CN104915000A - Multisensory biological recognition interaction method for naked eye 3D advertisement - Google Patents

Abstract

The present invention relates to a multi-sensory biometric interaction method for naked-eye 3D advertisements, the steps of which are: (1) human body detection in the scene: in the construction of a platform for automatically delivering targeted advertisement content, the camera monitors the human body in a fixed scene Monitoring and detection of moving human body; (2) crowd feature detection: extract the face features, hairstyle features and morphological features of the crowd to distinguish the age and gender of different crowds; (3) crowd intelligent recognition and classification: use the extracted crowd face Features and morphological characteristics, intelligently identify and classify the gender and age of the crowd; (4) Advertisement recommendation decision: According to the results of the crowd classification, different advertisements are classified and placed, and personalized recommendation technology is used to deliver different information to People who want to be designated; ⑸Intelligent human-computer interaction: Through somatosensory technology and voice recognition technology, the captured human motion and voice signals are converted into control signals to carry out human-computer intelligent interaction between people and advertising machines.

Description

Multi-sensory biometric interaction method for naked-eye 3D advertisement

technical field

The invention belongs to the field of intelligent identification, in particular to a multi-sensory biological identification interaction method for naked-eye 3D advertisements.

Background technique

3D display technology, especially glasses-free 3D technology and products are promoting a technological revolution from flat panel display to stereoscopic display. International research has been in full swing, such as Sony, Samsung, and LG companies have launched glasses-free 3D products. At present, my country's 3D industry is still in the early stage of development, and the naked eye 3D industry has just started. Although many domestic companies have actively invested in the research and development of naked eye 3D technology and products, the development time of the naked eye 3D technology market in China is short, and there is no professional and formal scientific research. At the same time, because stereoscopic imaging technology is a cross-border technology, there are very few people who study it. The price of naked-eye 3D products in the market is relatively high, and the market is in the stage of education and cultivation. Among the glasses-free 3D devices currently on the market, the parallax barrier 3D technology is a relatively common type. This technology is mostly used in small-sized devices such as 3D mobile phones (Sharp SH8168U) and portable game consoles (Nintendo 3DS). Toshiba's 56-inch 3D TV uses lenticular lens technology, because the TV has higher requirements on brightness and resolution, and the former is still difficult to meet the harsh application environment.

Glasses-free 3D advertising machine uses the parallax characteristic of human eyes to obtain realistic three-dimensional images with space and depth without any auxiliary equipment (such as 3D glasses, helmets, etc.). Things in the painting can protrude outside the picture, or can be hidden deep in the picture. Colorful, distinct, vivid and lifelike, it is a true three-dimensional image. Glasses-free three-dimensional images are favored by consumers for their true and vivid expressive power, elegant and elegant environmental appeal, and strong and shocking visual impact. This new, unique, and unique image product is widely used in advertising media, exhibitions, tourism investment, wedding photography, scientific research and teaching, game entertainment, industrial applications, architectural design, mobile phones and other industries.

Digital advertising machine (hereinafter referred to as advertising machine) is a new media concept, which refers to a multimedia professional audio-visual system that releases commercial, financial or entertainment information through large-screen terminal display equipment in shopping malls, airports and other public places where other people gather. As a substitute for traditional outdoor advertising, the advertising machine contains a large amount of information, can be easily updated at any time, and can be connected to the Internet, which is unmatched by traditional outdoor advertising. Advertising machines can not only release multimedia advertising information to the public and bring new commercial innovation value, but also are considered to be the driving force for the new growth of the display industry. After nearly five years of development, the advertising machine has gone through several stages such as stand-alone, network, and interactive, and is gradually moving towards intelligence. Shi Yangwei, director of Intel China, said: "Traditional content and advertisements based on graphic representations are becoming less and less attractive to consumers, and interactive media communication and marketing methods will soon become the mainstream."

As the "fifth media", the intelligent naked-eye 3D advertising machine products will provide an indispensable technical platform for the realization of interactive media communication applications. "Intelligent naked-eye 3D advertising machine means that the terminal display device automatically launches targeted naked-eye 3D advertising information according to the identified characteristics of different consumers. Based on the analysis of this trend, Intel believes that: intelligent naked-eye 3D advertising machine system Not only must it be able to support clearer and vivid naked-eye 3D images to attract consumers, it must have a more friendly touch-screen interactive operation interface, but it also needs to be able to identify consumer characteristics and automatically launch customized advertising information in real time. The realization of these applications requires Supported by powerful processor technology, hardware platform and software solutions.

With the development of naked-eye 3D advertising machine Internet technology, new applications, and the growth of massive data that needs to be processed, the naked-eye 3D advertising machine processing solution based on the "intelligent interactive platform" is particularly important. The hardware resources of smart naked-eye 3D advertising machines can meet the needs of different application modes. At the same time, hardware such as controllers and displays developed for the naked-eye 3D advertising machine industry is also becoming more and more abundant, and the relatively weak link is the corresponding software solutions. That is, intelligent recognition technology. Although some intelligent recognition technologies have been reported internationally, most of them are in the laboratory research and development stage, and the person objects and environments considered are relatively single. However, in fact, the application field of naked-eye 3D advertising machines is a completely open and complex environment, and the service objects and environments are extremely uncertain: the number of people entering the camera's field of view, gender and age distribution, standing or walking posture, etc. cannot be restricted. The environment in front of the camera, such as vehicles passing through, changes in light and shade, and weather effects, are also uncontrollable. All these bring difficulties to human body detection, tracking and identification. In addition, some existing recognition technologies are mostly based on the facial features of people, and a few are based on hair color and hearing features. However, the above-mentioned characteristics of the naked-eye 3D advertising machine application environment make it almost impossible to achieve accurate detection of a single feature, so it is not possible to rely solely on a few characteristics to make a decision.

In addition, human-computer interaction technologies such as somatosensory technology and speech recognition technology have emerged one after another. Through somatosensory sensors and voice collectors, the movements or voices of the human body are captured, and the work or voice of the human body is converted into corresponding instructions through information processing technology. It is of great significance to control the playback of intelligent advertising machines and perform corresponding human-computer interaction.

Contents of the invention

The purpose of the present invention is to fill in the gaps in the prior art, and provide a multi-sensory biometric interaction method for naked-eye 3D advertisements that extracts and fuses multiple features and has intelligent recommendation technology.

The technical solution adopted by the present invention to solve its technical problems is:

A multi-sensory biometric interaction method for naked-eye 3D advertisements, the steps of which are:

⑴Human detection in the scene: In the construction of the platform for automatic delivery of targeted advertising content, the camera monitors and detects the moving human body in the fixed scene. The moving human body detection includes human body detection and tracking, that is, detection and tracking occur in The crowd in the video; the number of people, that is, the statistics of the number of crowd in the video;

(2) Crowd feature detection: Extract the face features, hairstyle features, voice features and morphological features of the crowd to distinguish the age and gender of different groups;

(3) Crowd intelligent identification and classification: use the extracted characteristics of the crowd to intelligently identify and classify the gender and age of the crowd;

⑷ Advertisement recommendation placement decision: According to the results of crowd classification, different advertisements are classified and placed, and personalized recommendation technology is used to deliver different information to the desired crowd;

⑸Intelligent human-computer interaction: Through somatosensory technology and voice recognition technology, the captured human body movements and voice signals are converted into control signals to carry out human-computer intelligent interaction between people and advertising machines.

Moreover, the human body detection and tracking described in step (1) adopt static image human body detection, which adopts Haar-like features, EOH features, texture features, and color features in feature selection. In the design of classifiers, it uses statistical Learn how to classify.

Moreover, the number of people described in step (1) adopts a dynamic update model, and uses a Gaussian distribution to describe the probability distribution of the background point color, and the detection of a new moving human body is based on the comparison between the current frame image and the background model established by the mixed Gaussian model , to obtain all moving human bodies in the foreground image of the current frame, and then compare with the recorded moving human body data. In this way, many moving objects in the current frame are obtained, and the smaller moving objects that overlap with the tracked human body are discarded, and the remaining parts are sorted by size, and then 10 largest human bodies are newly added. Come and keep track.

Moreover, the face feature and morphological feature extraction described in step (2) is divided into two ways: overall feature extraction and local feature extraction. The overall feature extraction algorithm roughly recognizes the face, and the local feature extraction supplements the details.

And, the main steps of the hairstyle feature extraction described in step (2) are as follows:

a. Obtain a frame of image from the video for detection, acquire images at regular intervals or acquire a single image at regular intervals for processing when a person in the scene has been detected.

b. Image preprocessing.

c, hairstyle segmentation.

d. Formation of hairstyle model.

Moreover, the crowd intelligent identification and classification described in step (3) uses multiple classifiers to realize decision fusion.

And, the steps of the algorithm of the gender identification described in step (3) are as follows:

a. Various features such as human face, skin color, hairstyle, decoration, etc. are extracted, and facial features include overall and local features;

b. Use the linear discriminant method to extract the characteristics of the human body detected by training, and use the Fisher method to extract the characteristics of the training samples and target images;

c, utilize the dynamic clustering method to classify the target image, and calculate the average recognition rate of a single classifier;

d. Use the male and female training samples to train the SVR to obtain a set of parameter values, and use the trained SVR to classify the test samples;

e. Use various fusion rules for combined classification, and compare the classification accuracy of different fusion rules.

And, the steps of the algorithm of the age group identification described in step (3) are as follows:

a. Extract the human face, skin color, hairstyle, and various features of the shape. For the face, select the left and right eyes, nose and mouth with strong discrimination ability as the local area, reduce the computational complexity to a certain extent, and calibrate the feature points Segment the left and right eyes, nose and mouth areas of the face;

b. Use the linear discriminant method to extract the characteristics of the human body detected by training, and use the Fisher method to extract the characteristics of the training samples and target images;

c, utilize the dynamic clustering method to classify the target image, and calculate the average recognition rate of a single classifier;

d. Use the training samples of different age groups to train the SVR to obtain a set of parameter values, and use the trained SVR to classify the test samples;

e. Use various fusion rules for combined classification, and compare the classification accuracy of different fusion rules. According to the classification of the target image, the classification category is iteratively increased, and the age classification interval is narrowed, so as to improve the age estimation accuracy.

Moreover, the advertisement recommendation placement decision described in step (4) adopts demographic recommendation technology, and the advertisement placement adopts manual classification.

Advantage and positive effect of the present invention are:

The present invention aims at the intelligent naked-eye 3D advertising machine, through the extraction and fusion of various characteristics of the target population in a complex environment, and finally realizes the intelligent classification and placement decision of the naked-eye 3D advertising machine advertisement and based on face recognition, somatosensory technology and voice recognition. Human-computer multi-sensory intelligent interaction of technology. Therefore, through the extraction, calculation and classification of multiple features, it provides an important data basis for the decision-making of the naked-eye 3D advertising machine, and fills the gap of intelligent recognition technology in the application field of the naked-eye 3D advertising machine.

The present invention captures human body motions or voice signals through somatosensory sensors and voice collectors, and converts human body motions or voice signals into corresponding instructions through information processing technology, which is used to control the playback of intelligent advertising machines and perform corresponding man-machine interact. In this way, face recognition, somatosensory technology and voice recognition technology are integrated to realize the application of multi-sensory biometric technology on intelligent advertising machines, which can serve different groups and attract customers' attention.

Description of drawings

Fig. 1 is a flow chart of the multi-sensory biometric identification interaction method of the present invention;

Figure 2 is a block diagram of human body detection based on static images;

Fig. 3 is a block diagram of human body detection based on video;

Fig. 4 is the processing system block diagram of voice signal;

Fig. 5 is the calculation block diagram of MFCC feature parameter;

Fig. 6 is a block diagram of intelligent identification of gender and age group.

Detailed ways

The present invention will be further described in detail below through the specific examples, the following examples are only descriptive, not restrictive, and cannot limit the protection scope of the present invention with this.

A multi-sensory biometric interaction method for naked-eye 3D advertisements, the steps of which are:

⑴Human detection in complex scenes: In the construction of a platform for automatic delivery of targeted advertising content, monitoring and detection of moving human bodies in fixed scenes through cameras is the basis for further automatic identification of gender, age group, etc., mainly as follows Two study sub-contents:

Human detection and tracking: detection and tracking of crowds appearing in the video, multiple targets can be detected and tracked at the same time,

Number of people: Statistics of the number of people in the video.

Static image human detection, as shown in Figure 1, is actually a standard pattern recognition problem. The two most critical aspects are: feature selection and classifier design. In feature selection, Haar-like features are used. , EOH feature, texture feature, color feature, etc. In the design of the classifier, the method of statistical learning classification is adopted, and the Adaboost learning algorithm that has been successfully applied in the field of face detection is considered. Satisfactory results have been obtained in the design, and it is used for human body detection and tracking. The present invention is based on video-based moving human body detection and tracking. The scene background of the camera generally changes little, which brings great benefits to moving human body detection and tracking. Convenience, while greatly reducing the complexity of detection and tracking.

The block diagram of human body detection based on video is shown in Figure 2. Before processing the video to detect moving human body, the motion information in the video image is used to mark its moving area and remove the area that does not meet the requirements. Then use the trained classifier to detect the selected regions to get the final moving human instance. Foreground detection is mainly to divide the video image into the moving foreground area of the human body and the background area of the motionless part of the camera. The adaptive background model establishes a background model for the static background. By comparing the current image frame with the background model, the area with a large change is determined to be the foreground area. The calculation speed of this method is very fast, and a complete and accurate description of the moving target area can be obtained, but it is sensitive to the lighting conditions, large-area motion and noise in the scene, so it is necessary to use a certain algorithm for the background model in practical applications. Dynamically updated to adapt to changes in the environment.

The present invention uses a dynamic update model to describe the probability distribution of background point colors with Gaussian distribution (normal distribution). The detection of the new moving human body is based on comparing the current frame image with the background model established by the Gaussian mixture model to obtain all moving human bodies in the foreground image of the current frame. Secondly, it is compared with the recorded sports human body data. In this way, many moving targets in the current frame are obtained, and the smaller moving targets (probably caused by noise) that overlap with the tracked human body are discarded, and the remaining parts are sorted by size, and then Add 10 largest human bodies to continue tracking, and use them to count the number of people in the video.

(2) Feature detection of the crowd: Extract some features of the crowd to distinguish the age and gender of different groups. These features can be used comprehensively to improve the recognition accuracy and reliability of the system. The commonly used features are as follows:

a. Facial features: the overall or partial features of the human face.

b. Hairstyle features: mainly refers to long or short hair, and bald hair.

c. Sound characteristics: speech, coughing, wheezing, footsteps, etc.

b. Morphological characteristics: the height of the human body, the characteristics of key joint points.

The extraction of facial features and morphological features can be divided into two methods: overall feature extraction and local feature extraction. The overall feature extraction algorithm roughly recognizes the face, and the local feature extraction supplements the details. When local features (such as bald head, beard, etc.) are used, these local features can be directly used to determine gender, and the extraction of overall features and local features complement each other to achieve better results in face recognition.

In addition, feature fusion can also be successfully applied to local feature extraction. By processing an image multiple times, such as using local binary mode, multiple feature images can be obtained, and then fusion and feature extraction are performed, and finally artificial The face matches.

The four-dimensional local binary mode feature extraction is a new method based on the improvement of the local binary mode. The local binary mode only considers the local structure relationship between the center pixel and the area pixel, and describes the local texture of the image well. feature. However, the three-dimensional local binary mode ignores the influence of the pixel value of the center point of the image itself on the local detail features. Taking the gradient information into account, a new fused local binary pattern feature is formed.

Among them, for the extraction of hairstyle features,

The main steps of hair style feature extraction are as follows:

Step 1: Obtain a frame of image from the video for detection. Images can be acquired regularly or a single image can be acquired regularly for processing when a person in the scene has been detected.

The second step: image preprocessing. The effect of this step directly affects the effect of subsequent hairstyle model extraction. Hairstyle extraction is somewhat different from face extraction. The rigidity of the face model is relatively strong, and the reflectivity of the face is lower than that of the hair, that is, the sensitivity of the face to light is lower than that of the hair. Therefore, the treatment of this step is very important.

The third step: hairstyle segmentation. This step is the difficulty of the whole process, mainly because the hair is relatively flexible, its shape is not fixed, and it is greatly affected by changes in light, whether it is light intensity or light direction.

Step 4: The hairstyle model is formed. After segmentation, the hairstyle model can be extracted from the segmentation map, and the extracted hairstyle model is used for gender or age judgment.

For the extraction of sound features,

Sound feature extraction plays a decisive role. There are several links that must be experienced before the recognition system, including digitization, pre-filtering, sampling, A/D conversion, and pre-processing of sound signals (including framing and windowing). The specific process is shown in the figure 4.

For the sound feature extraction method, the Mel frequency scale cepstrum parameter (MFCC) is used. This method takes the auditory characteristics of the human ear into consideration, that is, the frequency spectrum of the sound is transformed into the nonlinear spectrum region of the Mel frequency scale, and then Then it is converted to the spectral domain through a certain homomorphic processing process, and its calculation flow chart is shown in Figure 5.

Compared with other traditional sound feature extraction methods, MFCC parameters have good recognition performance and anti-noise ability, and this parameter has better Lubang performance than LPCC based on channel model, which is more in line with the human ear characteristics, and still have good recognition performance when the signal-to-noise ratio decreases.

As for the method of classifying speech features, we use SVM (Support Vector Machine) for classification and recognition, and transfer the trained MFCC parameters to the SVM classifier, and then project them into a high-dimensional space to make them linearly separable. The principle of linear division is used to judge the classification boundary, and finally achieve the purpose of speech recognition.

(3) Crowd intelligent identification and classification: use the extracted crowd characteristics to intelligently identify and classify the gender and age groups of the crowd, and use statistics or artificial intelligence methods to identify and classify the extracted crowd features by gender and age group.

Gender Recognition: Classify males and females.

Age Group Identification: Distinguish and classify old, middle-aged, and young people.

Using a classifier to classify the age of the detected human body is the main method of age change research in the field of face recognition. It can be said that such an age estimation method is effective. But because a single classifier has its own advantages and disadvantages, the recognition results of different classifiers may be very different in the same situation. Multi-classifier decision fusion is a trend in pattern development, and its purpose is to improve classification accuracy and measure the density of patterns.

• Fisher's linear discriminant method. As a single classifier, Fisher's linear discriminant method has good recognition performance on different databases. However, when using Fisher's linear discriminant method for dimensionality reduction, some useful information will still be lost, which may be important for subsequent steps, which is also the shortcoming of this method.

· Dynamic clustering algorithm. The dynamic clustering algorithm uses the C-means algorithm, which is a commonly used unsupervised learning method based on the nearest neighbor rule. The basic idea is very simple. First, determine the required number of groups c, select c representative points, use these representative points as the initial type, and then find the nearest representative point for each sample X in the sample set H, and return X to Go to the group where the nearest representative point is located. In this way, H is initially divided into c groups by the nearest neighbor rule in the first iteration. In the next iteration, on this basis, the mean vector of each group obtained in the previous iteration is used as a new representative point, and H is divided into c groups according to the neighbor rule again until the grouping is stable.

Support Vector Regression (SVR for short). The support vector regression machine algorithm is the generalization of the support vector machine method on the regression problem. By introducing insensitive loss function and kernel function, it can be well applied to nonlinear regression analysis, and has good predictive performance for small sample set problems.

On the basis of the extracted features mentioned above, we can use the classifier to identify the gender and age of people, as shown in Figure 6. The detailed algorithm is described below:

① The main steps of the gender recognition algorithm are as follows:

a. According to the method introduced above, various features such as human face, skin color, hairstyle, decoration, etc. are extracted. Facial features include overall and local features.

b. Use the linear discriminant method to extract the features of the human body detected by training, and use the Fisher method to extract the features of the training samples and target images.

c. Use the dynamic clustering method to classify the target image, and calculate the average recognition rate of a single classifier.

d. Use the male and female training samples to train the SVR to obtain a set of parameter values. Use the trained SVR to classify the test samples.

e. Use various fusion rules for combined classification, and compare the classification accuracy of different fusion rules.

② The main steps of the age group identification algorithm are as follows:

a. According to the method introduced above, various features such as human face, skin color, hairstyle, decoration, etc. are extracted. For human faces, the left and right eyes, nose and mouth with strong discrimination ability are selected as local regions to reduce the computational complexity to a certain extent. The calibration feature points are used to segment the left and right eye, nose and mouth areas of the face.

b. Use the linear discriminant method to extract the features of the human body detected by training, and use the Fisher method to extract the features of the training samples and target images.

c. Use the dynamic clustering method to classify the target image, and calculate the average recognition rate of a single classifier.

d. Using training samples of different age groups (old, middle, young, etc.) to train the SVR to obtain a set of parameter values. Use the trained SVR to classify the test samples.

e. Use various fusion rules for combined classification, and compare the classification accuracy of different fusion rules. According to the classification of the target image, the classification category is iteratively increased, and the age classification interval is narrowed, so as to improve the age estimation accuracy. ⑷ Advertisement recommendation placement decision: According to the results of crowd classification, different advertisements are classified and placed. Using personalized recommendation technology, at the right time, the right information is delivered to the people who want to deliver it, so that the advertisement can expect to achieve better results.

Demographic-based recommendation technology is adopted. The most commonly used demographic characteristic variables of this technology are age, gender and geographical location, and sometimes income variables are also used (such as customers in high-end hotels). Age and gender are identified by the system, and geographic location is determined by the location of the system terminal.

In the initial stage of the advertisement, it is manually classified and marked with suitable crowd attributes (age, gender, geographical location and income, etc.). If possible, the system adds observations: the time a user watches an ad. The time is used to obtain the liking degree of a certain group of people for each advertisement, so as to adjust the playing strategy of each advertisement.

⑸Intelligent human-computer interaction: Through somatosensory technology, voice recognition technology, etc., the captured human motion and voice signals are converted into control signals to carry out the human-computer intelligent interaction of the advertising machine. Through face recognition technology, it is possible to identify a person's identity, age, and gender, and then play different advertisements for different groups of people. For example, play animation advertisements for children and cosmetic advertisements for women to achieve human-computer interaction. To a certain extent Can attract people's attention.

Through somatosensory sensors and voice collectors, human body movements or voice signals are captured, and through information processing technology, human body movements or voice signals are converted into corresponding instructions, which are used to control the playback of intelligent advertising machines and perform corresponding human-computer interaction.

Among them, the motion signals of the human body can be captured through the somatosensory sensor, such as left, right, up, down, and palm closing and other signals, and the human body’s motion signals can be converted into control signals of the intelligent advertising machine through signal processing. To control advertising video playback, pause, fast forward, fast rewind, source replacement, screen zoom in, zoom out and other functions, this technology can attract children and young advertising groups.

In addition, through voice recognition technology, human voices such as start, pause, play, fast forward, rewind, change of film source, screen zoom in, zoom out and other voice signals are converted into corresponding control signals to control the playback of advertising videos . The technology will greatly facilitate vulnerable groups such as the disabled.

In this way, face recognition, somatosensory technology and voice recognition technology are integrated to realize the application of multi-sensory biometric technology on intelligent advertising machines, which can serve different groups and attract customers' attention.

Claims (9)

1., for many perception bio-identification exchange method of bore hole 3D advertisement, it is characterized in that: the steps include:

(1) the human detection in scene: put in the construction of ad content platform targetedly at automatic switching, by camera, inspection and monitoring is carried out to the movement human in fixed scene, movement human detects and comprises human detection and tracking, and namely crowd in video appears in detection and tracking; The number of people, namely in video, the number of crowd is added up;

(2) the feature detection of crowd: extract the face characteristic of crowd, hair style feature, sound characteristic and morphological feature, is used for distinguishing age and the sex of different crowd;

(3) crowd's Intelligent Recognition and classification: the feature utilizing the crowd extracted, carries out the Intelligent Recognition classification of sex and age bracket to crowd;

(4) advertisement is recommended to throw in decision-making: according to the result of listener clustering, different advertisements put on by classification, adopt personalized recommended technology, different information is passed to the crowd wanting to specify;

(5) intelligent human-machine interaction: by body sense technology, speech recognition technology changes the human action of catching and voice signal into control signal, carries out the human-computer intellectualization of people and advertisement machine.

2. the many perception bio-identification exchange method for bore hole 3D advertisement according to claim 1, it is characterized in that: step (1) described in human detection and follow the tracks of and adopt still image human detection, it is in feature selecting, adopt Haar-like feature, EOH feature, textural characteristics, color characteristic, in the design of sorter, employing be statistical learning classification method.

3. the many perception bio-identification exchange method for bore hole 3D advertisement according to claim 1, it is characterized in that: the number of the people that step is (1) described adopts and dynamically updates model, utilize Gaussian distribution to describe the probability distribution of background dot color, the detection of new movement human contrasts according to current frame image and the background model set up by mixed Gauss model, obtain all movement humans in the foreground image of present frame, then, compare with the movement human data recorded again, obtain the target with kinetic property in many present frames thus, overlapping target is had to abandon less moving target and tracked human body, and remaining part is sorted according to size, just newly add 10 maximum human bodies afterwards to continue to follow the trail of.

4. the many perception bio-identification exchange method for bore hole 3D advertisement according to claim 1, it is characterized in that: the face characteristic that step is (2) described and morphological feature extraction are divided into global feature to extract and local shape factor two kinds of modes, global feature extraction algorithm carries out rough identification to face, and local shape factor supplements from details.

5. the many perception bio-identification exchange method for bore hole 3D advertisement according to claim 1, is characterized in that: the (2) described hair style feature extraction key step of step is as follows:

A, from video, obtain a two field picture detect, time image acquisitions or detected in scene that timing acquisition list note image processes when having personage.

B, Image semantic classification.

C, hair style are split.

D, hair style model are formed.

6. the many perception bio-identification exchange method for bore hole 3D advertisement according to claim 1, is characterized in that: crowd's Intelligent Recognition that step is (3) described and classification adopt multi-categorizer to realize Decision fusion.

7. the many perception bio-identification exchange method for bore hole 3D advertisement according to claim 6, is characterized in that: the step of the algorithm of the sex identification that step is (3) described is as follows:

The various features such as a, the face extracting human body, the colour of skin, hair style, decoration, face characteristic comprises whole and part feature;

B, utilize linear discriminant method to extract the feature of human body of training and detecting, utilize Fisher method to extract the feature of training sample and target image simultaneously;

C, utilize dynamic cluster method to classify to target image, and calculate the average recognition rate of single classifier;

D, utilize men and women's training sample to train SVR, obtain one group of parameter value, utilize the SVR trained to classify to test sample book;

E, various fusion rule is utilized to carry out assembled classification, the nicety of grading of more different fusion rule.

8. the many perception bio-identification exchange method for bore hole 3D advertisement according to claim 6, is characterized in that: the step of the algorithm of the age bracket identification that step is (3) described is as follows:

A, the face extracting human body, the colour of skin, hair style, the various feature of form, to face, choose distinguishing ability stronger left and right eyes, nose and face as regional area, reduce computation complexity to a certain extent, feature point for calibration is partitioned into the left and right eyes of face, nose and face region;

B, utilize linear discriminant method to extract the feature of human body of training and detecting, utilize Fisher method to extract the feature of training sample and target image simultaneously;

C, utilize dynamic cluster method to classify to target image, and calculate the average recognition rate of single classifier;

D, utilize Different age group training sample to train SVR, obtain one group of parameter value, utilize the SVR trained to classify to test sample book;

E, various fusion rule is utilized to carry out assembled classification, the nicety of grading of more different fusion rule.Increase class categories according to the classification iteration of target image, reduce age bracket class interval, to improve age estimated accuracy.

9. the many perception bio-identification exchange method for bore hole 3D advertisement according to claim 1, is characterized in that: the (4) described advertisement of step is recommended to throw in decision-making and adopted demographic recommended technology, and advertisement putting adopts manual type classification.