WO2018036241A1 - 一种年龄群体分类的方法及装置 - Google Patents

一种年龄群体分类的方法及装置 Download PDF

Info

Publication number
WO2018036241A1
WO2018036241A1 PCT/CN2017/087184 CN2017087184W WO2018036241A1 WO 2018036241 A1 WO2018036241 A1 WO 2018036241A1 CN 2017087184 W CN2017087184 W CN 2017087184W WO 2018036241 A1 WO2018036241 A1 WO 2018036241A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
classification
classifier
image
dimensional image
Prior art date
Application number
PCT/CN2017/087184
Other languages
English (en)
French (fr)
Inventor
陈彬
Original Assignee
中兴通讯股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Publication of WO2018036241A1 publication Critical patent/WO2018036241A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/245—Classification techniques relating to the decision surface
    • G06F18/2451—Classification techniques relating to the decision surface linear, e.g. hyperplane
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20004—Adaptive image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20224—Image subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30232—Surveillance

Definitions

  • the present application relates to the field of communication technologies, for example, to a method and apparatus for age group classification.
  • the classification of age groups is a very critical technology. From the latest research based on the 2015 China Mobile Internet Crowd Behavior Analysis Report, it can be seen that the era of the mobile Internet has arrived, and users have long been unable to meet the one-way guidance and indoctrination. Experience, engagement, and communication with each other are becoming more and more important to the user community. Being able to grasp the behavioral characteristics and trends of the Internet population, deepen insights, discover the rules, and use them in digital marketing is the premise to seize the user market, understand the needs of the user groups, fit different user groups, and create value for users. It is especially important. However, the age group classification method in the related art has low accuracy, which affects the analysis of the actual use of user data.
  • the embodiments of the present disclosure provide a method and apparatus for age group classification to solve the problem of low accuracy of the age group classification method in the related art.
  • a method of age group classification comprising:
  • the age group is classified into the characters in the video scene according to the acquired vital information.
  • the method may further include:
  • the cameras at different shooting angles with respect to the video scene are calibrated to obtain internal parameters and external parameters of each camera.
  • the step of establishing a three-dimensional image corresponding to the video scene according to the image information may include:
  • stereo feature matching is performed according to the feature points
  • the step of acquiring the vital sign information of the person in the three-dimensional image may include:
  • the age group is classified into the characters in the video scene according to the obtained height information and shoulder width information.
  • the step of acquiring the height information and the shoulder width information of the person in the three-dimensional image may include:
  • the length of the rectangular frame is determined as the height information of the person, and the width of the rectangular frame is determined as the shoulder width information of the person.
  • the age group to which each vital sign information belongs is determined.
  • the method may further include:
  • the multi-level classifier is constructed.
  • the step of constructing the multi-level classifier may include:
  • sample information from the three-dimensional image wherein the sample information includes: physical sign information of a character in the three-dimensional image and form information of a part of the object;
  • the sample information assigned the weighted value is sequentially input into a plurality of basic classifiers for classification training;
  • the plurality of basic classifiers that complete the classification training are constructed into a multi-level strong classifier.
  • the step of sequentially inputting the sample information of the assigned weight value into the plurality of basic classifiers for classifying training may include:
  • N c ⁇ (c-1)/2;
  • N represents the number of basic classifiers and c represents the number of age group classifications
  • the sample information assigned the weighted value is sequentially input into the N basic classifiers for classification training, wherein each basic classifier can classify two age groups.
  • an apparatus for classifying age groups comprising:
  • An image acquisition module configured to acquire image information of a video scene captured by at least one camera within a predetermined time period
  • An image creation module is configured to establish a three-dimensional image corresponding to the video scene according to the image information acquired by the image acquisition module;
  • a sign obtaining module configured to acquire the vital sign information of the person in the three-dimensional image established by the image establishing module
  • the classification module is configured to perform age group classification on the characters in the video scene according to the physical information acquired by the physical sign acquisition module.
  • the device may further include:
  • the parameter acquisition module is configured to calibrate the cameras at different shooting angles with respect to the video scene, and acquire internal parameters and external parameters of each camera.
  • the image creation module may include:
  • An extracting unit configured to extract feature points in the image information
  • a matching unit configured to perform stereo feature matching according to the feature points in image information captured at different angles
  • the image establishing unit is configured to establish a three-dimensional image corresponding to the video scene according to the stereo feature matching result and the internal parameter and the external parameter of the camera corresponding to each image information.
  • the sign acquisition module may include:
  • a sign obtaining unit configured to acquire height information and shoulder width information of a person in the three-dimensional image
  • the classification module includes:
  • the classification unit is configured to classify the characters in the video scene according to the acquired height information and shoulder width information.
  • the sign obtaining unit may include:
  • a first processing sub-unit configured to frame a person in the three-dimensional image with a rectangular frame according to the determined contour information
  • a first computing subunit configured to calculate a length and a width of the rectangular frame
  • the second processing subunit is configured to determine the length of the rectangular frame as the height information of the character, and determine the width of the rectangular frame as the shoulder width information of the character.
  • the classification module may include:
  • a first processing unit configured to input the acquired vital sign information into a preset multi-level classifier
  • the decision unit is configured to determine an age group to which each of the vital sign information belongs according to an output result of the multi-level classifier.
  • the device may further include:
  • a building block is configured to construct the multi-level classifier.
  • the building module can include:
  • a sample acquiring unit configured to acquire sample information from the three-dimensional image, wherein the sample information includes: physical sign information of a person in the three-dimensional image and form information of a partial object;
  • a weight allocation unit configured to allocate the same weight value for each sample information
  • a second processing unit configured to sequentially input the sample information of the assigned weight value into the plurality of basic classifiers for classification training
  • a third processing unit configured to re-adjust the weight value of each sample information according to the classification result output by each basic classifier, and input the classification training in the next basic classifier; a classification result output by the basic classifier, calculating a classification error value of each basic classifier, and calculating a weight value of each basic classifier according to the size of the classification error value;
  • the building unit is configured to construct a plurality of basic classifiers that complete the classification training into a multi-level strong classifier.
  • the second processing unit may include:
  • N represents the number of basic classifiers and c represents the number of age group classifications
  • the third processing sub-unit is configured to sequentially input the sample information of the assigned weight value into the N basic classifiers for classification training, wherein each basic classifier can classify the two age groups.
  • Embodiments of the present disclosure also provide a non-transitory computer readable storage medium storing computer executable instructions arranged to perform the above method.
  • An embodiment of the present disclosure further provides an electronic device, including:
  • At least one processor At least one processor
  • the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the method described above.
  • the above technical solution uses the two-dimensional image captured by the camera to perform three-dimensional reconstruction, acquires the physical sign information of the character in the established three-dimensional image, and classifies the characters in the video scene according to the acquired physical information, and the reconstruction technique is better.
  • the restored original image is more accurate and can adapt to the change of the lens angle, so it can obtain more accurate physical information in the establishment of the three-dimensional image, thereby improving the accuracy of age group classification.
  • FIG. 1 is a flow chart showing a method of age group classification provided by a first embodiment of the present disclosure
  • FIG. 2 is a flowchart showing a method for establishing a three-dimensional image according to a first embodiment of the present disclosure
  • FIG. 3 is a flow chart showing a method for acquiring height and shoulder width information provided by the first embodiment of the present disclosure
  • FIG. 4 is a flowchart showing a method of constructing a multi-level classifier according to a first embodiment of the present disclosure
  • FIG. 5 is a structural diagram of a multi-level classifier construction process according to a first embodiment of the present disclosure
  • FIG. 6 is a flow chart showing an apparatus for age group classification provided by a second embodiment of the present disclosure.
  • FIG. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure.
  • the embodiment of the present disclosure provides a method for classifying age groups, as shown in FIG. 1 , including:
  • One or more cameras may be arranged according to actual needs in a video scene to collect image information of different angles of the video scene.
  • the camera can be a monocular camera or a multi-camera camera.
  • the camera lens can be in a rotatable form or a non-rotatable form, which can be selected according to actual needs.
  • a dynamic three-dimensional image can be established according to the two-dimensional image information captured by the camera in real time, and the timeliness is good, and the information in the video scene can be accurately learned from the three-dimensional image.
  • the selection and quantity of the types of the vital information described herein may be determined according to the needs of the age group classification, and the physical information that can distinguish the age of the person is within the protection scope of the embodiments of the present disclosure.
  • S104 Perform age group classification on the characters in the video scene according to the obtained vital sign information.
  • the age groups are classified according to the vital information.
  • the two-dimensional image captured by the camera is used for three-dimensional reconstruction
  • the physical location information of the character is acquired in the established three-dimensional image
  • the age group of the characters in the video scene is classified according to the acquired physical information
  • the reconstruction technology can Better restoration of the original image, more accurate, and can adapt to changes in the angle of the lens, so it can obtain more accurate physical information in the establishment of three-dimensional images, thereby improving the accuracy of age group classification.
  • image information of a video scene may be acquired by using multiple cameras, and multiple shots are respectively arranged in different orientations in the video scene, thereby improving the accuracy and shooting efficiency of the image information.
  • the internal parameters include: focal length, imaging origin and distortion coefficient
  • external parameters include: optical characteristic parameter matrix and translation matrix.
  • the process of the camera calibration is as follows: first, prepare a sample of the checkerboard, and mark the feature points on the checkerboard.
  • a red solid dot with a diameter of 1 cm may be used.
  • Multiple shots sequentially acquire several sample images at different angles, and extract the contour of each feature point in the sample image by the Hough circle detection algorithm, fit into a circle, and locate the center of each circle to determine The center of the circle corresponds to the feature points marked on the checkerboard sample in the three-dimensional world, thereby calculating the internal parameters and external parameters of the camera.
  • the calibration operation needs to be performed at the calibration time to reduce the influence of the distortion parameters of the camera lens on the quality of the captured image, reduce the distortion of the captured image, and improve the authenticity of the displayed image.
  • the process of establishing a three-dimensional image corresponding to a video scene according to image information captured by the camera may include:
  • the selection of feature points can be determined according to actual needs.
  • the shift feature matching algorithm can be used for stereo feature matching.
  • the vital sign information to be acquired in the embodiment of the present disclosure may be the height information and the shoulder width information of the character.
  • the height shoulder width of international children and adults is used as reference data, so it is carried out by height and shoulder width information.
  • the age group classification has a good basis, and the program has good practicability and scalability.
  • the method for obtaining the height information and the shoulder width information of the character in the three-dimensional image includes:
  • target detection and feature extraction are performed on the three-dimensional image to extract the contour information of the human body.
  • the background difference method can be used to adaptive background learning of multi-frame image information, and the reference background is obtained. Then, according to the reference background, the image segmentation algorithm is used to segment the subsequent image information, and the difference target feature is extracted.
  • the feature is Gaubor filtering, and the image information of the difference target feature in the filtering range is excluded, and the image matching information is tracked by the shift tracking algorithm. When the image information of the difference target feature continues to exceed the preset frame When it is counted, mark it as a validly determined target (ie, a person).
  • the determined contour of the human body is framed by a rectangular frame, and the target person is tracked, and the shape of the rectangular frame is adjusted in real time according to the change of the shape of the character. If the character framed by the rectangle disappears in the image, the acquisition of the height and shoulder width information of the character is abandoned.
  • S304 Determine the length of the rectangular frame as the height of the character, and determine the width of the rectangular frame as the shoulder width of the character.
  • the physical sign information can be input into a preset multi-level classifier, and the age group to which each vital sign information belongs is determined according to the output result of the multi-level classifier.
  • the preset multi-level classifier described herein is a pre-trained classifier, which is a multi-level strong classification composed of a plurality of weak classification cascades, which can classify the vital information more accurately.
  • the steps of constructing a multi-level classifier include:
  • the sample information includes: the physical sign information (ie, the positive sample) of the character in the three-dimensional image and the morphological information of the partial object (ie, the negative sample), and the accuracy of the classifier training can be improved by adding the negative sample information.
  • each sample information At the beginning of constructing the multi-level classifier, first assign the same weight value to each sample information. If there are N sample information, the weight value of each sample information is 1/N, that is, each sample has the same utterance. right.
  • the sample information that is assigned the weighted value is sequentially input into a plurality of basic classifiers for classification training.
  • the sample information is sequentially input into a plurality of basic classifiers to perform classification training to construct a multi-level classifier.
  • S404 Re-adjust the weight value of each sample information according to the classification result output by each basic classifier, and input the classification training in the next basic classifier; and calculate each according to the classification result output by each basic classifier.
  • the classification error value of the base classifier, and the weight value of each base classifier is calculated according to the size of the classification error value.
  • the embodiment of the present disclosure may perform the training of the classifier by using the Adaboost algorithm, wherein each of the basic classifiers generates a classification result after the classification information is trained by the sample information, and reduces the weight value for the accurate classification of the sample information.
  • the wrong sample information increase its weight value, and then input the sample information of the redistributed weight value into the next basic classifier for training, as shown in Figure 5.
  • the classification error value of the basic classifier that completes the classification training is calculated, and the weight value of the basic classifier is calculated according to the classification error value. The smaller the general classification error value is, the larger the calculated weight value is.
  • each base classifier is assigned a weight value, and all the base classifiers are cascaded to form a multi-level strong classifier.
  • the age group in the embodiment of the present disclosure may include: a child group of 0 to 7 years old, a youth group of 8 to 17 years old, a youth group of 18 to 30 years old, and an age group of 31 to 45 years old.
  • the process of implementing age group classification is to identify which age group the person in the three-dimensional image belongs to.
  • N represents the number of basic classifiers and c represents the number of age group classifications.
  • c represents the number of age group classifications.
  • the present disclosure There are a total of 6 age groupings in the embodiment, and the number of basic classifiers required is 15.
  • the sample information assigned the weighted value is sequentially input into the N basic classifiers for classification training, wherein each basic classifier can classify two age groups.
  • the two age groups classified by each base classifier are different from the two age groups classified by other base classifiers.
  • the technical solution provided by the embodiment of the present disclosure performs three-dimensional reconstruction by using a two-dimensional image captured by a calibrated camera, acquires physical sign information of the character in the established three-dimensional image, and pairs the acquired vital information into the video scene.
  • the characters are classified by age group. Because the reconstruction technology can restore the original image better, it is more accurate, and can adapt to the change of the lens angle. Therefore, it is possible to obtain more accurate physical information in the establishment of the three-dimensional image, thereby improving the age group classification. The accuracy.
  • An embodiment of the present disclosure provides an apparatus for classifying age groups. As shown in FIG. 6, the apparatus includes:
  • the image obtaining module 601 is configured to acquire image information of a video scene captured by at least one camera within a predetermined time period.
  • One or more cameras may be arranged according to actual needs in a video scene to collect image information of different angles of the video scene.
  • the camera can be a monocular camera or a multi-camera camera.
  • the camera lens can be in a rotatable form or a non-rotatable form, which can be selected according to actual needs.
  • the image establishing module 602 is configured to establish a three-dimensional image corresponding to the video scene according to the image information acquired by the image acquiring module 601.
  • a dynamic three-dimensional image can be established according to the two-dimensional image information captured by the camera in real time, and the timeliness is good, and the information in the video scene can be accurately learned from the three-dimensional image.
  • the sign acquisition module 603 is configured to acquire the vital sign information of the person in the three-dimensional image established by the image establishing module 602.
  • the selection and quantity of the types of the vital information described herein may be determined according to the needs of the age group classification, and the physical information that can distinguish the age of the person is within the protection scope of the embodiments of the present disclosure.
  • the classification module 604 is configured to perform age group classification on the characters in the video scene according to the vital information acquired by the sign acquisition module 603.
  • the age groups are classified according to the vital information.
  • the device can also include:
  • the parameter acquisition module is configured to calibrate cameras located at different shooting angles with respect to the video scene, and obtain internal parameters and external parameters of each camera.
  • image information of a video scene is acquired by using multiple cameras, and the plurality of lenses are respectively arranged in different orientations in the video scene, thereby improving accuracy and shooting efficiency of the image information.
  • the internal parameters include: focal length, imaging origin and distortion coefficient, and external parameters include: optical characteristic parameter matrix and translation matrix.
  • the process of the camera calibration is as follows: first, prepare a sample of the checkerboard, and mark the feature points on the checkerboard.
  • a red solid dot with a diameter of 1 cm may be used.
  • Multiple shots sequentially acquire several sample images at different angles, and extract the contour of each feature point in the sample image by the Hough circle detection algorithm, fit into a circle, and locate the center of each circle to determine The center of the circle corresponds to the feature points marked on the checkerboard sample in the three-dimensional world, thereby calculating the internal parameters and external parameters of the camera.
  • the calibration operation needs to be performed at the calibration time to reduce the influence of the distortion parameters of the camera lens on the quality of the captured image, reduce the distortion of the captured image, and improve the authenticity of the displayed image.
  • the image building module can include:
  • An extracting unit configured to extract feature points in the image information.
  • the selection of feature points can be determined according to actual needs.
  • the matching unit is configured to perform stereo feature matching according to the feature points in the image information captured at different angles.
  • the shift feature matching algorithm can be used for stereo feature matching.
  • the image establishing unit is configured to establish a three-dimensional image corresponding to the video scene according to the stereo feature matching result and the internal parameters and the external parameters of the camera corresponding to each image information.
  • the sign acquisition module can include:
  • the sign acquisition unit is configured to acquire height information and shoulder width information of the person in the three-dimensional image.
  • the classification module includes:
  • the classification unit is configured to classify the age groups of the characters in the video scene according to the acquired height information and shoulder width information.
  • the vital sign information to be acquired in the embodiment of the present disclosure is the height information and the shoulder width information of the person.
  • the height and shoulder width of international children and adults as reference data, there is a good basis for age group classification through height and shoulder width information, and the program has good practicability and scalability.
  • the sign obtaining unit may include:
  • a determination subunit configured to determine contour information of a person in the three-dimensional image.
  • target detection and feature extraction are performed on the three-dimensional image to extract the contour information of the human body.
  • the background difference method can be used to adaptive background learning of multi-frame image information, and the reference background is obtained. Then, according to the reference background, the image segmentation algorithm is used to segment the subsequent image information, and the difference target feature is extracted.
  • the feature is Gaubor filtering, and the image information of the difference target feature in the filtering range is excluded, and the image matching information is tracked by the shift tracking algorithm. When the image information of the difference target feature continues to exceed the preset frame When it is counted, mark it as a validly determined target (ie, a person).
  • the first processing sub-unit is configured to frame the characters in the three-dimensional image with a rectangular frame according to the determined contour information.
  • the determined contour of the human body is framed by a rectangular frame, and the target person is tracked, and the shape of the rectangular frame is adjusted in real time according to the change of the shape of the character. If the character framed by the rectangle disappears in the image, the acquisition of the height and shoulder width information of the character is abandoned.
  • the first calculation subunit is configured to calculate the length and width of the rectangular frame.
  • the second processing subunit is configured to determine the length of the rectangular frame as the height of the character and the width of the rectangular frame as the shoulder width of the character.
  • the classification module can include:
  • the first processing unit is configured to input the acquired vital sign information into a preset multi-level classifier.
  • the decision unit is configured to determine an age group to which each vital sign information belongs according to an output result of the multi-level classifier.
  • the physical sign information can be input into a preset multi-level classifier, and the age group to which each vital sign information belongs is determined according to the output result of the multi-level classifier.
  • the preset multi-level classifier described herein is a pre-trained classifier, which is a multi-level strong classification composed of a plurality of weak classification cascades, which can classify the vital information more accurately.
  • the device may also include:
  • a build module that is configured to build a multi-level classifier.
  • the building block can include:
  • a sample acquisition unit configured to acquire sample information from the three-dimensional image.
  • the sample information includes: the physical sign information (ie, the positive sample) of the character in the three-dimensional image and the morphological information of the partial object (ie, the negative sample), and the accuracy of the classifier training can be improved by adding the negative sample information.
  • a weight allocation unit configured to assign the same weight value for each sample information.
  • each sample information At the beginning of constructing the multi-level classifier, first assign the same weight value to each sample information. If there are N sample information, the weight value of each sample information is 1/N, that is, each sample has the same utterance. right.
  • the second processing unit is configured to sequentially input the sample information of the assigned weight value into the plurality of basic classifiers for classification training.
  • the sample information is sequentially input into a plurality of basic classifiers to perform classification training to construct a multi-level classifier.
  • a third processing unit configured to: re-adjust the weight value of each sample information according to the classification result output by each basic classifier, and input the classification training in the next basic classifier; and output according to each basic classifier
  • the classification result is calculated, the classification error value of each basic classifier is calculated, and the weight value of each basic classifier is calculated according to the size of the classification error value.
  • the embodiment of the present disclosure may perform the training of the classifier by using the Adaboost algorithm, wherein each of the basic classifiers generates a classification result after the classification information is trained by the sample information, and reduces the weight value for the accurate classification of the sample information.
  • the wrong sample information increase its weight value, and then input the sample information of the redistributed weight value into the next basic classifier for training, as shown in Figure 5.
  • the classification error value of the basic classifier that completes the classification training is calculated, and the weight value of the basic classifier is calculated according to the classification error value. The smaller the general classification error value is, the larger the calculated weight value is.
  • the building unit is configured to construct a plurality of basic classifiers that complete the classification training into a multi-level strong classifier.
  • each base classifier is assigned a weight value, and all the base classifiers are cascaded to form a multi-level strong classifier.
  • the second processing unit can include:
  • N represents the number of basic classifiers and c represents the number of age group classifications.
  • N represents the number of basic classifiers and c represents the number of age group classifications.
  • c represents the number of age group classifications.
  • the third processing sub-unit is configured to sequentially input the sample information of the assigned weight value into the N basic classifiers for classification training.
  • Each of the basic classifiers can classify two age groups.
  • the two age groups classified by each base classifier are different from the two age groups classified by other base classifiers.
  • the age group in the embodiment of the present disclosure includes: a child group of 0 to 7 years old, a youth group of 8 to 17 years old, a youth group of 18 to 30 years old, and an age group of 31 to 45 years old.
  • the process of implementing age group classification is to identify which age group the person in the three-dimensional image belongs to.
  • Embodiments of the present disclosure also provide a non-transitory computer readable storage medium storing computer executable instructions arranged to perform the method of any of the above embodiments.
  • the embodiment of the present disclosure further provides a schematic structural diagram of an electronic device.
  • the electronic device includes:
  • At least one processor 70 which is exemplified by a processor 70 in FIG. 7; and a memory 71, may further include a communication interface 72 and a bus 73.
  • the processor 70, the communication interface 72, and the memory 71 can complete communication with each other through the bus 73.
  • Communication interface 72 can be used for information transfer.
  • Processor 70 can invoke logic instructions in memory 71 to perform the methods of the above-described embodiments.
  • logic instructions in the memory 71 described above can be implemented in the form of software functional units and When sold or used as a stand-alone product, it can be stored on a computer readable storage medium.
  • the memory 71 is a computer readable storage medium, and can be used to store a software program, a computer executable program, a program instruction/module corresponding to the method in the embodiment of the present disclosure.
  • the processor 70 executes the function application and the data processing by executing the software programs, instructions, and modules stored in the memory 71, that is, the method of realizing the age group classification in the above method embodiments.
  • the memory 71 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application required for at least one function; the storage data area may store data created according to usage of the terminal device, and the like. Further, the memory 71 may include a high speed random access memory, and may also include a nonvolatile memory.
  • the technical solution of the embodiments of the present disclosure may be embodied in the form of a software product stored in a storage medium, including one or more instructions for causing a computer device (which may be a personal computer, a server, or a network) The device or the like) performs all or part of the steps of the method described in the embodiments of the present disclosure.
  • the foregoing storage medium may be a non-transitory storage medium, including: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and the like.
  • the technical solution provided by the embodiment of the present disclosure performs three-dimensional reconstruction by using a two-dimensional image captured by a calibrated camera, acquires physical sign information of the character in the established three-dimensional image, and pairs the acquired vital information into the video scene.
  • the characters are classified by age group. Because the reconstruction technology can restore the original image better, it is more accurate, and can adapt to the change of the lens angle. Therefore, it is possible to obtain more accurate physical information in the establishment of the three-dimensional image, thereby improving the age group classification. The accuracy.
  • the device for classifying the age group in the embodiment of the present disclosure is a device corresponding to the method for classifying the age group in the first embodiment, and all the implementation manners in the foregoing method embodiments are applicable to the embodiment of the device. In the same way, the same technical effect can be achieved.
  • the method and apparatus for classifying age groups provided by the present application can improve the accuracy of age group classification.

Abstract

一种年龄群体分类的方法及装置，其中，该年龄群体分类的方法包括：获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息（S101），根据图像信息，建立与视频场景对应的三维图像（S102），获取三维图像中的人物的体征信息（S103），根据获取的体征信息，对视频场景中的人物进行年龄群体分类（S104）。所述方法能够提高年龄群体分类的准确性。

Description

一种年龄群体分类的方法及装置 技术领域
本申请涉及通信技术领域，例如涉及一种年龄群体分类的方法及装置。
背景技术
在全民移动的时代，年龄群体的分类研究是一个非常关键的技术。从最新的《基于2015年中国移动互联网人群行为分析报告》的研究可以看出，移动互联网的群体时代已到来，用户早已不满足单向的引导和灌输。体验感、参与度以及互相之间的交流，对用户群体来说已经越来越重要。能够把握互联网人群的行为特征及趋势，深入洞察，发现规律，并将其用在数字营销中，是抢占用户市场的前提，能够了解用户群体需求，贴合不同的用户群体，能够为用户创造价值就显得尤为重要。但相关技术中的年龄群体分类方法准确度较低，影响对实际使用用户数据的分析。
发明内容
本公开实施例提供了一种年龄群体分类的方法及装置，以解决相关技术中的年龄群体分类方法准确度低的问题。
为了解决上述技术问题，本公开采用如下技术方案：
第一方面，提供了一种年龄群体分类的方法，包括：
获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息；
根据所述图像信息，建立与所述视频场景对应的三维图像；
获取所述三维图像中的人物的体征信息；
根据获取的体征信息，对所述视频场景中的人物进行年龄群体分类。
所述获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息的步骤之前，所述方法还可以包括：
对相对于所述视频场景位于不同拍摄角度的所述摄像机进行标定，获取每一摄像机的内部参数和外部参数。
所述根据所述图像信息，建立与所述视频场景对应的三维图像的步骤可以包括：
提取所述图像信息中的特征点；
在不同角度拍摄的图像信息中，根据所述特征点进行立体特征匹配；
根据立体特征匹配结果和每一图像信息对应的摄像机的内部参数和外部参数，建立与所述视频场景对应的三维图像。
所述获取所述三维图像中的人物的体征信息的步骤可以包括：
获取所述三维图像中的人物的身高信息和肩宽信息；
其中，所述根据获取的体征信息，对所述视频场景中的人物进行年龄群体分类的步骤包括：
根据获取的身高信息和肩宽信息，对所述视频场景中的人物进行年龄群体分类。
所述获取所述三维图像中的人物的身高信息和肩宽信息的步骤可以包括：
确定所述三维图像中的人物的轮廓信息；
根据确定的轮廓信息，用矩形框框出所述三维图像中的人物；
计算所述矩形框的长度和宽度；
将所述矩形框的长度确定为人物的身高信息，将所述矩形框的宽度确定为人物的肩宽信息。
所述根据获取的体征信息，对所述视频场景中人物的年龄群体进行分类的步骤可以包括：
将获取的体征信息输入预设的多级分类器；
根据所述多级分类器的输出结果，确定每一体征信息所属的年龄群体。
在所述将获取的体征信息输入预设的多级分类器的步骤之前，所述方法还可以包括：
构建所述多级分类器。
所述构建所述多级分类器的步骤可以包括：
从所述三维图像中获取样本信息，其中样本信息包括：所述三维图像中的人物的体征信息和部分物体的形态信息；
分配每一样本信息相同的权重值；
将分配有权重值的样本信息依次输入多个基础分类器中进行分类训练；
根据每一基础分类器输出的分类结果，重新调整每一样本信息的权重值，并输入下一基础分类器中进行分类训练；以及根据每一基础分类器输出的分类结果，计算每一基础分类器的分类误差值，并根据分类误差值的大小，计算每一基础分类器权重值；
将完成分类训练的多个基础分类器构建成多级强分类器。
所述将分配有权重值的样本信息依次输入多个基础分类器中进行分类训练的步骤可以包括：
根据预设公式：N＝c×(c-1)/2，计算所述多级分类器中的基础分类器的数量；
其中，N表示基础分类器的数量，c表示年龄群体分类的数量；
将分配有权重值的样本信息依次输入N个基础分类器中进行分类训练，其中每一基础分类器能够分类出两种年龄群体。
第二方面，提供了一种年龄群体分类的装置，包括：
图像获取模块，被配置为获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息；
图像建立模块，被配置为根据所述图像获取模块所获取的所述图像信息，建立与所述视频场景对应的三维图像；
体征获取模块，被配置为获取所述图像建立模块所建立的所述三维图像中的人物的体征信息；
分类模块，被配置为根据所述体征获取模块所获取的体征信息，对所述视频场景中的人物进行年龄群体分类。
所述装置还可以包括：
参数获取模块，被配置为对相对于所述视频场景位于不同拍摄角度的所述摄像机进行标定，获取每一摄像机的内部参数和外部参数。
所述图像建立模块可以包括：
提取单元，被配置为提取所述图像信息中的特征点；
匹配单元，被配置为在不同角度拍摄的图像信息中，根据所述特征点进行立体特征匹配；
图像建立单元，被配置为根据立体特征匹配结果和每一图像信息对应的摄像机的内部参数和外部参数，建立与所述视频场景对应的三维图像。
所述体征获取模块可以包括：
体征获取单元，被配置为获取所述三维图像中的人物的身高信息和肩宽信息；
其中，所述分类模块包括：
分类单元，被配置为根据获取的身高信息和肩宽信息，对所述视频场景中的人物进行年龄群体分类。
所述体征获取单元可以包括：
确定子单元，被配置为确定所述三维图像中的人物的轮廓信息；
第一处理子单元，被配置为根据确定的轮廓信息，用矩形框框出所述三维图像中的人物；
第一计算子单元，被配置为计算所述矩形框的长度和宽度；
第二处理子单元，被配置为将所述矩形框的长度确定为人物的身高信息，将所述矩形框的宽度确定为人物的肩宽信息。
所述分类模块可以包括：
第一处理单元，被配置为将获取的体征信息输入预设的多级分类器；
决策单元，被配置为根据所述多级分类器的输出结果，确定每一体征信息所属的年龄群体。
所述装置还可以包括：
构建模块，被配置为构建所述多级分类器。
所述构建模块可以包括：
样本获取单元，被配置为从所述三维图像中获取样本信息，其中样本信息包括：所述三维图像中的人物的体征信息和部分物体的形态信息；
权重分配单元，被配置为分配每一样本信息相同的权重值；
第二处理单元，被配置为将分配有权重值的样本信息依次输入多个基础分类器中进行分类训练；
第三处理单元，被配置为根据每一基础分类器输出的分类结果，重新调整每一样本信息的权重值，并输入下一基础分类器中进行分类训练；以及根据每 一基础分类器输出的分类结果，计算每一基础分类器的分类误差值，并根据分类误差值的大小，计算每一基础分类器权重值；
构建单元，被配置为将完成分类训练的多个基础分类器构建成多级强分类器。
所述第二处理单元可以包括：
第二计算子单元，被配置为根据预设公式：N＝c×(c-1)/2，计算所述多级分类器中的基础分类器的数量；
其中，N表示基础分类器的数量，c表示年龄群体分类的数量；
第三处理子单元，被配置为将分配有权重值的样本信息依次输入N个基础分类器中进行分类训练，其中每一基础分类器能够分类出两种年龄群体。
本公开实施例还提供了一种非暂态计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令设置为执行上述方法。
本公开实施例还提供了一种电子设备，包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器执行上述的方法。
本公开的有益效果是：
上述技术方案，利用摄像机拍摄的二维图像进行三维重建，在建立的三维图像中获取人物的体征信息，并根据获取的体征信息对视频场景中的人物进行年龄群体分类，由于重建技术能较好的复原原始图像，较为精确，并且能够适应镜头角度的变化，因此能够在建立三维图像中，获取到更加精确的体征信息，从而提高年龄群体分类的准确性。
附图概述
图1表示本公开第一实施例提供的年龄群体分类的方法的流程图；
图2表示本公开第一实施例提供的三维图像建立方法的流程图；
图3表示本公开第一实施例提供的身高和肩宽信息获取方法的流程图；
图4表示本公开第一实施例提供的构建多级分类器的方法的流程图；
图5表示本公开第一实施例提供的多级分类器构建过程结构图；
图6表示本公开第二实施例提供的年龄群体分类的装置的流程图；以及
图7表示本公开实施例提供的电子设备的结构示意图。
具体实施方式
下面将参照附图更详细地描述本公开的示例性实施例。虽然附图中显示了本公开的示例性实施例，然而应当理解，可以以各种形式实现本公开而不应被这里阐述的实施例所限制。相反，提供这些实施例是为了能够更透彻地理解本公开，并且能够将本公开的范围完整的传达给本领域的技术人员。
第一实施例
本公开实施例提供了一种年龄群体分类的方法，如图1所示，包括：
S101、获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息。
其中，在一视频场景中可根据实际需求布置一个或多个摄像机，以采集视频场景不同角度的图像信息。摄像机可以是单目摄像机也可以是多目摄像机，摄像机的镜头可以是可转动形式，也可以是不可转动形式，可根据实际需求选择。
S102、根据图像信息，建立与视频场景对应的三维图像。
本公开实施例中可实时根据摄像机拍摄的二维图像信息，建立动态的三维图像，及时性好，能够准确的从三维图像中获悉视频场景中的信息。
S103、获取三维图像中的人物的体征信息。
其中，这里所述的体征信息种类的选择和数量可以根据年龄群体分类的需求确定，凡是能够进行区分人物年龄的体征信息均在本公开实施例的保护范围内。
S104、根据获取的体征信息，对视频场景中的人物进行年龄群体分类。
在获取到体征信息后，根据体征信息对视频场景中的人物进行年龄群体分类。
本公开实施例中，利用摄像机拍摄的二维图像进行三维重建，在建立的三维图像中获取人物的体征信息，并根据获取的体征信息对视频场景中的人物进行年龄群体分类，由于重建技术能较好的复原原始图像，较为精确，并且能够适应镜头角度的变化，因此能够在建立三维图像中，获取到更加精确的体征信息，从而提高年龄群体分类的准确性。
本公开实施例中可以采用多个摄像机获取视频场景的图像信息，多个镜头分别布置于视频场景中的不同方位，从而提高图像信息的准确性和拍摄效率。其中，在利用摄像机拍摄的图像信息进行三维重建之前，还需对相对于视频场景位于不同拍摄角度的摄像机进行标定，获取每一摄像机的内部参数和外部参数，还原摄像机的图像坐标系和世界坐标系的对应关系。其中内部参数包括：焦距、成像原点和畸变系数，外部参数包括：光学特征参数矩阵和平移矩阵。
其中，摄像机标定的过程为：首先准备一份棋盘格的样本，在棋盘格上标定特征点，本公开实施例中可采用红色的直径为1cm的实心圆点标注。多个镜头在不同角度下依次采集若干张样本图像，通过霍夫圆检测算法，提取样本图像中每个特征点的轮廓，拟合为圆形，并定位出每个圆形的圆心，以确定的圆心作为对应于三维世界中棋盘格样本上标注的特征点，从而计算出摄像机的内部参数和外部参数。其中，在标定时还需进行校正操作，降低摄像机镜头的畸变参数对拍摄图像质量的影响，减小拍摄图像的失真，提高显示画面的真实性。
如图2所示，在根据摄像机拍摄的图像信息，建立与视频场景对应的三维图像的过程可以包括：
S201、提取图像信息中的特征点。
其中，特征点的选择可根据实际需求确定。
S202、在不同角度拍摄的图像信息中，根据提取的特征点进行立体特征匹配。
其中，可采用shift特征匹配算法进行立体特征匹配。
S203、根据立体特征匹配结果和每一图像信息对应的摄像机的内部参数和外部参数，建立与视频场景对应的三维图像。
本公开实施例中所要获取的体征信息可以为人物的身高信息和肩宽信息。具有国际儿童和成人的身高肩宽作为参考数据，因此通过身高和肩宽信息进行 年龄群体分类有较好的依据，方案的实用性和可扩展性较好。
其中，如图3所示，获取三维图像中的人物的身高信息和肩宽信息的实现方法包括：
S301、确定三维图像中的人物的轮廓信息。
在建立三维图像后，对三维图像进行目标检测和特征提取，以将人体轮廓信息提取出来。
其中，可通过背景差分法，对多帧图像信息进行自适应背景学习，获得参考背景，然后根据参考背景，采用图像分割算法，对后续的图像信息进行分割处理，提取差异目标特征，对差异目标特征做Gabor滤波处理，排除差异目标特征在过滤范围内的图像信息，通过shift跟踪算法对图像信息进行差异目标特征的运动匹配跟踪，当持续存在差异目标特征的图像信息的帧数超过预设帧数时，将其标记为有效确定的目标(即人物)。
S302、根据确定的轮廓信息，用矩形框框出三维图像中的人物。
在确定人体轮廓信息后，用矩形框框住确定的人体轮廓，并对框住的人物进行目标跟踪，并根据人物形态的变化实时调整矩形框的形状。若矩形框框住的人物消失在图像中，则放弃对该人物的身高和肩宽信息的获取。
S303、计算矩形框的长度和宽度。
S304、将矩形框的长度确定为人物的身高，将矩形框的宽度确定为人物的肩宽。
在获取到的体征信息后，可将体征信息输入预设的多级分类器，并根据多级分类器的输出结果，确定每一体征信息所属的年龄群体。
其中，这里所述的预设的多级分类器为预先训练好的分类器，是由多个弱分类级联构成的多级强分类，能够更准确的对体征信息进行分类。
因此，在将获取的体征信息输入预设的多级分类器的步骤之前，还需构建多级分类器。其中，构建多级分类器的步骤包括：
S401、从三维图像中获取样本信息。
其中，样本信息包括：三维图像中的人物的体征信息(即正样本)和部分物体的形态信息(即负样本)，通过增加负样本信息，可提高分类器训练的准确性。
S402、分配每一样本信息相同的权重值。
在开始构建多级分类器，首先对每一样本信息分配相同的权重值，如若存在N个样本信息，则每一个样本信息的权重值为1/N，也就是每个样本都具有相同的话语权。
S403、将分配有权重值的样本信息依次输入多个基础分类器中进行分类训练。
在对每一样本信息分配权重值之后，将样本信息依次输入多个基础分类器重进行分类训练，以构建多级分类器。
S404、根据每一基础分类器输出的分类结果，重新调整每一样本信息的权重值，并输入下一基础分类器中进行分类训练；以及根据每一基础分类器输出的分类结果，计算每一基础分类器的分类误差值，并根据分类误差值的大小，计算每一基础分类器权重值。
本公开实施例可以采用Adaboost算法，进行分类器的训练，其中，每个基础分类器经过样本信息进行分类训练后，会产生一个分类结果，对于分类准确的样本信息，降低其权重值，对于分类错误的样本信息，增加其权重值，然后将重新分配权重值的样本信息输入下一个基础分类器进行训练，如图5所示。同时计算完成分类训练的基础分类器的分类误差值，根据分类误差值，计算该基础分类器的权重值，一般分类误差值越小，计算得到的权重值越大。
S405、将完成分类训练的多个基础分类器构建成多级强分类器。
当全部基础分类器完成训练后，每一基础分类器均会被分配一权重值，所有基础分类器级联在一起，构成多级强分类器。
本公开实施例中年龄群组可以包括：年龄段在0～7岁的幼儿组，在8～17岁的少年组、年龄段在18～30岁的青年组、年龄段在31～45岁的中青年组、年龄段在46～60岁的中年组以及年龄段在60岁以上的老年组，共6个分组。其中，实施年龄群组分类的过程，就是识别出三维图像中的人物归属于上面哪一个年龄群组。
其中，强分类器中的基础分类器的数量，可根据年龄群组的数据进行计算，即根据预设公式：N＝c×(c-1)/2，计算多级分类器中的基础分类器的数量。
其中，N表示基础分类器的数量，c表示年龄群体分类的数量。例如本公开 实施例中共有6个年龄群体分组，那么需要的基础分类器的数量为15。当进行分类器训练时，将分配有权重值的样本信息依次输入N个基础分类器中进行分类训练，其中每一基础分类器能够分类出两种年龄群体。一般情况下，每个基础分类器分类出的两个年龄群体不同于其他基础分类器所分类处的两个年龄群体。
综上所述，本公开实施例提供的技术方案，利用标定过的摄像机拍摄的二维图像进行三维重建，在建立的三维图像中获取人物的体征信息，并根据获取的体征信息对视频场景中的人物进行年龄群体分类，由于重建技术能较好的复原原始图像，较为精确，并且能够适应镜头角度的变化，因此能够在建立三维图像中，获取到更加精确的体征信息，从而提高年龄群体分类的准确性。
第二实施例
本公开实施例提供了一种年龄群体分类的装置，如图6所示，该装置包括：
图像获取模块601，被配置为获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息。
其中，在一视频场景中可根据实际需求布置一个或多个摄像机，以采集视频场景不同角度的图像信息。摄像机可以是单目摄像机也可以是多目摄像机，摄像机的镜头可以是可转动形式，也可以是不可转动形式，可根据实际需求选择。
图像建立模块602，被配置为根据图像获取模块601所获取的图像信息，建立与视频场景对应的三维图像。
本公开实施例中可实时根据摄像机拍摄的二维图像信息，建立动态的三维图像，及时性好，能够准确的从三维图像中获悉视频场景中的信息。
体征获取模块603，被配置为获取图像建立模块602所建立的三维图像中的人物的体征信息。
其中，这里所述的体征信息种类的选择和数量可以根据年龄群体分类的需求确定，凡是能够进行区分人物年龄的体征信息均在本公开实施例的保护范围内。
分类模块604，被配置为根据体征获取模块603所获取的体征信息，对视频场景中的人物进行年龄群体分类。
在获取到体征信息后，根据体征信息对视频场景中的人物进行年龄群体分类。
该装置还可以包括：
参数获取模块，被配置为对相对于视频场景位于不同拍摄角度的摄像机进行标定，获取每一摄像机的内部参数和外部参数。
本公开实施例中采用多个摄像机获取视频场景的图像信息，多个镜头分别布置于视频场景中的不同方位，从而提高图像信息的准确性和拍摄效率。
其中内部参数包括：焦距、成像原点和畸变系数，外部参数包括：光学特征参数矩阵和平移矩阵。
其中，摄像机标定的过程为：首先准备一份棋盘格的样本，在棋盘格上标定特征点，本公开实施例中可采用红色的直径为1cm的实心圆点标注。多个镜头在不同角度下依次采集若干张样本图像，通过霍夫圆检测算法，提取样本图像中每个特征点的轮廓，拟合为圆形，并定位出每个圆形的圆心，以确定的圆心作为对应于三维世界中棋盘格样本上标注的特征点，从而计算出摄像机的内部参数和外部参数。其中，在标定时还需进行校正操作，降低摄像机镜头的畸变参数对拍摄图像质量的影响，减小拍摄图像的失真，提高显示画面的真实性。
该图像建立模块可以包括：
提取单元，被配置为提取图像信息中的特征点。
其中，特征点的选择可根据实际需求确定。
匹配单元，被配置为在不同角度拍摄的图像信息中，根据特征点进行立体特征匹配。
其中，可采用shift特征匹配算法进行立体特征匹配。
图像建立单元，被配置为根据立体特征匹配结果和每一图像信息对应的摄像机的内部参数和外部参数，建立与视频场景对应的三维图像。
该体征获取模块可以包括：
体征获取单元，被配置为获取三维图像中的人物的身高信息和肩宽信息。
其中，该分类模块包括：
分类单元，被配置为根据获取的身高信息和肩宽信息，对视频场景中的人物进行年龄群体分类。
本公开实施例中所要获取的体征信息为人物的身高信息和肩宽信息。具有国际儿童和成人的身高肩宽作为参考数据，因此通过身高和肩宽信息进行年龄群体分类有较好的依据，方案的实用性和可扩展性较好。
该体征获取单元可以包括：
确定子单元，被配置为确定三维图像中的人物的轮廓信息。
在建立三维图像后，对三维图像进行目标检测和特征提取，以将人体轮廓信息提取出来。
其中，可通过背景差分法，对多帧图像信息进行自适应背景学习，获得参考背景，然后根据参考背景，采用图像分割算法，对后续的图像信息进行分割处理，提取差异目标特征，对差异目标特征做Gabor滤波处理，排除差异目标特征在过滤范围内的图像信息，通过shift跟踪算法对图像信息进行差异目标特征的运动匹配跟踪，当持续存在差异目标特征的图像信息的帧数超过预设帧数时，将其标记为有效确定的目标(即人物)。
第一处理子单元，被配置为根据确定的轮廓信息，用矩形框框出三维图像中的人物。
在确定人体轮廓信息后，用矩形框框住确定的人体轮廓，并对框住的人物进行目标跟踪，并根据人物形态的变化实时调整矩形框的形状。若矩形框框住的人物消失在图像中，则放弃对该人物的身高和肩宽信息的获取。
第一计算子单元，被配置为计算矩形框的长度和宽度。
第二处理子单元，被配置为将矩形框的长度确定为人物的身高，将矩形框的宽度确定为人物的肩宽。
该分类模块可以包括：
第一处理单元，被配置为将获取的体征信息输入预设的多级分类器。
决策单元，被配置为根据多级分类器的输出结果，确定每一体征信息所属的年龄群体。
在获取到的体征信息后，可将体征信息输入预设的多级分类器，并根据多级分类器的输出结果，确定每一体征信息所属的年龄群体。
其中，这里所述的预设的多级分类器为预先训练好的分类器，是由多个弱分类级联构成的多级强分类，能够更准确的对体征信息进行分类。
所该装置还可以包括：
构建模块，被配置为构建多级分类器。
在将获取的体征信息输入预设的多级分类器的步骤之前，还需构建多级分类器。
该构建模块可以包括：
样本获取单元，被配置为从三维图像中获取样本信息。
其中，样本信息包括：三维图像中的人物的体征信息(即正样本)和部分物体的形态信息(即负样本)，通过增加负样本信息，可提高分类器训练的准确性。
权重分配单元，被配置为分配每一样本信息相同的权重值。
在开始构建多级分类器，首先对每一样本信息分配相同的权重值，如若存在N个样本信息，则每一个样本信息的权重值为1/N，也就是每个样本都具有相同的话语权。
第二处理单元，被配置为将分配有权重值的样本信息依次输入多个基础分类器中进行分类训练。
在对每一样本信息分配权重值之后，将样本信息依次输入多个基础分类器重进行分类训练，以构建多级分类器。
第三处理单元，被配置为根据每一基础分类器输出的分类结果，重新调整每一样本信息的权重值，并输入下一基础分类器中进行分类训练；以及根据每一基础分类器输出的分类结果，计算每一基础分类器的分类误差值，并根据分类误差值的大小，计算每一基础分类器权重值。
本公开实施例可以采用Adaboost算法，进行分类器的训练，其中，每个基础分类器经过样本信息进行分类训练后，会产生一个分类结果，对于分类准确的样本信息，降低其权重值，对于分类错误的样本信息，增加其权重值，然后将重新分配权重值的样本信息输入下一个基础分类器进行训练，如图5所示。同时计算完成分类训练的基础分类器的分类误差值，根据分类误差值，计算该基础分类器的权重值，一般分类误差值越小，计算得到的权重值越大。
构建单元，被配置为将完成分类训练的多个基础分类器构建成多级强分类器。
当全部基础分类器完成训练后，每一基础分类器均会被分配一权重值，所有基础分类器级联在一起，构成多级强分类器。
该第二处理单元可以包括：
第二计算子单元，被配置为根据预设公式：N＝c×(c-1)/2，计算多级分类器中的基础分类器的数量。
其中，N表示基础分类器的数量，c表示年龄群体分类的数量。例如本公开实施例中共有6个年龄群体分组，那么需要的基础分类器的数量为15。
第三处理子单元，被配置为将分配有权重值的样本信息依次输入N个基础分类器中进行分类训练。
其中每一基础分类器能够分类出两种年龄群体。一般情况下，每个基础分类器分类出的两个年龄群体不同于其他基础分类器所分类处的两个年龄群体。
本公开实施例中年龄群组包括：年龄段在0～7岁的幼儿组，在8～17岁的少年组、年龄段在18～30岁的青年组、年龄段在31～45岁的中青年组、年龄段在46～60岁的中年组以及年龄段在60岁以上的老年组，共6个分组。其中，实施年龄群组分类的过程，就是识别出三维图像中的人物归属于上面哪一个年龄群组。
本公开实施例还提供了一种非暂态计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令设置为执行上述任一实施例中的方法。
本公开实施例还提供了一种电子设备的结构示意图。参见图7，该电子设备包括：
至少一个处理器(processor)70，图7中以一个处理器70为例；和存储器(memory)71，还可以包括通信接口(Communications Interface)72和总线73。其中，处理器70、通信接口72、存储器71可以通过总线73完成相互间的通信。通信接口72可以用于信息传输。处理器70可以调用存储器71中的逻辑指令，以执行上述实施例的方法。
此外，上述的存储器71中的逻辑指令可以通过软件功能单元的形式实现并 作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。
存储器71作为一种计算机可读存储介质，可用于存储软件程序、计算机可执行程序，如本公开实施例中的方法对应的程序指令/模块。处理器70通过运行存储在存储器71中的软件程序、指令以及模块，从而执行功能应用以及数据处理，即实现上述方法实施例中的年龄群体分类的方法。
存储器71可包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需的应用程序；存储数据区可存储根据终端设备的使用所创建的数据等。此外，存储器71可以包括高速随机存取存储器，还可以包括非易失性存储器。
本公开实施例的技术方案可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括一个或多个指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本公开实施例所述方法的全部或部分步骤。而前述的存储介质可以是非暂态存储介质，包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等多种可以存储程序代码的介质，也可以是暂态存储介质。
综上所述，本公开实施例提供的技术方案，利用标定过的摄像机拍摄的二维图像进行三维重建，在建立的三维图像中获取人物的体征信息，并根据获取的体征信息对视频场景中的人物进行年龄群体分类，由于重建技术能较好的复原原始图像，较为精确，并且能够适应镜头角度的变化，因此能够在建立三维图像中，获取到更加精确的体征信息，从而提高年龄群体分类的准确性。
需要说明的是，本公开实施例中的年龄群体分类的装置是与第一实施例中的年龄群体分类的方法的对应的装置，上述方法实施例中所有实现方式均适用于该装置的实施例中，也能达到相同的技术效果。
以上所述的是本公开的实施方式，应当指出对于本技术领域的普通人员来说，在不脱离本公开实施例的范围的前提下还可以作出若干改进和润饰，这些 改进和润饰也在本公开的保护范围内。
工业实用性
本申请提供的年龄群体分类的方法及装置，能够提高年龄群体分类的准确性。

Claims (19)

  1. 一种年龄群体分类的方法，包括：
    获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息；
    根据所述图像信息，建立与所述视频场景对应的三维图像；
    获取所述三维图像中的人物的体征信息；
    根据获取的体征信息，对所述视频场景中的人物进行年龄群体分类。
  2. 根据权利要求1所述的方法，其中，所述获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息的步骤之前，所述方法还包括：
    对相对于所述视频场景位于不同拍摄角度的所述摄像机进行标定，获取每一摄像机的内部参数和外部参数。
  3. 根据权利要求1或2所述的方法，其中，所述根据所述图像信息，建立与所述视频场景对应的三维图像的步骤包括：
    提取所述图像信息中的特征点；
    在不同角度拍摄的图像信息中，根据所述特征点进行立体特征匹配；
    根据立体特征匹配结果和每一图像信息对应的摄像机的内部参数和外部参数，建立与所述视频场景对应的三维图像。
  4. 根据权利要求1所述的方法，其中，所述获取所述三维图像中的人物的体征信息的步骤包括：
    获取所述三维图像中的人物的身高信息和肩宽信息；
    其中，所述根据获取的体征信息，对所述视频场景中的人物进行年龄群体分类的步骤包括：
    根据获取的身高信息和肩宽信息，对所述视频场景中的人物进行年龄群体分类。
  5. 根据权利要求4所述的方法，其中，所述获取所述三维图像中的人物的身高信息和肩宽信息的步骤包括：
    确定所述三维图像中的人物的轮廓信息；
    根据确定的轮廓信息，用矩形框框出所述三维图像中的人物；
    计算所述矩形框的长度和宽度；
    将所述矩形框的长度确定为人物的身高信息，将所述矩形框的宽度确定为 人物的肩宽信息。
  6. 根据权利要求1所述的方法，其中，所述根据获取的体征信息，对所述视频场景中人物的年龄群体进行分类的步骤包括：
    将获取的体征信息输入预设的多级分类器；
    根据所述多级分类器的输出结果，确定每一体征信息所属的年龄群体。
  7. 根据权利要求6所述的方法，其中，在所述将获取的体征信息输入预设的多级分类器的步骤之前，所述方法还包括：
    构建所述多级分类器。
  8. 根据权利要求7所述的方法，其中，所述构建所述多级分类器的步骤包括：
    从所述三维图像中获取样本信息，其中样本信息包括：所述三维图像中的人物的体征信息和部分物体的形态信息；
    分配每一样本信息相同的权重值；
    将分配有权重值的样本信息依次输入多个基础分类器中进行分类训练；
    根据每一基础分类器输出的分类结果，重新调整每一样本信息的权重值，并输入下一基础分类器中进行分类训练；以及根据每一基础分类器输出的分类结果，计算每一基础分类器的分类误差值，并根据分类误差值的大小，计算每一基础分类器权重值；
    将完成分类训练的多个基础分类器构建成多级强分类器。
  9. 根据权利要求8所述的方法，其中，所述将分配有权重值的样本信息依次输入多个基础分类器中进行分类训练的步骤包括：
    根据预设公式：N＝c×(c-1)/2，计算所述多级分类器中的基础分类器的数量；
    其中，N表示基础分类器的数量，c表示年龄群体分类的数量；
    将分配有权重值的样本信息依次输入N个基础分类器中进行分类训练，其中每一基础分类器能够分类出两种年龄群体。
  10. 一种年龄群体分类的装置，包括：
    图像获取模块，被配置为获取至少一摄像机在预定时间段内拍摄的一视频场景的图像信息；
    图像建立模块，被配置为根据所述图像获取模块所获取的所述图像信息，建立与所述视频场景对应的三维图像；
    体征获取模块，被配置为获取所述图像建立模块所建立的所述三维图像中的人物的体征信息；
    分类模块，被配置为根据所述体征获取模块所获取的体征信息，对所述视频场景中的人物进行年龄群体分类。
  11. 根据权利要求10所述的装置，还包括：
    参数获取模块，被配置为对相对于所述视频场景位于不同拍摄角度的所述摄像机进行标定，获取每一摄像机的内部参数和外部参数。
  12. 根据权利要求10或11所述的装置，其中，所述图像建立模块包括：
    提取单元，被配置为提取所述图像信息中的特征点；
    匹配单元，被配置为在不同角度拍摄的图像信息中，根据所述特征点进行立体特征匹配；
    图像建立单元，被配置为根据立体特征匹配结果和每一图像信息对应的摄像机的内部参数和外部参数，建立与所述视频场景对应的三维图像。
  13. 根据权利要求10所述的装置，其中，所述体征获取模块包括：
    体征获取单元，被配置为获取所述三维图像中的人物的身高信息和肩宽信息；
    其中，所述分类模块包括：
    分类单元，被配置为根据获取的身高信息和肩宽信息，对所述视频场景中的人物进行年龄群体分类。
  14. 根据权利要求13所述的装置，其中，所述体征获取单元包括：
    确定子单元，被配置为确定所述三维图像中的人物的轮廓信息；
    第一处理子单元，被配置为根据确定的轮廓信息，用矩形框框出所述三维图像中的人物；
    第一计算子单元，被配置为计算所述矩形框的长度和宽度；
    第二处理子单元，被配置为将所述矩形框的长度确定为人物的身高信息，将所述矩形框的宽度确定为人物的肩宽信息。
  15. 根据权利要求10所述的装置，其中，所述分类模块包括：
    第一处理单元，被配置为将获取的体征信息输入预设的多级分类器；
    决策单元，被配置为根据所述多级分类器的输出结果，确定每一体征信息所属的年龄群体。
  16. 根据权利要求15所述的装置，还包括：
    构建模块，被配置为构建所述多级分类器。
  17. 根据权利要求16所述的装置，其中，所述构建模块包括：
    样本获取单元，被配置为从所述三维图像中获取样本信息，其中样本信息包括：所述三维图像中的人物的体征信息和部分物体的形态信息；
    权重分配单元，被配置为分配每一样本信息相同的权重值；
    第二处理单元，被配置为将分配有权重值的样本信息依次输入多个基础分类器中进行分类训练；
    第三处理单元，被配置为根据每一基础分类器输出的分类结果，重新调整每一样本信息的权重值，并输入下一基础分类器中进行分类训练；以及根据每一基础分类器输出的分类结果，计算每一基础分类器的分类误差值，并根据分类误差值的大小，计算每一基础分类器权重值；
    构建单元，被配置为将完成分类训练的多个基础分类器构建成多级强分类器。
  18. 根据权利要求17所述的装置，其中，所述第二处理单元包括：
    第二计算子单元，被配置为根据预设公式：N＝c×(c-1)/2，计算所述多级分类器中的基础分类器的数量；
    其中，N表示基础分类器的数量，c表示年龄群体分类的数量；
    第三处理子单元，被配置为将分配有权重值的样本信息依次输入N个基础分类器中进行分类训练，其中每一基础分类器能够分类出两种年龄群体。
  19. 一种非暂态计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令设置为执行权利要求1-9中任一项的方法。
PCT/CN2017/087184 2016-08-23 2017-06-05 一种年龄群体分类的方法及装置 WO2018036241A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610711538.2A CN107766782A (zh) 2016-08-23 2016-08-23 一种年龄群体分类的方法及装置
CN201610711538.2 2016-08-23

Publications (1)

Publication Number Publication Date
WO2018036241A1 true WO2018036241A1 (zh) 2018-03-01

Family

ID=61245445

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/087184 WO2018036241A1 (zh) 2016-08-23 2017-06-05 一种年龄群体分类的方法及装置

Country Status (2)

Country Link
CN (1) CN107766782A (zh)
WO (1) WO2018036241A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112148910A (zh) * 2019-06-28 2020-12-29 富士胶片株式会社 图像处理装置、图像处理方法及存储有图像处理程序的记录介质
US11373063B2 (en) * 2018-12-10 2022-06-28 International Business Machines Corporation System and method for staged ensemble classification

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109829454A (zh) * 2019-01-17 2019-05-31 柳州康云互联科技有限公司 一种基于预训练标识的图像特征采集方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102496001A (zh) * 2011-11-15 2012-06-13 无锡港湾网络科技有限公司 一种视频监控目标自动检测的方法和系统
CN102697508A (zh) * 2012-04-23 2012-10-03 中国人民解放军国防科学技术大学 采用单目视觉的三维重建来进行步态识别的方法
CN102930454A (zh) * 2012-10-07 2013-02-13 乐配(天津)科技有限公司 基于多感知技术的智能3d广告推荐方法
CN104112209A (zh) * 2013-04-16 2014-10-22 苏州和积信息科技有限公司 显示终端的受众统计方法和系统
CN104408412A (zh) * 2014-11-20 2015-03-11 苏州福丰科技有限公司 一种用于保险柜的三维人脸识别方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8351647B2 (en) * 2002-07-29 2013-01-08 Videomining Corporation Automatic detection and aggregation of demographics and behavior of people
US7848548B1 (en) * 2007-06-11 2010-12-07 Videomining Corporation Method and system for robust demographic classification using pose independent model from sequence of face images
CN104537353A (zh) * 2015-01-07 2015-04-22 深圳市唯特视科技有限公司 基于三维点云的三维人脸年龄分类装置及方法
CN104915000A (zh) * 2015-05-27 2015-09-16 天津科技大学 用于裸眼3d广告的多感知生物识别交互方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102496001A (zh) * 2011-11-15 2012-06-13 无锡港湾网络科技有限公司 一种视频监控目标自动检测的方法和系统
CN102697508A (zh) * 2012-04-23 2012-10-03 中国人民解放军国防科学技术大学 采用单目视觉的三维重建来进行步态识别的方法
CN102930454A (zh) * 2012-10-07 2013-02-13 乐配(天津)科技有限公司 基于多感知技术的智能3d广告推荐方法
CN104112209A (zh) * 2013-04-16 2014-10-22 苏州和积信息科技有限公司 显示终端的受众统计方法和系统
CN104408412A (zh) * 2014-11-20 2015-03-11 苏州福丰科技有限公司 一种用于保险柜的三维人脸识别方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11373063B2 (en) * 2018-12-10 2022-06-28 International Business Machines Corporation System and method for staged ensemble classification
CN112148910A (zh) * 2019-06-28 2020-12-29 富士胶片株式会社 图像处理装置、图像处理方法及存储有图像处理程序的记录介质

Also Published As

Publication number Publication date
CN107766782A (zh) 2018-03-06

Similar Documents

Publication Publication Date Title
CN110147721B (zh) 一种三维人脸识别方法、模型训练方法和装置
US11928800B2 (en) Image coordinate system transformation method and apparatus, device, and storage medium
CN109829396B (zh) 人脸识别运动模糊处理方法、装置、设备及存储介质
WO2019200702A1 (zh) 去网纹系统训练方法、去网纹方法、装置、设备及介质
CN107622497B (zh) 图像裁剪方法、装置、计算机可读存储介质和计算机设备
WO2020088029A1 (zh) 活体检验方法、存储介质和电子设备
CN111144284B (zh) 深度人脸图像的生成方法、装置、电子设备及介质
WO2018036241A1 (zh) 一种年龄群体分类的方法及装置
CN112200056B (zh) 人脸活体检测方法、装置、电子设备及存储介质
CN112836625A (zh) 人脸活体检测方法、装置、电子设备
CN111783593A (zh) 基于人工智能的人脸识别方法、装置、电子设备及介质
CN113902781A (zh) 三维人脸重建方法、装置、设备及介质
CN112802081A (zh) 一种深度检测方法、装置、电子设备及存储介质
US10791321B2 (en) Constructing a user's face model using particle filters
JP2022133378A (ja) 顔生体検出方法、装置、電子機器、及び記憶媒体
CN110007764B (zh) 一种手势骨架识别方法、装置、系统及存储介质
CN113902849A (zh) 三维人脸模型重建方法、装置、电子设备及存储介质
WO2022246605A1 (zh) 一种关键点标定方法和装置
CN113762009B (zh) 一种基于多尺度特征融合及双注意力机制的人群计数方法
CN112396016B (zh) 一种基于大数据技术的人脸识别系统
CN111353325A (zh) 关键点检测模型训练方法及装置
CN115880643B (zh) 一种基于目标检测算法的社交距离监测方法和装置
WO2020244076A1 (zh) 人脸识别方法、装置、电子设备及存储介质
EP3699865B1 (en) Three-dimensional face shape derivation device, three-dimensional face shape deriving method, and non-transitory computer readable medium
CN111461971B (zh) 图像处理方法、装置、设备及计算机可读存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17842666

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17842666

Country of ref document: EP

Kind code of ref document: A1