CN110135894A - A method of vehicles advertisement is passed through based on passenger's identification - Google Patents
A method of vehicles advertisement is passed through based on passenger's identification Download PDFInfo
- Publication number
- CN110135894A CN110135894A CN201910325858.8A CN201910325858A CN110135894A CN 110135894 A CN110135894 A CN 110135894A CN 201910325858 A CN201910325858 A CN 201910325858A CN 110135894 A CN110135894 A CN 110135894A
- Authority
- CN
- China
- Prior art keywords
- information
- passenger
- cell
- advertisement
- vehicles
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0265—Vehicular advertisement
- G06Q30/0266—Vehicular advertisement based on the position of the vehicle
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Abstract
The invention discloses a kind of methods by vehicles advertisement based on passenger's identification, ad distribution platform and passenger identification systems are provided in the vehicles, comprising the following steps: step (1): passenger identification systems obtain the gender of current passenger and/or age information and/or fat or thin information and/or wear information；Step (2): the information of current passenger is sent to ad distribution platform by passenger identification systems；Step (3): ad distribution platform goes out matching advertisement according to the information sifting of current passenger and is ranked up according to matching degree to the advertisement filtered out and is sequentially played.The present invention can identify the information of passenger in a vehicle, then match according to different passengers and push corresponding advertisement, receiving and ratings of the passenger to advertisement are substantially increased in this way, to improve the effect of publicity of advertisement.
Description
Technical field
The invention belongs to media advertising technical fields, specifically a kind of to be pushed away based on what passenger identified by the vehicles
The method for sending advertisement.
Background technique
Traditional advertisement promotion channel mainly includes TV, newspaper, website, the electronic display of setting in public places
Deng.These ways of promotion mainly pay certain ad distribution expense to distributor from advertiser or advertising company, distributor's
Media platform promulgating advertisement, to show consumer.The main problem of this advertisement promotion mode first is that can not determine wide
Accuse whether information is effectively communicated to consumer.For example, may play many advertisements in TV programme, consumer is being watched
After complete program, attention may be will disperse, and can not pay attention to the content in advertisement.In another example in the media such as newspaper, portal website,
Many advertisements may be shown on the page, consumer can ignore page in order to take a glance at the newspaper, when the main contents of portal website intentionally
Advertisement on face.Consumer is this to ignore the waste that can be seen as a kind of advertising resource in attention.
Net about vehicle and taxi have become going out for many people as the vehicles important in city in modern city
The first choice of the row vehicles, and pass through investigation discovery, this part population is largely booming income crowd, and as people are to low
Carbon life pursuit and Modern Traffic more congestion, more and more people select public transport trip for example public transport and
Subway, therefore realize that advertising is an effective means by the vehicles, how to have to these crowds
For advertising, be the emphasis of each advertising corporation research.As it is common using taxi, net about vehicle as ad distribution
Media be it is very universal, mainly to the LED billboard of roof, seat Static planar advertisement and dynamic video advertisement.Dynamically
Video ads are beautiful in colour, contain much information, and had application on the taxi in some cities, but lack there is also some
Point, the mainly content of advertisement are all pre-set in systems, for all passengers all be roll loop play these
Video, so many passengers are substantially love and see or look over so as to check and forget to ad content, because firstly, advertisement is broadcast
It puts and passenger is not effectively directed to and is selected, different crowds (such as different genders, different ages, difference
Income level) can all have certain selectivity to advertisement, be not that everyone can be interested in all advertisements, therefore leads
Advertising inefficient is caused, is taken less than ideal effect, meanwhile, scientific and technological now increasingly developed, face recognition technology is also increasingly complete
It is kind, the expression information of people can be identified by face recognition technology, or even gender, the age of people are analyzed from face information
Deng wherein outstanding has Microsoft open source projects Oxford, wherein mainly there is the function such as face recognition, Gender Classification and age detection
Can, and it is provided freely to developer's use, and development difficulty is reduced, certainly, producer oneself can also write corresponding code,
Corresponding data model is established, realizes the function of recognition of face, and the basis of Gender Classification and age detection is face recognition skill
Art, age detection and Gender Classification are in a computer based on face characteristic expression, training data collection, recurrence/disaggregated model
Foundation and model optimization carry out regression analysis and statistical classification, to obtain corresponding gender information and age information.
Secondly, the advertisement that passenger can play these is treated as conventional common advertisement, therefore most people
Does is what relationship phychology all: these advertisements and oneself having? just think, the advertisement that a passenger sees be apart from oneself tens even
The advertisement of advertiser (such as a food and drink quotient) outside kilometers up to a hundred, and so remote place will not be gone to eat in oneself short time
Meal, even if so he by advertisement to this family food and drink quotient produce interest be likely to be merely resting on the level of thought without
It can try out at once, and the so more than ten seconds time that the effectiveness of advertisement is namely seen at him can exist, so extensively
The effect played accused also just is had a greatly reduced quality.
Summary of the invention
The present invention provides a kind of pushing by the vehicles based on passenger's identification in view of the above problems in the prior art
The method of advertisement.
To achieve the goals above, The technical solution adopted by the invention is as follows:
A method of based on passenger's identification by vehicles advertisement, advertisement is provided in the vehicles
Distribution platform and passenger identification systems, comprising the following steps:
Step (1): the gender and/or age information of the passenger identification systems acquisition current passenger and/or fat or thin information
And/or wear information；
Step (2): the passenger identification systems by the gender of current passenger and/or age information and/or fat or thin information and/
Or it wears information and is sent to ad distribution platform；
Step (3): the ad distribution platform is according to the gender and/or age information and/or fat or thin information of current passenger
And/or it wears information sifting and goes out matching advertisement and the advertisement filtered out is ranked up according to matching degree and is sequentially broadcast
It puts, the matching degree can be set as various ways, such as be ranked up according to the degree of correlation of a certain information (such as gender).
Further, the passenger identification systems include camera and control analysis module, and the camera absorbs passenger
Image information obtains passenger's human face image information from image information, wears information, and control analysis module is based on recognition of face skill
Art obtains human face characteristic point from human face image information and obtains gender according to human face characteristic point progress gender and the analysis of/character classification by age
Information, control analysis module are based on face characteristic and figure aspectual character and standard faces size and figure profile size to score
Fat or thin information is analysed, control analysis module wears information based on the brand for wearing information analysis clothing dress ornament to obtain.
Further, control analysis module is based on face recognition technology and obtains human face characteristic point and root from human face image information
Gender is carried out according to human face characteristic point and the analysis of/character classification by age show that the process of gender information includes:
After the human face image information for obtaining passenger, image is pre-processed: Haar being utilized to the image of input first
Feature cascades detection of classifier face, positions position of human eye, obtains eyes distance and eyes tilt angle；Then the angle is utilized
Two dimensional affine transformation is done, face is rotated, solves the different influence of posture angle；Face part is normalized again, is kept away
Exempt from face influence not of uniform size；Brightness normalizing finally is carried out using histogram equalization, noise is eliminated using smoothing processing, reaches
To the effect of human face light equilibrium to obtain the facial image with uniform characteristics；
Gabor wavelet transformation is carried out to pretreated image, two-dimensional Gabor function representation is
Wherein,U, v are direction and the scale of core；Z (x, y) is direction vector,
X, y are two-dimensional coordinate；kU, vFor the centre frequency of filter；F is sampling step length；σ is Gauss window width and wavelength ratio, face
Gabor wavelet transformation be with the Gabor filter convolution of facial image and different scale different directions, f (x, y) is image
Intensity profile has
G (x, y, u, v)=f (x, v) * ψU, v(z)
Facial image and 5 scales after 40 Wavelet Kernel whole convolution in 8 directions, will obtain 40 filtering figures
Picture, so far available facial image by Gabor wavelet transformation after three-dimensional matrice, can by three-dimensional matrice sequence store at
The image pattern resolution ratio of one-dimensional feature vector, input is M × N, then feature vector dimension is M × N × 40, obtained above
5 scales, there are bulk redundancies for the feature vector in 8 directions, but the feature on most of directions is not obvious, therefore utilize
The method of weight matrix, by assigning certain weight to feature on direction, so that the feature on the direction under certain scales
It is more prominent, and dimension is significantly reduced, the method that the calculating of weight matrix uses gradient Nearest Neighbor with Weighted Voting is specific as follows: will
The connected region Block that image is divided into the connected region Cell not overlapped and has overlapping, wherein horizontal line grid rectangle is one
Cell, cross grid frame rectangle are a Block, and oblique line filling rectangle is the Cell being overlapped between 2 Block；
The size for defining Cell is 8 × 8, and unit is pixel；The size of Block is 2 × 2, and unit is Cell, due in ladder
The Nearest Neighbor with Weighted Voting spent in direction Histogram nomography and SIFT algorithm is undirected nine channel, here in order to which same Gabor wavelet is decomposed
8 directions be consistent, need for π to be divided into 8 parts, the angle size in each direction are as follows:For
Each Cell counts its gradient direction, carries out a Nearest Neighbor with Weighted Voting, and weight is the modulus value of the gradient, obtains so each
The directional statistics information of Cell, the region Block to comprising the region Cell be normalized, normalization algorithm use L2-
Then norm carries out a global normalization to each Cell, then all statistical informations of relative position is taken the channel mean value
It merges, a L is carried out again to the eight channel statistical informations of Cell1- norm normalization, so far each region Cell has 8
A weight, entire picture can be reserved for as a weight matrix
Wherein, m is the number of Cell, ti(i=1 ..., m) be composed by the weight in 8 directions of i-th of Cell to
Amount；
The feature vector for classification is obtained by weight matrix weighted array to the feature vector that Gabor is extracted, specifically
Process is as follows: it can be seen from the above, image is after Gabor transformation, 5 scales are obtained to each pixel, 8 directions
40 convolution values, due to weight matrix be obtained as unit of Cell, so the value after Gabor transformation also can be regarded as with
Cell is that unit is stored in a three-dimensional matrice, and when storage continuously saves the numerical value for belonging to same scale, for i-th
A pixel in Cell, can be expressed as by the value after Gabor transformation under j-th of scale
Gj8=(gj1..., gj8)
Wherein gjk(k=1 ..., 8) is the vector composed by the value in 8 directions under jth (j=1 ..., 5) a scale,
Then using weights matrix is weighted combination to features described above matrix, i.e., with 8 weights of i-th of Cell with obtain i-th
The value in 8 directions in a Cell under each scale of each pixel makees inner product:
F=Gj8·(ti)T, i=1 ..., m
Direction value of each pixel under 5 scales in i-th of Cell is obtained, to each pixel in all Cell
Same processing to be done, a new three-dimensional matrice is obtained, the number of plies of this three-dimensional matrice is reduced from U × V layers to V layers,
Middle U is general direction number, and V is out to out number, then carries out size adjusting to image, is adjusted with linear interpolation method to small size, i.e.,
By the Image Adjusting of original M × N to (M/5) × (N/5), dimension is finally fallen below into (M × N)/5 from the dimension of original M × N × 40
Dimension, last journal matrix form one-dimensional feature vector, so as to complete the weighted array converted based on Gabor wavelet
The extraction of feature；
By the face character feature vector of said extracted, sample set D is constituted, is existed using tree classifier
The lane database of WebFace or FaceScrub or Adience is trained and tests, and utilizes tree classifier processing higher-dimension
Data select the feature more important to gender and/or character classification by age to obtain required gender and/or age information.
Further, the ad distribution platform includes memory module, and the memory module includes several according to gender
And/or age information and/or fat or thin information and/or the storage section for wearing information setting, the ad datas of several species according to
Gender and/or age information and/or fat or thin information and/or wear information classified storage several storage sections in, store mould
Root tuber according to gender and/or age information and/or fat or thin information and/or wear information with several storage sections matched, and
And the advertisement for transferring the storage section of pairing plays out.
Further, positioning system is additionally provided in the vehicles, the positioning system obtains working as the vehicles
The geographical location information is simultaneously sent to external server by preceding geographical location information, and the external server is according to the geography
The information for the advertiser that it is N kilometers as the center of circle, radius using the geographical location that location information, which is obtained, the external server is by institute
The information for stating advertiser is sent to ad distribution platform, and the memory module is according to gender and/or age information and/or fat or thin letter
It ceases and/or wears information and matched with several storage sections, and transfer the advertisement of the advertiser in the storage section of pairing
It plays out.
Further, the vehicles are net about vehicles or taxi.
Further, the positioning system is GPS system.
Further, the ad distribution platform includes being arranged in net about vehicle or taxi front-row seats back and/or pair
The display module on storage tank is driven, the advertisement is played with the mode of video by display module
Further, the value range of the N are as follows: 30 >=N >=1.
Further, the address in the advertisement including advertiser and/or the information apart from current geographic position distance.
Compared with prior art, the present invention its remarkable advantage is:
(1) present invention introduces passenger identification systems on the basis of existing technology, can identify multiply in a vehicle
The gender of visitor and/or age information and/or fat or thin information and/or wear information, then according to different passenger's matchings and pushes
Corresponding advertisement substantially increases receiving and ratings of the passenger to advertisement, to improve the effect of publicity of advertisement in this way；
(2) the invention proposes a kind of method of accurately and efficiently acquisition of gender and/or age information, effectively realize,
Cooperate and support advertisement sending method of the invention；
(3) present invention introduces positioning system on the basis of existing technology, and the advertisement so played will no longer
It is the pre- constant sequence set, but with current location apart from very close businessman, if such passenger produces ad content
Raw interest is just likely to go to businessman's experience or consumption, because distance is very close, passenger goes to businessman and also easily holds very much
It easily realizes, to substantially increase advertising effect and practical changing effect by means of the present invention.
Detailed description of the invention
Fig. 1 is the flow chart of the method by vehicles advertisement identified the present invention is based on passenger.
Fig. 2 is the schematic diagram of the Cell and Block in image.
Fig. 3 is the multiple expression schematic diagram in the region Cell.
Fig. 4 is the forming process schematic diagram of weight matrix.
Fig. 5 is tree classifier flow chart.
Specific embodiment
With reference to the accompanying drawings of the specification, the present invention is further illustrated.
In conjunction with Fig. 1, a method of based on passenger's identification by vehicles advertisement, set in the vehicles
It is equipped with ad distribution platform and passenger identification systems, comprising the following steps:
Step (1): the gender and/or age information of the passenger identification systems acquisition current passenger and/or fat or thin information
And/or wear information；
Step (2): the passenger identification systems by the gender of current passenger and/or age information and/or fat or thin information and/
Or it wears information and is sent to ad distribution platform；
Step (3): the ad distribution platform is according to the gender and/or age information and/or fat or thin information of current passenger
And/or it wears information sifting and goes out matching advertisement and the advertisement filtered out is ranked up according to matching degree and is sequentially broadcast
It puts.
Further, the passenger identification systems include camera and control analysis module, and the camera absorbs passenger
Image information obtains passenger's human face image information from image information, wears information, and control analysis module is based on recognition of face skill
Art obtains human face characteristic point from human face image information and carries out gender and/or obtaining property of character classification by age analysis according to human face characteristic point
Not and/or age information, detailed process include:
After the human face image information for obtaining passenger, since in the image of most of reality scene, it is big that there is faces
It is small different, uneven illumination weighing apparatus, posture angle not parity problem, in order to eliminate these problems bring adverse effect, it is necessary to figure
As being pre-processed, the present embodiment pretreatment is largely divided into 5 steps: utilizing Haar feature cascade classifier to the image of input first
Face is detected, position of human eye is positioned, obtains eyes distance and eyes tilt angle；Then two dimensional affine change is done using the angle
It changes, rotates face, solve the different influence of posture angle；Face part is normalized again, avoids face size not
One influence；Brightness normalizing finally is carried out using histogram equalization, noise is eliminated using smoothing processing, it is equal to reach human face light
The effect of weighing apparatus.The facial image with uniform characteristics is obtained after above-mentioned pretreatment, is extracted work for subsequent characteristics and is done
Prepare.
Gabor wavelet transformation is carried out to pretreated image, Gabor wavelet all has good office in frequency domain and airspace
Portion's property can fully describe image texture, for extracting local feature significant effect；It belongs to adding window Fourier transformation, is Gabor
The product that theoretical and wavelet theory combines, two-dimensional Gabor function can be expressed as
Wherein,U, v are direction and the scale of core, determine Gaussian window
Size；Z (x, y) is direction vector, and x, y are two-dimensional coordinate；kU, vIt is maximum sample frequency for the centre frequency of filter；f
For sampling step length；σ is Gauss window width and wavelength ratio, and formula (1) effectively defines one group of self similarity wavelet function race, parameter
Different choosing values are presented as rotation and the change of scale of Gabor filter, and the Gabor wavelet for being used in recognition of face direction becomes
It changes, following selection parameter effect is best: kmax takesF takesσ takes8 directions (u=0,1 ..., 7), 5 scale v
=(0,1,2,3,4), the Gabor wavelet transformation of face are to be filtered with the Gabor of facial image and different scale different directions
Device convolution.F (x, y) is image grayscale distribution, that is, is had
G (x, y, u, v)=f (x, y) * ψU, v(z)
Facial image and 5 scales after 40 Wavelet Kernel whole convolution in 8 directions, will obtain 40 filtering figures
Violent variation tendency is presented in certain parts of picture, face on specific scale and direction, converts it by Gabor wavelet
Afterwards, these variations can be showed intuitively very much.Therefore former by after the convolution nuclear convolution in different scales and direction
Corresponding part has very strong response in beginning input picture.
In small scale, eyes, nose, mouth in face just have very high response；And in large scale, the wheel of face
Exterior feature seems more prominent.So far three-dimensional matrice of the available facial image after Gabor wavelet transformation, can be by three-dimensional square
Battle array sequence storage is at one-dimensional feature vector.If the image pattern resolution ratio of input is M × N, the feature vector of the sample
Dimension is M × N × 40, needs to carry out dimension-reduction treatment to it using certain methods, avoids the occurrence of dimension crisis.
5 scales obtained above, there are bulk redundancies for the feature vector in 8 directions, but the spy on most of directions
Sign is not obvious, therefore using the method for weight matrix, by assigning certain weight to feature on direction, so that in certain rulers
The feature on direction under degree is more prominent, and significantly reduces dimension.The calculating of weight matrix uses gradient Nearest Neighbor with Weighted Voting
Method, it is specific as follows: to divide the image into the connected region Cell not overlapped and the connected region Block, such as Fig. 2 that have overlapping
It is shown, wherein horizontal line grid rectangle is a Cell, and cross grid frame rectangle is a Block, and it is 2 that oblique line, which fills rectangle,
The Cell being overlapped between Block.
In this algorithm, the size for defining Cell is 8 × 8, and unit is pixel；The size of Block is 2 × 2, and unit is
Cell.Due in gradient orientation histogram (histograms of oriented gradients, HOG) algorithm and SIFT algorithm
In Nearest Neighbor with Weighted Voting be undirected nine channel, here in order to which 8 directions that same Gabor wavelet is decomposed are consistent, need π point
At 8 parts, the angle size in each direction are as follows:Its gradient direction is counted for each Cell, into
Nearest Neighbor with Weighted Voting of row, weight are the modulus value of the gradient, obtain the directional statistics information of each Cell in this way.In order to eliminate office
The influence of portion's illumination, need the region Block to comprising the region Cell be normalized, normalization algorithm use L2- norm,
Because different Block has the part of overlapping, i.e. majority Cell can be expressed repeatedly, as shown in Figure 3.In Fig. 3, each square
Indicate that the gradient statistical information of a Cell, x, y-axis respectively represent the transverse and longitudinal coordinate axis of image, z-axis represents what Cell was expressed
Number.For piece image, the Cell of quadrangle is expressed 1 time, and the Cell on boundary is expressed 2 times (not including quadrangle), in
Between be expressed 4 times, this is because normalized Cell is relative to the Block belonging to it every time, Block difference is to its institute
The normalization result of the Cell of category also can be different.This 4 layers statistical information can be indicated with a three-dimensional matrice, then right
Each Cell carries out a global normalization, then takes the channel mean value to merge all statistical informations of relative position, then
Three-dimensional matrice just will become a common two-dimensional matrix, and the purpose that this part calculates is that eight channel weightings of some Cell are thrown
The directive weighting coefficient of institute that ticket result is decomposed as Gabor wavelet, believes so to count to eight channels of the Cell acquired
Breath carries out a L again1- norm normalization.
So far there are 8 weights in each region Cell, and entire picture can be reserved for as a weight matrix
Wherein, m is the number of Cell, ti(i=1 ..., m) be composed by the weight in 8 directions of i-th of Cell to
Amount, the flow chart of the above process are as shown in Figure 4.
The feature vector for classification is obtained by weight matrix weighted array to the feature vector that Gabor is extracted, specifically
Process is as follows.It can be seen from the above, image is after Gabor transformation, 5 scales are obtained to each pixel, 8 directions
40 convolution values, due to weight matrix be obtained as unit of Cell, so the value after Gabor transformation also can be regarded as with
Cell is that unit is stored in a three-dimensional matrice, and when storage continuously saves the numerical value for belonging to same scale, facilitates following
Weighted calculation can be expressed as a pixel in i-th of Cell by the value after Gabor transformation under j-th of scale
Gj8=(gj1..., gj8)
Wherein gjk(k=1 ..., 8) is the vector composed by the value in 8 directions under jth (j=1 ..., 5) a scale,
Then using weights matrix is weighted combination to features described above matrix, i.e., with 8 weights of i-th of Cell with obtain i-th
The value in 8 directions in a Cell under each scale of each pixel makees inner product.
F=Gj8·(ti)T, i=1 ..., m
Direction value of each pixel under 5 scales in available i-th of Cell, to each picture in all Cell
Element all does same processing, an available new three-dimensional matrice, and the number of plies of this three-dimensional matrice is reduced from U × V layers
To V layers, (U is general direction number, and V is out to out number, and wherein U takes 8, V to take 5), while face characteristic can be protected to greatest extent
It stays.In addition to this, in addition it is also necessary to size adjusting be carried out to image, adjusted with linear interpolation method to small size, i.e., by original M × N's
Dimension is fallen below (M × N)/5 from the dimension of original M × N × 40 and tieed up by Image Adjusting to (M/5) × (N/5), final the method, most
Journal matrix afterwards forms one-dimensional feature vector, so as to complete the weighted array feature converted based on Gabor wavelet
It extracts.
By the face character feature vector of said extracted, sample set D is constituted, is existed using tree classifier
The lane database of WebFace or FaceScrub or Adience is trained and tests, and utilizes tree classifier processing higher-dimension
Data select the feature more important to gender and/or character classification by age to obtain required gender and/or age information.
Tree classifier is that Leo breiman is proposed, is made of k classification tree, its basic thought is will be multiple
Weak Classifier is integrated into a strong classifier.Classification tree is made of root node, internal node and leaf node.Wherein, root node
Training set is represented, each internal node represents Weak Classifier, and by sample according to a certain Attribute transposition, each leaf node is that have
Input data is categorized into several subsets by the training of label or test set.The last result of decision of tree classifier is all
Classification tree is voted the optimal result of selection, is mainly comprised the steps that
1) it is extracted with putting back to from original sample collection D using bootstrap repeat replication, k-th of sample set is remembered
Make Dk, and generates random vector θ for kth classification treek, θkWith the random vector independent same distribution of front.Use h (Dk, θk) table
Show kth classification-tree method；
2) classification tree is established to k sample respectively.The generation of classification tree is exactly the process of recurrence building binary class tree, is made
Binary tree is divided with the smallest feature of gini index.Assuming that input sample DkIt is the feature vector of M dimension, in each node of tree
Place, m (m < < M) a feature is randomly selected from this M feature and, as candidate feature, is then chosen from this m feature optimal
Feature and best two-value cut-off divide the node.Step 2) is repeated, k classification tree is established.
3) according to each classification tree as a result, choosing last classification results in a vote.Construct the process of tree classifier
As shown in Figure 5.
Further, control analysis module is based on face characteristic and figure aspectual character and standard faces size and figure wheel
The wide fat or thin information of size comparative analysis, control analysis module wear the brand message of dress ornament clothes based on information analysis is worn to obtain
The brand of clothes, packet of passenger etc. can be recognized by wearing information for example.
Further, the ad distribution platform includes memory module, and the memory module includes several according to gender
And/or age information and/or fat or thin information and/or the storage section for wearing information setting, the ad datas of several species according to
Gender and/or age information and/or fat or thin information and/or wear information classified storage several storage sections in, store mould
Root tuber according to gender and/or age information and/or fat or thin information and/or wear information with several storage sections matched, and
And the advertisement for transferring the storage section of pairing plays out.
Further, positioning system is additionally provided in the vehicles, the positioning system obtains working as the vehicles
The geographical location information is simultaneously sent to external server by preceding geographical location information, and the external server is according to the geography
The information for the advertiser that it is N kilometers as the center of circle, radius using the geographical location that location information, which is obtained, the external server is by institute
The information for stating advertiser is sent to ad distribution platform, and the memory module is according to gender and/or age information and/or fat or thin letter
It ceases and/or wears information and matched with several storage sections, and transfer the advertisement of the advertiser in the storage section of pairing
It plays out.
Further, the vehicles are net about vehicles or taxi.
Further, the positioning system is GPS system.
Further, the ad distribution platform includes being arranged in net about vehicle or taxi front-row seats back and/or pair
The display module such as liquid crystal display on storage tank (namely glove box) is driven, can guarantee multiplying in the passenger side or heel row in this way
Visitor is it can be seen that advertisement.When display module is arranged on copilot storage tank, a support frame can be set to fix branch
Display module is supportted, support frame as described above has elevating function, and when vehicle normally travel, support frame as described above drives display module to rise one
Fixed height, in this way vacates the position of the passenger side air bag, avoids the occurrence of the case where display module has blocked air bag；Or
The position for keeping left or keeping right display module can also be arranged in, air bag is arranged in such face passenger middle position, in this way
One substantially increases travel safety, can smoothly pop up when there is accident safety airbag, display module will not be formed it
It hinders.
Further, the advertisement is played with the mode of video by display module.
Further, the value range of the N are as follows: 30 >=N >=1, the N can be specifically chosen according to actual needs.
Further, the address in the advertisement including advertiser and/or the information apart from current geographic position distance.
Further, Customer information variation (having changed a collection of passenger) Shi Huichong that the method is captured whenever camera
It is new to obtain Customer information and push matching advertisement again, it can preferably refresh primary new geographical position every t minutes
It sets the new order ads of acquisition or refreshes primary new geographical location after current advertisement plays and obtain new advertisement row
Sequence guarantees not weigh as far as possible with the advertisement that mobile the played advertisement of the vehicles also can be updated and be played therewith in this way
It is multiple.
Further, the address in the advertisement including advertiser or the information apart from current geographic position distance, such as
The address of businessman and the information of distance of the address apart from passenger can be provided in advertisement video, allow passenger more to have in this way
The businessman impression very close from oneself, if the idea that passenger was interested in or produced impulsion consumption to advertisement so goes to quotient
It is very easy to accomplish that family, which is experienced or consumed,.
The present invention introduces passenger identification systems on the basis of existing technology, can identify passenger's in a vehicle
Gender and/or age information and/or fat or thin information and/or information is worn, is then matched according to different passengers and push is corresponding
Advertisement, substantially increase receiving and ratings of the passenger to advertisement in this way, to improve the effect of publicity of advertisement, such as
The type advertisements such as automobile, house property, game can be pushed for male, the types such as cuisines, beauty, dress ornament can be pushed for women
Advertisement can push cuisines, body-building, weight-reducing advertisement for obese people, can identify passenger to a certain degree for information is worn
Consuming capacity level (such as brand of the identification clothes of passenger, packet) advertisement of corresponding level is pushed according to passenger's consuming capacity
(such as passenger's clothes and packet are all the advertisements that famous brand or luxury brand just push same class or more high-grade brand)；The present invention
Also introduce positioning system on the basis of existing technology, the advertisement so played will no longer be it is pre- set it is constant
Sequentially, but with current location apart from very close businessman, such passenger is very possible if generating interest to ad content
Businessman's experience or consumption can be gone to, because distance is very close, passenger goes to businessman and is also easily easily achieved, to pass through this
The method of invention substantially increases advertising effect and practical changing effect；Since the present invention has preferable advertising results,
Can be very good that businessman is attracted to launch advertisement, ad distribution platform can cooperate with such as net about vehicle platform, to fare into
The certain subsidy of row (or even realize and freely call a taxi), it is final preferential to passenger in this way, then passenger will preferably net about Che Pingtai
Trip has very big prospect to realize an extraordinary benign business chain.
Basic principles and main features and advantage of the invention have been shown and described above.The technical staff of the industry should
Understand, the present invention is not limited to the above embodiments, and the above embodiments and description only describe originals of the invention
Reason, without departing from the spirit and scope of the present invention, various changes and improvements may be made to the invention, these changes and improvements
It all fall within the protetion scope of the claimed invention.The claimed scope of the invention is by appended claims and its equivalent circle
It is fixed.
Claims (10)
1. a kind of method by vehicles advertisement based on passenger's identification is provided with advertisement hair in the vehicles
Cloth platform and passenger identification systems, it is characterised in that the following steps are included:
Step (1): the passenger identification systems obtain current passenger gender and/or age information and/or fat or thin information and/or
Wear information；
Step (2): the passenger identification systems are by the gender of current passenger and/or age information and/or fat or thin information and/or wear
Information be sent to ad distribution platform；
Step (3): the ad distribution platform according to the gender and/or age information of current passenger and/or fat or thin information and/or
Information sifting is worn to go out matching advertisement and be ranked up the advertisement filtered out according to matching degree and sequentially play.
2. the method by vehicles advertisement according to claim 1 based on passenger's identification, which is characterized in that
The passenger identification systems include camera and control analysis module, and the camera absorbs passenger image information, are believed from image
Passenger's human face image information is obtained in breath, wears information, and control analysis module is based on face recognition technology from human face image information
It obtains human face characteristic point and gender and/or age information is obtained according to human face characteristic point progress gender and/or character classification by age analysis,
It is fat based on face characteristic and figure aspectual character and standard faces size and figure profile size comparative analysis to control analysis module
Thin information, control analysis module wear information based on the brand for wearing information analysis clothing dress ornament to obtain.
3. the method by vehicles advertisement according to claim 2 based on passenger's identification, which is characterized in that
Control analysis module is based on face recognition technology and obtains human face characteristic point from human face image information and carried out according to human face characteristic point
Gender and/or character classification by age analysis show that the process of gender information includes:
After the human face image information for obtaining passenger, image is pre-processed: Haar feature being utilized to the image of input first
Cascade classifier detects face, positions position of human eye, obtains eyes distance and eyes tilt angle；Then two are done using the angle
Affine transformation is tieed up, face is rotated, solves the different influence of posture angle；Face part is normalized again, avoids people
Face influence not of uniform size；Brightness normalizing finally is carried out using histogram equalization, noise is eliminated using smoothing processing, reaches people
The effect of face illumination equilibrium is to obtain the facial image with uniform characteristics；
Gabor wavelet transformation is carried out to pretreated image, two-dimensional Gabor function representation is
Wherein,U, v are direction and the scale of core；Z (x, y) is direction vector, and x, y are
Two-dimensional coordinate；ku,vFor the centre frequency of filter；F is sampling step length；σ be Gauss window width and wavelength ratio, face
Gabor wavelet transformation is with the Gabor filter convolution of facial image and different scale different directions, and f (x, y) is that image is grey
Degree distribution, that is, have
G (x, y, u, v)=f (x, y) * ψU, v(z)
Facial image and 5 scales after 40 Wavelet Kernel whole convolution in 8 directions, will obtain 40 filtering images, until
Three-dimensional matrice of this available facial image after Gabor wavelet transformation, can store three-dimensional matrice sequence at one-dimensional
Feature vector, the image pattern resolution ratio of input is M × N, then feature vector dimension is M × N × 40,5 obtained above
Scale, there are bulk redundancies for the feature vector in 8 directions, but the feature on most of directions is not obvious, therefore exploitation right
The method of value matrix, by assigning certain weight to feature on direction, so that the feature on the direction under certain scales is more
It is prominent, and dimension is significantly reduced, the method that the calculating of weight matrix uses gradient Nearest Neighbor with Weighted Voting is specific as follows: will to scheme
As the connected region Block for being divided into the connected region Cell not overlapped and having overlapping, wherein horizontal line grid rectangle is one
Cell, cross grid frame rectangle are a Block, and oblique line filling rectangle is the Cell being overlapped between 2 Block；
The size for defining Cell is 8 × 8, and unit is pixel；The size of Block is 2 × 2, and unit is Cell, due in gradient side
Nearest Neighbor with Weighted Voting into histogramming algorithm and SIFT algorithm is undirected nine channel, 8 decomposed here for same Gabor wavelet
Direction is consistent, and needs for π to be divided into 8 parts, the angle size in each direction are as follows:For each
Cell counts its gradient direction, carries out a Nearest Neighbor with Weighted Voting, and weight is the modulus value of the gradient, obtains the side of each Cell in this way
To statistical information, the region Block to comprising the region Cell be normalized, normalization algorithm uses L2- norm is then right
Each Cell carries out a global normalization, then takes the channel mean value to merge all statistical informations of relative position, right
The eight channel statistical informations of Cell carry out a L again1- norm normalization, so far there are 8 weights in each region Cell, entirely
Picture can be reserved for as a weight matrix
Wherein, m is the number of Cell, ti(i=1 ..., m) is vector composed by the weight in 8 directions of i-th of Cell；
The feature vector for classification, detailed process are obtained by weight matrix weighted array to the feature vector that Gabor is extracted
It is as follows: it can be seen from the above, image is after Gabor transformation, 5 scales are obtained to each pixel, 40 of 8 directions
Convolution value, since weight matrix is obtained as unit of Cell, so the value after Gabor transformation also can be regarded as with Cell
It is stored in a three-dimensional matrice for unit, when storage continuously saves the numerical value for belonging to same scale, in i-th of Cell
A pixel, can be expressed as by the value after Gabor transformation under j-th of scale
Gj8=(gj1..., gj8)
Wherein gjk(k=1 ..., 8) is the vector composed by the value in 8 directions under jth (j=1 ..., 5) a scale, then
Using weights matrix is weighted combination to features described above matrix, i.e., with 8 weights of i-th of Cell and obtain i-th
The value in 8 directions in Cell under each scale of each pixel makees inner product:
F=Gj8·(ti)T, i=1 ..., m
Direction value of each pixel under 5 scales in i-th of Cell is obtained, each pixel in all Cell is done together
The processing of sample obtains a new three-dimensional matrice, and the number of plies of this three-dimensional matrice is reduced from U × V layers to V layers, and wherein U is
General direction number, V are out to out number, then carry out size adjusting to image, are adjusted with linear interpolation method to small size, i.e., will be original
Dimension is finally fallen below (M × N)/5 from the dimension of original M × N × 40 and tieed up, most by the Image Adjusting of M × N to (M/5) × (N/5)
Journal matrix afterwards forms one-dimensional feature vector, so as to complete the weighted array feature converted based on Gabor wavelet
It extracts；
By the face character feature vector of said extracted, constituted sample set D, using tree classifier in WebFace or
The lane database of FaceScrub or Adience is trained and tests, and using the data of tree classifier processing higher-dimension, selects
The feature more important to gender and/or character classification by age is to obtain required gender and/or age information.
4. the method by vehicles advertisement according to claim 2 based on passenger's identification, which is characterized in that
The ad distribution platform includes memory module, the memory module include several according to genders and/or age information and/or
Fat or thin information and/or the storage section for wearing information setting, the ad data of several species is according to gender and/or age information
And/or fat or thin information and/or wear information classified storage several storage sections in, memory module is according to gender and/or year
It age information and/or fat or thin information and/or wears information and is matched with several storage sections, and transfer the storage area of pairing
Between advertisement play out.
5. the method by vehicles advertisement according to claim 4 based on passenger's identification, which is characterized in that
Positioning system is additionally provided in the vehicles, the positioning system obtains the current geographic position information of the vehicles and will
The geographical location information is sent to external server, and the external server is obtained according to the geographical location information with described
The information for the advertiser that geographical location is the center of circle, radius is N kilometers, the external server send the information of the advertiser
To ad distribution platform, the memory module according to gender and/or age information and/or fat or thin information and/or wear information with
Several storage sections are matched, and the advertisement for transferring the advertiser in the storage section of pairing plays out.
6. the method by vehicles advertisement according to claim 1-5 based on passenger's identification,
It is characterized in that, the vehicles are net about vehicles or taxi.
7. the method by vehicles advertisement according to claim 6 based on passenger's identification, which is characterized in that
The positioning system is GPS system.
8. the method by vehicles advertisement according to claim 7 based on passenger's identification, which is characterized in that
The ad distribution platform includes the display being arranged on net about vehicle or taxi front-row seats back and/or copilot storage tank
Module, the advertisement are played with the mode of video by display module.
9. the method by vehicles advertisement according to claim 6 based on passenger's identification, which is characterized in that
The value range of the N are as follows: 30 >=N >=1.
10. the method by vehicles advertisement according to claim 6 based on passenger's identification, feature exist
In the address in the advertisement including advertiser and/or the information apart from current geographic position distance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910325858.8A CN110135894A (en) | 2019-04-23 | 2019-04-23 | A method of vehicles advertisement is passed through based on passenger's identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910325858.8A CN110135894A (en) | 2019-04-23 | 2019-04-23 | A method of vehicles advertisement is passed through based on passenger's identification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110135894A true CN110135894A (en) | 2019-08-16 |
Family
ID=67570622
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910325858.8A Pending CN110135894A (en) | 2019-04-23 | 2019-04-23 | A method of vehicles advertisement is passed through based on passenger's identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110135894A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705391A (en) * | 2019-09-17 | 2020-01-17 | 四川大学锦城学院 | Seat distribution system based on human body image recognition method |
| CN111160962A (en) * | 2019-12-20 | 2020-05-15 | 恒银金融科技股份有限公司 | Micro-expression recognition marketing pushing method and system |
| CN111382353A (en) * | 2019-12-27 | 2020-07-07 | 上海学人信息科技有限公司 | Accurate information advertisement matching and releasing device and system for passengers in vehicle |
| CN112132616A (en) * | 2020-09-23 | 2020-12-25 | 范玲珍 | Mobile multimedia advertisement intelligent pushing management system based on big data |
| CN113034169A (en) * | 2019-12-25 | 2021-06-25 | 丰田自动车株式会社 | Information processing apparatus, information processing system, storage medium, and information processing method |
| CN115829655A (en) * | 2022-12-30 | 2023-03-21 | 广州掌购网络科技有限公司 | Advertisement effect intelligent analysis management system and method based on big data analysis |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102902986A (en) * | 2012-06-13 | 2013-01-30 | 上海汇纳网络信息科技有限公司 | Automatic gender identification system and method |
| CN103747086A (en) * | 2014-01-10 | 2014-04-23 | 江苏矽望电子科技有限公司 | Intelligent bus on-board advertisement management system based on cloud platform |
| CN103942705A (en) * | 2014-03-25 | 2014-07-23 | 惠州Tcl移动通信有限公司 | Advertisement classified match pushing method and system based on human face recognition |
| CN105046513A (en) * | 2015-06-19 | 2015-11-11 | 长沙待霁电子科技有限公司 | Vehicle-mounted regionalization intelligence positioning advertisement method |
| CN107301820A (en) * | 2017-07-18 | 2017-10-27 | 广东长虹电子有限公司 | It is a kind of to recognize the intelligent advisement player and its control method of spectators' type |
| CN108197988A (en) * | 2017-12-29 | 2018-06-22 | 厦门理工学院 | Underground tunnel advertisement control method for playing back and its equipment, storage medium and system based on face recognition technology |
| CN109118266A (en) * | 2018-06-27 | 2019-01-01 | 北京蜂盒科技有限公司 | Advertisement sending method, device, storage medium and electronic equipment |
-
2019
- 2019-04-23 CN CN201910325858.8A patent/CN110135894A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102902986A (en) * | 2012-06-13 | 2013-01-30 | 上海汇纳网络信息科技有限公司 | Automatic gender identification system and method |
| CN103747086A (en) * | 2014-01-10 | 2014-04-23 | 江苏矽望电子科技有限公司 | Intelligent bus on-board advertisement management system based on cloud platform |
| CN103942705A (en) * | 2014-03-25 | 2014-07-23 | 惠州Tcl移动通信有限公司 | Advertisement classified match pushing method and system based on human face recognition |
| CN105046513A (en) * | 2015-06-19 | 2015-11-11 | 长沙待霁电子科技有限公司 | Vehicle-mounted regionalization intelligence positioning advertisement method |
| CN107301820A (en) * | 2017-07-18 | 2017-10-27 | 广东长虹电子有限公司 | It is a kind of to recognize the intelligent advisement player and its control method of spectators' type |
| CN108197988A (en) * | 2017-12-29 | 2018-06-22 | 厦门理工学院 | Underground tunnel advertisement control method for playing back and its equipment, storage medium and system based on face recognition technology |
| CN109118266A (en) * | 2018-06-27 | 2019-01-01 | 北京蜂盒科技有限公司 | Advertisement sending method, device, storage medium and electronic equipment |
Non-Patent Citations (4)
| Title |
|---|
| 刘劲松: "《专业学位研究生实验课程 工科篇》", 30 November 2017 * |
| 宓超等: "《装卸机器视觉及其应用》", 31 January 2016 * |
| 申蔚等: "《虚拟现实技术》", 30 September 2002 * |
| 韩志艳: "《面向语音与面部表情信号的多模式情感识别技术研究》", 31 January 2017 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705391A (en) * | 2019-09-17 | 2020-01-17 | 四川大学锦城学院 | Seat distribution system based on human body image recognition method |
| CN110705391B (en) * | 2019-09-17 | 2023-09-19 | 四川大学锦城学院 | Seat distribution system based on human body image recognition method |
| CN111160962A (en) * | 2019-12-20 | 2020-05-15 | 恒银金融科技股份有限公司 | Micro-expression recognition marketing pushing method and system |
| CN113034169A (en) * | 2019-12-25 | 2021-06-25 | 丰田自动车株式会社 | Information processing apparatus, information processing system, storage medium, and information processing method |
| CN111382353A (en) * | 2019-12-27 | 2020-07-07 | 上海学人信息科技有限公司 | Accurate information advertisement matching and releasing device and system for passengers in vehicle |
| CN112132616A (en) * | 2020-09-23 | 2020-12-25 | 范玲珍 | Mobile multimedia advertisement intelligent pushing management system based on big data |
| CN112132616B (en) * | 2020-09-23 | 2021-06-01 | 广东省广汽车数字营销有限公司 | Mobile multimedia advertisement intelligent pushing management system based on big data |
| CN115829655A (en) * | 2022-12-30 | 2023-03-21 | 广州掌购网络科技有限公司 | Advertisement effect intelligent analysis management system and method based on big data analysis |
| CN115829655B (en) * | 2022-12-30 | 2023-09-22 | 广州掌购网络科技有限公司 | Advertisement effect intelligent analysis management system and method based on big data analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110135894A (en) | A method of vehicles advertisement is passed through based on passenger's identification | |
| CN110111138A (en) | A method of passing through vehicles advertisement | |
| Richards et al. | Using image recognition to automate assessment of cultural ecosystem services from social media photographs | |
| CN106815601B (en) | Hyperspectral image classification method based on recurrent neural network | |
| CN105718873B (en) | Stream of people's analysis method based on binocular vision | |
| CN103578119B (en) | Target detection method in Codebook dynamic scene based on superpixels | |
| CN111898523A (en) | Remote sensing image special vehicle target detection method based on transfer learning | |
| CN109344736A (en) | A kind of still image people counting method based on combination learning | |
| CN105809198B (en) | SAR image target recognition method based on depth confidence network | |
| CN106991382A (en) | A kind of remote sensing scene classification method | |
| Yang et al. | Superpixel-based unsupervised band selection for classification of hyperspectral images | |
| CN109002492B (en) | Performance point prediction method based on LightGBM | |
| DE102012111303A1 (en) | Usage measurement methods and systems for interactive advertising | |
| CN106815604A (en) | Method for viewing points detecting based on fusion of multi-layer information | |
| CN108647595B (en) | Vehicle weight identification method based on multi-attribute depth features | |
| CN107657271B (en) | Hyperspectral image classification method based on long memory network in short-term | |
| CN105184298A (en) | Image classification method through fast and locality-constrained low-rank coding process | |
| CN102799635A (en) | Image set ordering method driven by user | |
| CN109492593A (en) | Hyperspectral image classification method based on principal component analysis network and space coordinate | |
| CN107895139A (en) | A kind of SAR image target recognition method based on multi-feature fusion | |
| US20220292328A1 (en) | Convolutional artificial neural network based recognition system in which registration, search, and reproduction of image and video are divided between and performed by mobile device and server | |
| Yu et al. | AI-based targeted advertising system | |
| Hussein et al. | A comprehensive evaluation framework and a comparative study for human detectors | |
| CN108460400A (en) | A kind of hyperspectral image classification method of combination various features information | |
| CN110288368A (en) | A kind of method that compartmentalization passes through vehicles advertisement |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190816 |
|
| RJ01 | Rejection of invention patent application after publication |