between license plate candidate area and face candidate region, N _i(μ _i, δ _i) train the Gauss model obtained, i ∈ { big, middle, samll} for passing through in step one.

Carry out car plate and the meticulous location of driver's face according to deformable part model, both find L ^*make

$L^{*}=\{{l^{*}}_{\mathrm{plate}},{l^{*}}_{\mathrm{face}}\}=\arg \underset{L}{\max }(m\left(l_{\mathrm{plate}}\right)+m\left(l_{\mathrm{face}}\right)+m(l_{\mathrm{platte}},l_{\mathrm{face}}))---\left(6\right)$

Can extract car plate subsequent sections by step 2, can extract face candidate region by step 3, we, according to the confidence level of candidate region, sort to candidate region.If the license plate candidate area after sequence is l _{plate_1}... l _{plate_n}, n is the number of license plate candidate area, and the driver's face candidate region after sequence is l _{face_1}... l _{face_m}, m is driver's face candidate region.The process of then carrying out car plate and driver's Face detection by deformable part model is: corresponding j=1 ..., m, calculates marking value record the car plate of maximum marking value and correspondence thereof and the driver's face location result as final car plate and driver's Face detection.

Step 5: the relative position relation based on car plate and driver's face carries out vehicle cab recognition

Best car plate position l is found by step 4 ^* _plateand driver face location l ^* _faceafter, by calculating

$i^{*}=\arg \underset{i}{\max }P(d({l^{*}}_{\mathrm{plate}},{l^{*}}_{\mathrm{face}})\&Element;N_i({\mathrm{\&mu;}}_i,{\mathrm{\&delta;}}_i))---\left(7\right)$

{ big, middle, samll}, can obtain final vehicle cab recognition result to i ∈, and both having obtained this car is large car, in-between car or compact car.

Claims (2)

1. based on car plate and the driver's Face detection method of deformable part model, it is characterized in that, a kind of car plate based on deformable part model and driver's Face detection method, specific implementation step is as follows:

Step one: the deformable part model setting up front vehicle

Using car plate and driver's face as parts, set up the deformable part model of front vehicle, by training data, obtain the position relationship between car plate and driver's face, as model parameter;

The deformable part model M of the vehicle in front view is defined as follows

M＝{part _plate,part _face,pos _plate,face} (1)

Wherein, part _platerepresent the car plate parts in model, part _facerepresent the face component in model, represent the position relationship between car plate and driver's face; Wherein p _{plate, face}represent the spatial relation of car plate and driver's face; For country variant and area, because the position at driver place is different, this position relationship is also different; For Continental Area, plate.x < face.x need be met, plate.y < face.y, wherein plate.x, face.x represent the x coordinate of car plate and face, plate.y, face.y represents the y coordinate of car plate and face, and namely car plate is in the lower left of face; And

d _plate,face∈N _i(μ _i,δ _i),i∈{big,middle,samll} (2)

D _{plate, face}represent the distance between car plate and face;

N (μ, δ) represents that average is μ, and variance is the Gauss model of δ;

Distance between the car plate marked by statistics and face, is obtained all referring to and variance of the Gauss model corresponding to vehicle of each type, has namely arrived the deformable part model of vehicle;

Step 2: adopt the license plate locating method based on paired morphological operator to carry out the coarse positioning of car plate, obtains the candidate region of car plate and corresponding confidence level

Step 3: adopt the method for detecting human face based on AdaBoost to carry out the coarse positioning of driver's face, obtains the candidate region of face and corresponding confidence level

Step 4: the face candidate region that the license plate candidate area that the deformable part model of front vehicle set up based on step one, step 2 obtain and confidence level and step 3 obtain and confidence level carry out the meticulous location of car plate and driver's face

If L={l _plate, l _faceit is a model M realization in the picture; Wherein l _platerepresent car plate position in the picture, l _facerepresent the position of face in car plate; If m is (l _plate) represent that car plate position is at l _plateconfidence level, m (l _face) represent that face location is at l _faceconfidence level; M (l _plate, l _face) represent the degree of conformity of position relationship between car plate and face and model, and

$m(l_{\mathrm{plate}},l_{\mathrm{face}})=\arg \underset{i}{\max }P(d_{l_{\mathrm{plate}},l_{\mathrm{face}}}\&Element;N_i({\mathrm{\&mu;}}_i,{\mathrm{\&delta;}}_i))---\left(3\right)$

Wherein for the distance between license plate candidate area and face candidate region, N _i(μ _i, δ _i) train the Gauss model obtained, i ∈ { big, middle, samll} for passing through in step one;

Carry out car plate and the meticulous location of driver's face according to deformable part model, both find L ^*make

$L^{*}=\{{l^{*}}_{\mathrm{plate}},{l^{*}}_{\mathrm{face}}\}=\arg \underset{L}{\max }(m\left(l_{\mathrm{plate}}\right)+m\left(l_{\mathrm{face}}\right)+m(l_{\mathrm{plate}},l_{\mathrm{face}}))---\left(4\right)$

Car plate subsequent sections can be extracted by step 2, face candidate region can be extracted by step 3, according to the confidence level of candidate region, be sorted in candidate region; If the license plate candidate area after sequence is l _{plate_1 ...}l _{plate_n}, n is the number of license plate candidate area, and the driver's face candidate region after sequence is l _{face_1 ...}l _{face_m}, m is driver's face candidate region; The process of then carrying out car plate and driver's Face detection by deformable part model is: corresponding i=1 ..., n, j=1 ..., m, calculates marking value m (l _{plate_i})+m (l _{face_j})+m (l _{plate_i}, l _{face_j}), record the car plate of maximum marking value and correspondence thereof and the driver's face location result as final car plate and driver's Face detection;

Step 5: vehicle cab recognition is carried out in the position of the car plate obtained based on step 4 and driver's face

Best car plate position l is found by step 4 ^* _plateand driver face location l ^* _faceafter, by calculating

$i^{*}=\arg \underset{i}{\max }P(d({l^{*}}_{\mathrm{plate}},{l^{*}}_{\mathrm{face}})\&Element;N_i({\mathrm{\&mu;}}_i,{\mathrm{\&delta;}}_i))---\left(5\right)$

{ big, middle, samll} can obtain final vehicle cab recognition result to i ∈.

2. a kind of car plate based on deformable part model according to claim 1 and driver's Face detection method, is characterized in that, adopts the license plate locating method based on paired morphological operator to carry out the coarse positioning of car plate.