WO2020019457A1 - 用户指令匹配方法、装置、计算机设备及存储介质 - Google Patents

用户指令匹配方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2020019457A1
WO2020019457A1 PCT/CN2018/106432 CN2018106432W WO2020019457A1 WO 2020019457 A1 WO2020019457 A1 WO 2020019457A1 CN 2018106432 W CN2018106432 W CN 2018106432W WO 2020019457 A1 WO2020019457 A1 WO 2020019457A1
Authority
WO
WIPO (PCT)
Prior art keywords
participant
image
matching
gesture
current
Prior art date
Application number
PCT/CN2018/106432
Other languages
English (en)
French (fr)
Inventor
朱文和
Original Assignee
平安科技（深圳）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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2020019457A1 publication Critical patent/WO2020019457A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/107—Static hand or arm
    • G06V40/113—Recognition of static hand signs
    • G—PHYSICS
    • G07—CHECKING-DEVICES
    • G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
    • G07C15/00—Generating random numbers; Lottery apparatus

Definitions

  • the present application relates to the field of data processing, and in particular, to a user instruction matching method, device, computer device, and storage medium.
  • the traditional lottery method is manually drawn by the lottery staff. With the development of the society, the lottery form has begun to be networked and automated. Generally, the lottery information of each user is collected to the network database end. The network database end will meet the conditions after certain information screening. The information is sent to the lottery system, for example, only the first one hundred of the lottery information are selected according to the sending time. Further, through the computer lottery system, a plurality of winners are randomly selected from the first hundred pieces of lottery information.
  • the lottery process needs to meet transparency and fairness, if the network database receives more information, it will continue to send information, and frequent information transmission causes the system stability to deteriorate, or the server and mobile terminal are communicating. In the process, because of the instability of the network, information may be lost. However, if the instruction is in the waiting stage due to incomplete instructions, it will make the entire instruction match take too much time and reduce the efficiency of the entire instruction match.
  • the embodiments of the present application provide a user instruction matching method, device, computer equipment, and storage medium to solve the problem of low user instruction matching efficiency.
  • a user instruction matching method includes:
  • a user instruction matching device includes:
  • a current image acquisition module configured to acquire a current image of each participant, where the current image includes a face area and a gesture area;
  • a matching participant identification acquisition module configured to match the face region of each of the current images in a participant image database to obtain a matching participant identification of each of the current images
  • a participant instruction information acquisition module configured to identify the gesture area of each of the current images, obtain participant instruction information, and associate the participant instruction information with the corresponding matching participant identifier;
  • a gesture image recognition module configured to obtain a gesture image of a decision maker, identify the gesture image of the decision maker, and obtain decision instruction information of the decision maker;
  • the instruction matching result acquisition module is configured to match the participant instruction information corresponding to each of the matching participant identifiers with the decision instruction information of the decision maker through a mapping relationship table to obtain an instruction matching result for each participant.
  • a computer device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor.
  • the processor executes the computer-readable instructions, the following steps are implemented:
  • One or more non-volatile readable storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to perform the following steps:
  • FIG. 1 is a schematic diagram of an application environment of a user instruction matching method according to an embodiment of the present application
  • FIG. 2 is an example diagram of a user instruction matching method according to an embodiment of the present application.
  • FIG. 3 is another example diagram of a user instruction matching method according to an embodiment of the present application.
  • FIG. 4 is another example diagram of a user instruction matching method in an embodiment of the present application.
  • FIG. 5 is another example diagram of a user instruction matching method in an embodiment of the present application.
  • FIG. 6 is another example diagram of a user instruction matching method according to an embodiment of the present application.
  • FIG. 7 is another example diagram of a user instruction matching method according to an embodiment of the present application.
  • FIG. 8 is a schematic block diagram of a user instruction matching device according to an embodiment of the present application.
  • FIG. 9 is a schematic diagram of a computer device according to an embodiment of the present application.
  • the user instruction matching method provided in this application can be applied in the application environment as shown in FIG. 1, in which a client (computer device) communicates with a server through a network.
  • the client collects the participant's current image and the decision-maker's gesture image and sends it to the server.
  • the server processes the acquired current image and gesture image to obtain the instruction matching result of each participant.
  • the client computer device
  • the server can be implemented by an independent server or a server cluster composed of multiple servers.
  • a method for matching user instructions is provided.
  • the method is applied to the server in FIG. 1 as an example, and includes the following steps:
  • S10 Acquire a current image of each participant, where the current image includes a face area and a gesture area.
  • Participants refer to users who need to perform instruction matching.
  • the current image refers to the participant's image collected by the client.
  • the current image includes the face area and gesture area. That is, the client collects participants 'images, and the collected participants' images include the participant's face area and gesture area.
  • the current image of each participant can be collected centrally through one client or separately through multiple clients.
  • the timing of the current image collection of each participant can be staggered or performed simultaneously.
  • a predetermined number of clients can be set according to the location of participants, and a client can be set to collect the current image of a participant, or a client can be set to collect the current number of participants image. Understandably, the timing of collecting the current images of the participants by different clients may be performed simultaneously or may be staggered according to actual needs, which is not specifically limited herein.
  • a plurality of clients are set, where each client collects current images of a predetermined number of participants, and the number of participants collected by each client can be determined according to the location set by the client, and can be set Each client captures the current images of all participants in the client's captured image area. Moreover, the plurality of clients are set to simultaneously collect current images of participants. This acquisition method can improve the collection efficiency of the current image.
  • S20 Match the face area of each current image in the participant image database, and obtain the matching participant ID of each current image.
  • the participant image library is an image library in which all participant face images are stored in advance, and the matching participant identification is used to determine which face area in the current image matches which face image in the participant image library. , And it is reflected by a logo.
  • the matching participant identification may be an employee number, an ID number, a phone number of the participant, or other information that can uniquely identify the participant.
  • a corresponding face area is first obtained from the current image, and the face area can be determined by a face feature point detection algorithm.
  • the face feature point detection algorithm is used to identify the face area, and then the face area is rectangular. Box is marked.
  • the face feature point detection algorithm includes, but is not limited to, a deep learning-based face feature point detection algorithm, a model-based face feature point detection algorithm, or a cascade shape regression-based face feature point detection algorithm.
  • the face area of each current image is matched in the participant image database, and the identity of the successfully matched face image in the participant image database is output to obtain the matching participant identity of the current image.
  • the participant instruction information refers to instruction information determined according to the gesture area of the participant.
  • the participant instruction information is gesture instruction information, wherein the gesture instruction information is instruction information representing different gesture actions, such as: scissors, rock, or cloth, or gesture actions representing different numbers, such as: 1, 2 , 3, 4 or 5.
  • a gesture area is determined from each current image, and then the gesture area of each current image is identified to obtain participant instruction information.
  • an edge detection algorithm can be used to obtain an edge image of the current image, and then each knuckle point and two wrist points in the current image are found according to the curvature of the edge image.
  • the edge detection algorithm may use a differential edge detection algorithm, a Sobel edge detection algorithm, or a Reborts edge detection algorithm. After finding each knuckle point and two wrist points in the current image, a gesture region is obtained according to each knuckle point and two wrist points. Specifically, a fixed area may be determined according to each knuckle point and two wrist points.
  • the fixed area may be a rectangle or a circle, and the fixed area includes a gesture area.
  • the fixed area is set to a circle, and the center of the circle is determined according to the two wrist points and one of the knuckle points, and then multiplied by a certain distance according to the distance between the wrist point and one of the knuckle points.
  • the radius of the circle is determined by the ratio of the ratio. The ratio can be set according to actual needs, and is not specifically limited here. After obtaining the gesture area, you can use the imageclipper tool to capture the gesture area. The imageclipper tool can quickly capture the gesture area from the gesture image.
  • the gesture area is input to a pre-trained gesture image recognition model for recognition, and participant instruction information is output, such as scissors, rock, or cloth.
  • participant instruction information is output, such as scissors, rock, or cloth.
  • the participant instruction information is associated with the corresponding matching participant identification, that is, the participant instruction information corresponding to each matching participant identification is marked.
  • S40 Acquire a gesture image of a decision maker, identify the gesture image of the decision maker, and obtain decision instruction information of the decision maker.
  • a decision maker refers to a user who provides a gesture image to assist in matching the participants' instructions.
  • the gesture image of the decision maker may be collected through a predetermined client, which may be a video camera, a camera, a scanner, or another device with a photographing function.
  • the client After the client obtains the instruction to acquire the gesture image of the decision maker from the server, the client collects the image of the location of the decision maker, and then sends the image of the location of the decision maker to the server.
  • the server After the server obtains the image of the decision maker's location, it can obtain a gesture image of the decision maker by taking a screenshot.
  • the method for acquiring the gesture image of the decision maker is the same as the method for acquiring the gesture area in step S30, and details are not described herein again.
  • the gesture image After obtaining the gesture image of the decision maker, the gesture image is input to a pre-trained gesture image recognition model for recognition, and the decision instruction information of the decision maker is output. Understandably, the decision instruction information corresponds to the participant instruction information. If the participant instruction information is gesture instruction information, the decision instruction information is also gesture instruction information. For example, if the participant instruction information represents scissors, stones, or cloth, the decision instruction information also represents scissors, stones, or cloth.
  • S50 Match the participant instruction information corresponding to each matching participant ID with the decision instruction information of the decision maker through a mapping relationship table to obtain the instruction matching result of each participant.
  • the mapping relation table is a preset table, and the mapping relation table can be matched with the decision instruction information of the decision maker according to the participant instruction information of the participants, and the corresponding instruction matching result is found in the mapping relationship table.
  • a current image of each participant is obtained first, where the current image includes a face area and a gesture area; the face area of each current image is matched in the participant image database to obtain each current Image matching participant identification; identifying the gesture area of each current image to obtain participant instruction information, correlating the participant instruction information with the corresponding matching participant identification; obtaining a gesture image of a decision maker, and identifying the gesture of the decision maker
  • the image is used to obtain the decision instruction information of the decision maker.
  • the participant instruction information corresponding to each matching participant identifier is matched with the decision instruction information of the decision maker through a mapping relationship table to obtain the instruction matching result of each participant.
  • the data is acquired by collecting the current image of the participant and the gesture image of the decision maker separately.
  • each participant it is not necessary for each participant to interact with the server through a terminal, avoiding the waiting period due to incomplete data acquisition. , Which improves the efficiency of user instruction matching. Moreover, after obtaining the current image of the participant and the gesture image of the decision maker, the corresponding instruction information is identified and matched by the mapping relationship table to obtain the instruction matching result of each participant, which also ensures that the user instruction matches. effectiveness.
  • the participant image database includes a reference participant identifier, a reference position identifier, and a reference face image.
  • the reference face image is a face image of a participant collected in advance. After the reference face image is collected, the reference face image of each participant is assigned a corresponding reference participant identifier and a reference position identifier.
  • the reference face image is used for subsequent matching with each current image.
  • a photo of the participant's ID or a photo on the work ID can be used as the reference face image.
  • the reference participant ID is an identifier used to determine which participant each reference face image belongs to.
  • the reference participant ID can be the participant's employee number, ID card number, phone number, or other uniquely identifiable participant identity Information.
  • the reference position identifier refers to an identifier of a position assigned to each participant in advance, and the reference position identifier may be represented by numbers, letters, or other computer-recognizable symbols.
  • this embodiment is applied to a conference or event site.
  • a client is set up to collect a reference face image of each participant.
  • the server numbers each participant in the order of collecting the participant's reference face image, and uses this number as the reference to participate
  • the identity of the participant is sent to the client, and the identity information of the participant can also be obtained through the client.
  • the identity information of the participant can be the employee number, ID card number, phone number, or other unique information that can identify the participant.
  • adding the participant's name to the information that can uniquely identify the participant can better identify the participant's identity.
  • the participant identity information is used as the reference participant ID, and the client sends the reference participant ID to the server.
  • the client's reference face image is collected through the client, and each time the server obtains a participant's reference face image, the participant is automatically assigned a location, and the reference location identifier corresponding to the location is recorded.
  • the server Record the reference participant ID, reference position ID, and reference face image of each participant to form a participant image library.
  • the current image includes a current location identifier.
  • the current position identifier refers to the identifier of the position where the participant is located when collecting the current image.
  • the current location identifier may be located according to the location of the client that collected the current image.
  • matching the face area of each current image in the participant image database to obtain the matching participant ID of each current image includes the following steps:
  • S21 Query the participant's image library for the corresponding reference position identifier according to the current position identifier of each current image, and match the face area of the current image with the reference face image of the corresponding reference position identifier.
  • the corresponding reference position identifier is queried in the participant image database according to the current position identifier of each current image, where the corresponding reference position identifier means that the reference position identifier is the same as the current position identifier. For example, if the current position identifier is 007, the reference position identifier that is also 007 in the participant image database corresponds to the current position identifier. After obtaining the reference position identifier corresponding to the current position identifier, the face region of the current image is matched with the reference face image of the reference position identifier. Specifically, the face area of the current image and the reference face image identified by the reference position can be respectively converted into feature vectors, and then the similarity of the two feature vectors is used to determine whether or not they match. Optionally, a similarity threshold may be set, and then the calculated similarity and the similarity threshold are compared to obtain a result of successful matching or failed matching.
  • the reference participant ID of the reference face image of the successfully matched participant image library is used as the corresponding participant ID of the current image.
  • all the reference face images corresponding to the reference position identifiers that failed to match and the corresponding reference participant identifiers are used as the reference face image database.
  • the feature vector converted from the face area of each current image and the feature vector of each reference face image in the reference face image database are used to calculate the feature vector similarity, and the reference face image with the highest feature vector similarity value That is, matching with the corresponding current image is successful.
  • the reference participant ID of the reference face image of the reference face image library that has been successfully matched is used as the matching participant ID of the corresponding current image.
  • the corresponding face image is first searched for matching according to the current position identifier and the reference position identifier, which avoids that each current image needs to be matched with all the reference face images in the participant image library when matching. Matching improves the efficiency of face matching. After a matching failure occurs, the remaining reference face images are matched by one-to-one comparison, which also ensures the integrity of the current image matching.
  • matching the face area of the current image with a reference face image corresponding to a reference position identifier specifically includes the following steps:
  • the matching feature vector refers to the feature vector of the face region of the current image, and is a vector used to characterize the image information features of the face region of the current image.
  • a projection-based feature vector such as PCA (Principal Component Analysis), Component analysis) feature vector
  • direction-based feature vectors such as HOG (Histogram of Oriented Gradient, gradient direction histogram) feature vectors
  • deep learning-based feature vectors such as convolutional neural network feature vectors.
  • the feature vector can represent the image information with simple data, and the subsequent comparison process can be simplified by extracting the feature vector of the face image.
  • the matching feature vector in this embodiment may be a feature vector based on deep learning.
  • deep convolutional neural network for feature extraction. Because deep learning can automatically learn from the data of the face image, it can be applied to a variety of environments, and it eliminates the need for complex preprocessing operations. Instead, it uses features based on projection, direction, and center of gravity. Vectors can often only extract one feature, such as color features or shape features, which are difficult to apply to realistic and complex environments. Therefore, matching feature vectors to feature vectors based on deep learning can improve the accuracy of subsequent face matching.
  • the reference face feature vector is a feature vector of the reference face image, and is a vector used to characterize the image information features of the reference face image.
  • the feature vector transformation method of the reference face feature vector is the same as the feature vector transformation method of the matching feature vector in step S221, and details are not described herein again.
  • the feature vector similarity between the two is calculated.
  • the feature vector similarity can be calculated by Euclidean distance algorithm, Manhattan distance algorithm, Minkowski distance algorithm or cosine similarity algorithm.
  • the Euclidean distance algorithm can be used to calculate the feature vector similarity between the matching feature vector and the reference face feature vector:
  • Feature vector where x i is the vector element in the matching feature vector, and y i is the vector element in the reference face feature vector, i is a positive integer, and 0 ⁇ i ⁇ n.
  • the feature vector similarity between the matching feature vector and the reference face feature vector is calculated.
  • the matching after calculating the similarity between the matching feature vector and the feature vector of the reference face feature vector, it is determined whether the matching is successful by comparing the magnitude between the feature vector similarity and a preset similarity threshold. If the feature vector similarity is greater than or equal to a preset similarity threshold, the matching is successful. If the feature vector similarity is less than a preset similarity threshold, the matching fails.
  • the similarity of the feature vector is calculated by calculating the feature vector similarity between the face area of the current image and the feature vector corresponding to the reference face image identified by the corresponding reference position, thereby ensuring the accuracy and efficiency of image matching.
  • acquiring a gesture image of a decision maker, identifying the gesture image of the decision maker, and obtaining the decision instruction information of the decision maker specifically includes the following steps:
  • S41 Acquire a gesture image of a decision maker, and intercept a gesture area image from the gesture image.
  • the gesture image of the decision maker is obtained. Because the gesture image of the decision maker is obtained, the area that the gesture image may encompass is relatively large, which is not conducive to subsequent recognition processing. Therefore, the gesture region needs to be extracted from the gesture image to obtain the gesture region image.
  • an edge detection algorithm may be used to obtain an edge image of the gesture image, and then each knuckle point and two wrist points in the gesture image are found according to the curvature of the edge image.
  • the edge detection algorithm may use a differential edge detection algorithm, a Sobel edge detection algorithm, or a Reborts edge detection algorithm.
  • a gesture region is obtained according to each knuckle point and two wrist points, and finally a gesture region image is obtained according to the gesture region.
  • a fixed area may be determined according to each knuckle point and two wrist points.
  • the fixed area may be a rectangle or a circle, and the fixed area includes a gesture area.
  • the fixed area is set to a circle, and the center of the circle is determined according to the two wrist points and one of the knuckle points, and then multiplied by a certain distance according to the distance between the wrist point and one of the knuckle points.
  • the radius of the circle is determined by the ratio of the ratio. The ratio can be set according to actual needs, and is not specifically limited here.
  • S42 Input the gesture area image into the gesture image recognition model for recognition, and obtain decision instruction information of a decision maker.
  • the output of the gesture recognition model can be used to obtain decision instruction information corresponding to the gesture area image.
  • the gesture image recognition model is a pre-trained recognition model. Through the gesture image recognition model, the type of the gesture area image can be quickly identified, and the decision instruction information of the decision maker is output to facilitate subsequent matching.
  • the gesture area image including the gesture area is captured, and then the gesture area image is input into the gesture image recognition model to identify the current gesture type, and the decision instruction information of the decision maker is obtained accordingly, thereby improving the accuracy of gesture recognition. Sex and efficiency.
  • the instruction matching method further includes the following steps:
  • S421 Obtain original images, classify and label each original image to form a gesture training image.
  • the original image is an image including different gestures collected in advance.
  • the images of different gestures can be obtained through cameras, or they can be obtained through existing data sets on the network.
  • each original image such as: "stone”, "scissors" or "cloth”.
  • each original image contains corresponding labeling data, and the original image containing the labeling data is used as a gesture training image.
  • S422 Use a gesture training image to train the convolutional neural network model to obtain a gesture image recognition model.
  • the Convolutional Neural Network (CNN) model is a feed-forward neural network. Its artificial neurons can respond to a part of the surrounding cells in the coverage area, and are often applied to the processing of large images.
  • Convolutional neural networks usually include at least two non-linearly trainable convolutional layers, at least two non-linear pooling layers, and at least one fully connected layer, that is, including at least five hidden layers, in addition to an input layer and an output Floor.
  • a gesture image recognition model obtained by training a convolutional neural network model through a gesture training image can more accurately classify the gesture training image.
  • the current gesture image is input into the gesture image recognition model for recognition, and the current gesture type corresponding to the current gesture image is obtained.
  • each original image is classified and labeled to form a gesture training image.
  • the convolutional neural network model is trained with gesture training images to obtain a gesture image recognition model, which better guarantees the recognition efficiency and accuracy of subsequent gesture area images.
  • the participant instruction information corresponding to each matching participant identifier is matched with the decision instruction information of the decision maker through a mapping relationship table to obtain the instruction matching result of each participant. Specifically, It includes the following steps:
  • the participant label refers to the numerical or symbolized information obtained by converting the participant's instruction information of the participant.
  • the participant's instruction information of the participant includes stones, scissors, and cloth
  • the three instruction information are correspondingly converted into numerical or symbolized information for the convenience of subsequent processing.
  • the "stone” instruction is mapped to " 1 "
  • scissors “instructions are mapped to” 2 "
  • cloth “instructions are mapped to” 3 ".
  • the decision maker label refers to the numerical or symbolized information obtained by converting the decision instruction information of the decision maker. Understandably, the decision instruction information of the decision maker corresponds to the target instruction information or random instruction information of the participants, such as: stone, scissors, and cloth.
  • the specific conversion method is also the same as step S51, and is not repeated here.
  • mapping relationship table After obtaining the participant labels and decision maker labels, matching is performed through the mapping relationship table, and the instruction matching result of each participant can be obtained. Specifically, a mapping table between participant labels and decision maker labels can be set in advance, and corresponding instruction matching results can be queried in the mapping relationship table according to different participant labels and different decision maker labels. For example: if the match result of the instruction includes three cases of win, draw, and negative, X, Y, and Z can be used to represent the three cases, respectively.
  • the instruction matching result of each participant is obtained by converting the corresponding instruction information into a label, and then obtaining the instruction matching result of each participant through a preset mapping relationship table, thereby improving the efficiency of obtaining the instruction matching result.
  • a user instruction matching device corresponds to the user instruction matching method in the foregoing embodiment.
  • the user instruction matching device includes a current image acquisition module 10, a matching participant identification acquisition module 20, a participant instruction information acquisition module 30, a gesture image recognition module 40, and an instruction matching result acquisition module 50.
  • the detailed description of each function module is as follows:
  • the current image acquisition module 10 is configured to acquire a current image of each participant, where the current image includes a face area and a gesture area.
  • the matching participant identification acquisition module 20 is configured to match a face area of each current image in a participant image database, and obtain a matching participant identification of each current image.
  • the participant instruction information acquisition module 30 is configured to identify a gesture area of each current image, obtain participant instruction information, and associate the participant instruction information with a corresponding matching participant identifier.
  • the gesture image recognition module 40 is configured to acquire a gesture image of a decision maker, and recognize the gesture image of the decision maker, to obtain decision instruction information of the decision maker.
  • the instruction matching result obtaining module 50 is configured to match the participant instruction information corresponding to each matching participant identifier with the decision instruction information of the decision maker through a mapping relationship table to obtain the instruction matching result of each participant.
  • the matching participant identification obtaining module 20 includes a first matching unit, a first matching participant identification obtaining unit, a second matching unit, and a second matching participant identification obtaining unit.
  • a first matching unit configured to query the corresponding reference position identifier in the participant image database according to the current position identifier of each current image, and match the face area of the current image with the reference face image corresponding to the reference position identifier .
  • the first matching participant ID acquiring unit is configured to use the reference participant ID of the reference face image of the participant image library that has been successfully matched as the matching participant ID of the corresponding current image.
  • the second matching unit is configured to obtain all reference face images corresponding to the reference position identifiers that failed to match and the corresponding reference participant identifiers, as a reference face image database, to compare the face region and reference of each current image that fails to match. Each reference face image in the face image library is matched.
  • the second matching participant identifier obtaining unit is configured to use the reference participant identifier of the reference face image of the reference face image library that has been successfully matched as the matching participant identifier of the corresponding current image.
  • the first matching participant identification obtaining unit includes a matching feature vector transformation subunit, a reference face feature vector obtaining subunit, and a feature vector similarity calculation subunit.
  • the matched feature vector transformation subunit is used to transform the face region of the current image into a matched feature vector.
  • a reference face feature vector acquisition subunit is used to obtain a reference face feature vector, wherein the reference face feature vector is obtained by performing a feature vector transformation on a reference face image corresponding to a corresponding reference position identifier.
  • Feature vector similarity calculation subunit configured to calculate the feature vector similarity between the matching feature vector and the corresponding reference face feature vector.
  • the gesture image recognition module 40 includes a gesture area image acquisition unit and a decision instruction information acquisition unit.
  • a gesture area image acquisition unit is configured to acquire a gesture image of a decision maker, and capture a gesture area image from the gesture image.
  • a decision instruction information acquisition unit is configured to input a gesture area image into a gesture image recognition model for recognition, and obtain decision instruction information of a decision maker.
  • the user instruction matching device further includes a gesture training image acquisition module and a gesture image recognition model acquisition module.
  • a gesture training image acquisition module is used to obtain original images, classify and label each original image, and form a gesture training image.
  • a gesture image recognition model acquisition module is used to train a convolutional neural network model with a gesture training image to obtain a gesture image recognition model.
  • the instruction matching result acquisition module 50 includes a participant label conversion unit, a decision maker label conversion unit, and an instruction matching result acquisition unit.
  • the participant label conversion unit is configured to convert the participant instruction information corresponding to each matching participant identifier into a participant label.
  • a decision maker label conversion unit is used to convert decision information of a decision maker into a decision maker label.
  • the instruction matching result obtaining unit is configured to match each participant label and decision maker label through a mapping relationship table to obtain an instruction matching result of each participant.
  • Each module in the above-mentioned user instruction matching device may be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the hardware form or independent of the processor in the computer device, or may be stored in the memory of the computer device in the form of software, so that the processor calls and performs the operations corresponding to the above modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 9.
  • the computer device includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, computer-readable instructions, and a database.
  • the internal memory provides an environment for operating the operating system and computer-readable instructions in a non-volatile storage medium.
  • the database of the computer device is used to store the current image of the participant and the instruction information of the participant, the gesture image and decision instruction information of the decision maker, and the gesture picture recognition model.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by a processor to implement a user instruction matching method.
  • a computer device including a memory, a processor, and computer-readable instructions stored on the memory and executable on the processor.
  • the processor executes the computer-readable instructions, the processor is implemented as in the foregoing embodiment User instruction matching method steps.
  • one or more non-volatile readable storage media storing computer-readable instructions are provided, and when the computer-readable instructions are executed by one or more processors, the one or more Each processor executes the steps of the user instruction matching method in the above embodiment.
  • Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Synchlink DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDRSDRAM dual data rate SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM synchronous chain Synchlink DRAM
  • Rambus direct RAM
  • DRAM direct memory bus dynamic RAM
  • RDRAM memory bus dynamic RAM

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Image Analysis (AREA)

Abstract

一种用户指令匹配方法、装置、计算机设备及存储介质，通过分别采集参与者的当前图像和决策者的手势图像的方式来获取数据，不需要每一参与者都通过一个终端来和服务端进行数据交互，避免了由于数据获取不完整而一直处于等待阶段，提高了用户指令匹配的效率。而且在分别获取到参与者的当前图像和决策者的手势图像之后，识别出对应的指令信息，并通过映射关系表进行匹配，得到每一参与者的指令匹配结果，也保证了用户指令匹配的效率。

Description

用户指令匹配方法、装置、计算机设备及存储介质
本申请以2018年07月27日提交的申请号为201810841703.5，名称为“用户指令匹配方法、装置、计算机设备及存储介质”的中国发明专利申请为基础，并要求其优先权。
技术领域
本申请涉及数据处理领域，尤其涉及一种用户指令匹配方法、装置、计算机设备及存储介质。
背景技术
传统的抽奖方法通过抽奖人员手动抽取，随着社会的发展，抽奖形式开始网络化、自动化，一般是将各个用户的抽奖信息收集到网络数据库端，网络数据库端经过一定的信息筛选将符合条件的信息发送至抽奖系统，例如根据发送时间，只选出抽奖信息的前一百条。进一步地，通过电脑抽奖系统，随机的从所述前一百条抽奖信息中选出若干名中奖者。
由于抽奖的过程需要满足透明性和公平性，如果网络数据库接收到的信息比较多，就会持续不断地发送信息，而频繁的信息发送导致系统的稳定性变差，或者服务器和移动终端在通信的过程中，因为网络的不稳定，都有可能造成信息的丢失。而如果由于指令不完整而一直处于等待阶段的话，又会使得整个指令匹配花费太多的时间，降低了整个指令匹配的效率。
发明内容
本申请实施例提供一种用户指令匹配方法、装置、计算机设备及存储介质，以解决用户指令匹配效率较低的问题。
一种用户指令匹配方法，包括：
获取每一参与者的当前图像，其中，所述当前图像包括人脸区域和手势区域；
将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识；
识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信息和对应的所述匹配参与者标识进行关联；
获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息；
将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
一种用户指令匹配装置，包括：
当前图像获取模块，用于获取每一参与者的当前图像，其中，所述当前图像包括人脸区域和手势区域；
匹配参与者标识获取模块，用于将每一所述当前图像的所述人脸区域在参与者图像库 中进行匹配，获取每一所述当前图像的匹配参与者标识；
参与者指令信息获取模块，用于识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信息和对应的所述匹配参与者标识进行关联；
手势图像识别模块，用于获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息；
指令匹配结果获取模块，用于将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
获取每一参与者的当前图像，其中，所述当前图像包括人脸区域和手势区域；
将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识；
识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信息和对应的所述匹配参与者标识进行关联；
获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息；
将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
一个或多个存储有计算机可读指令的非易失性可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
获取每一参与者的当前图像，其中，所述当前图像包括人脸区域和手势区域；
将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识；
识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信息和对应的所述匹配参与者标识进行关联；
获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息；
将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
本申请的一个或多个实施例的细节在下面的附图和描述中提出，本申请的其他特征和优点将从说明书、附图以及权利要求变得明显。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例的描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获 得其他的附图。
图1是本申请一实施例中用户指令匹配方法的一应用环境示意图；
图2是本申请一实施例中用户指令匹配方法的一示例图；
图3是本申请一实施例中用户指令匹配方法的另一示例图；
图4是本申请一实施例中用户指令匹配方法的另一示例图；
图5是本申请一实施例中用户指令匹配方法的另一示例图；
图6是本申请一实施例中用户指令匹配方法的另一示例图；
图7是本申请一实施例中用户指令匹配方法的另一示例图；
图8是本申请一实施例中用户指令匹配装置的一原理框图；
图9是本申请一实施例中计算机设备的一示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请提供的用户指令匹配方法，可应用在如图1的应用环境中，其中，客户端(计算机设备)通过网络与服务端进行通信。客户端采集参与者的当前图像和决策者的手势图像并发送到服务端，服务端基于获取到的当前图像和手势图像进行处理，得到每一参与者的指令匹配结果。其中，客户端(计算机设备)可以但不限于是摄像机、照相机、扫描仪或其他带有拍照功能的设备。服务端可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在一实施例中，如图2所示，提供一种用户指令匹配方法，以该方法应用在图1中的服务端为例进行说明，包括如下步骤：
S10：获取每一参与者的当前图像，其中，当前图像包括人脸区域和手势区域。
其中，参与者是指需要进行指令匹配的用户。当前图像是指客户端采集的参与者的图像，当前图像包括人脸区域和手势区域。即客户端采集参与者的图像，且采集到的参与者的图像是包括参与者的人脸区域和手势区域的。
每个参与者的当前图像可以通过一个客户端进行集中采集，也可以通过复数个客户端进行分别采集。而每一参与者的当前图像的采集时机是可以错开，也可以同时进行。在一个具体实施方式中，可以根据参与者的位置排布来设置预定数量的客户端，而且可以设置一个客户端采集一个参与者的当前图像，或者设置一个客户端采集预定数量的参与者的当前图像。可以理解地，不同客户端采集参与者的当前图像的时机可以是同时进行，也可以根据实际需要错开来采集，在此不做具体限定。
在一个具体应用场景中，设置复数个客户端，其中每个客户端采集预定数量的参与者的当前图像，每个客户端采集的参与者的数量可以根据客户端设置的位置来确定，可以设 置每个客户端采集该客户端的采集图像区域内的所有参与者的当前图像。而且设定该复数个客户端同时进行对参与者的当前图像的采集。通过该采集方式可以提高对当前图像的采集效率。
S20：将每一当前图像的人脸区域在参与者图像库中进行匹配，获取每一当前图像的匹配参与者标识。
其中，参与者图像库为预先存储有所有参与者人脸图像的图像库，而匹配参与者标识是用于确定当前图像中的人脸区域是和参与者图像库中哪副人脸图像匹配的，而且通过一个标识来体现。具体地，匹配参与者标识可以是参与者的员工号、身份证号、电话号码或其他唯一可识别参与者身份的信息。
具体地，先从当前图像中获取到对应的人脸区域，该人脸区域可以通过人脸特征点检测算法确定，通过人脸特征点检测算法识别出人脸区域，然后将人脸区域用矩形框标出。可选地，人脸特征点检测算法包括但不限于基于深度学习的人脸特征点检测算法、基于模型的人脸特征点检测算法或者基于级联形状回归的人脸特征点检测算法等。
在获取到人脸区域之后，将每一当前图像的人脸区域在参与者图像库中进行匹配，输出参与者图像库中匹配成功的人脸图像的标识，得到当前图像的匹配参与者标识。
S30：识别每一当前图像的手势区域，得到参与者指令信息，将参与者指令信息和对应的匹配参与者标识进行关联。
其中，参与者指令信息是指根据参与者的手势区域确定出的指令信息。优选地，参与者指令信息为手势指令信息，其中，手势指令信息即是代表不同手势动作的指令信息，例如：剪刀、石头或布，又或者是代表不同数字的手势动作，例如：1、2、3、4或5。
从每一当前图像中确定出手势区域，再对每一当前图像的手势区域进行识别，得到参与者指令信息。具体地，可以采用边缘检测算法获取当前图像的边缘图像，再根据边缘图像的曲率找到当前图像中的各个指关节点和两个手腕点。可选地，边缘检测算法可以采用差分边缘检测算法、Sobel边缘检测算法或者Reborts边缘检测算法等。在找到当前图像中的各个指关节点和两个手腕点之后，根据各个指关节点和两个手腕点来获取到手势区域。具体地，可以根据各个指关节点和两个手腕点来确定一个固定区域，该固定区域可以为一矩形或者圆形，该固定区域包括了手势区域。可选地，将该固定区域设置为一圆形，根据两个手腕点和其中的一个指关节点确定圆形的圆心，然后再根据手腕点和其中的一个指关节点的距离来乘以一定的比例来确定该圆形的半径，该比例可以根据实际需要来设定，在此不做具体限定。在获取到手势区域之后，可以使用imageclipper工具截取手势区域，imageclipper工具能快速的从手势图像中截取手势区域。
得到手势区域之后，将手势区域输入到预先训练好的手势图像识别模型中进行识别，输出参与者指令信息，例如：剪刀、石头或布。在得到参与者指令信息之后，将参与者指令信息和对应的匹配参与者标识进行关联，即标注每一匹配参与者标识对应的参与者指令信息。
S40：获取决策者的手势图像，并识别决策者的手势图像，得到决策者的决策指令信息。
其中，决策者是指提供手势图像，辅助进行参与者指令匹配的用户。可以通过一预定的客户端来采集决策者的手势图像，该预定的客户端可以为摄像机、照相机、扫描仪或其他带有拍照功能的设备。
当客户端获取到服务端发出的获取决策者的手势图像的指令之后，客户端采集决策者所在位置的图像，然后发送该决策者所在位置的图像至服务端。服务端获取到决策者所在位置的图像之后，可以通过截屏的方式获取决策者的手势图像。具体地，该决策者的手势图像的获取方式和步骤S30中手势区域的获取方式相同，在此不再赘述。
获取到决策者的手势图像之后，将手势图像输入到预先训练好的手势图像识别模型中进行识别，输出决策者的决策指令信息。可以理解地，决策指令信息是和参与者指令信息对应的。若参与者指令信息为手势指令信息，则决策指令信息也是手势指令信息，例如，若参与者指令信息代表的是剪刀、石头或布，则决策指令信息代表的也是剪刀、石头或布。
S50：将每一匹配参与者标识对应的参与者指令信息与决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
其中，映射关系表是一个预设的表格，通过该映射关系表可以根据参与者的参与者指令信息与决策者的决策指令信息进行匹配，并在映射关系表中查询到对应的指令匹配结果。
在该步骤中，通过将每一匹配参与者标识对应的参与者指令信息与决策者的决策指令信息通过映射关系表进行匹配，就可以得到每一参与者的指令匹配结果。
在本实施例中，首先获取每一参与者的当前图像，其中，当前图像包括人脸区域和手势区域；将每一当前图像的人脸区域在参与者图像库中进行匹配，获取每一当前图像的匹配参与者标识；识别每一当前图像的手势区域，得到参与者指令信息，将参与者指令信息和对应的匹配参与者标识进行关联；获取决策者的手势图像，并识别决策者的手势图像，得到决策者的决策指令信息；将每一匹配参与者标识对应的参与者指令信息与决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。通过分别采集参与者的当前图像和决策者的手势图像的方式来获取数据，不需要每一参与者都通过一个终端来和服务端进行数据交互，避免了由于数据获取不完整而一直处于等待阶段，提高了用户指令匹配的效率。而且在分别获取到参与者的当前图像和决策者的手势图像之后，识别出对应的指令信息，并通过映射关系表进行匹配，得到每一参与者的指令匹配结果，也保证了用户指令匹配的效率。
在一实施例中，参与者图像库包括基准参与者标识、基准位置标识和基准人脸图像。基准人脸图像为预先采集的参与者的人脸图像，在采集基准人脸图像后为每一参与者的基准人脸图像分配对应的基准参与者标识和基准位置标识。
其中，基准人脸图像用于后续和每一当前图像进行匹配。可选地，可以采用参与者证 件照片或者工作证件上的照片作为基准人脸图像。基准参与者标识是用于确定每一基准人脸图像是属于哪一参与者的一个标识，基准参与者标识可以是参与者的员工号、身份证号、电话号码或其他唯一可识别参与者身份的信息。基准位置标识是指预先为每一参与者分配的位置的一个标识，基准位置标识可以用数字、字母或其他计算机可识别的符号表示。
在一个具体实施方式中，将本实施例应用在一个会议或者活动现场，在每一参与者入场时，设置一客户端，用于采集每一参与者的基准人脸图像，可以在该客户端采集基准人脸图像的时候，在客户端将基准人脸图像发送到服务端时，服务端按照采集参与者的基准人脸图像的顺序给每一参与者编号，并将这个编号作为基准参与者标识发送至客户端，也可以通过该客户端获取参与者身份信息，参与者的身份信息可以是参与者的员工号、身份证号、电话号码或其他唯一可识别参与者身份的信息。优选地，在唯一可识别参与者身份的信息中加入参与者姓名，可以更好地辨识参与者的身份。将参与者身份信息作为基准参与者标识，客户端将基准参与者标识发送到服务端。通过客户端采集每个参与者的基准人脸图像，服务端每获取到一幅参与者的基准人脸图像，自动为该参与者分配位置，并记录该与位置对应的基准位置标识，服务端记录每一参与者的基准参与者标识、基准位置标识和基准人脸图像，形成参与者图像库。
在本实施例中，通过预先建立参与者图像库，保证了获取到每一参与者的当前图像之后，可以通过该参与者图像库快速地实现人脸图像的匹配。
在一实施例中，当前图像包括当前位置标识。其中，当前位置标识是指参与者在采集当前图像时所处的位置的标识。具体地，可以根据采集该当前图像的客户端的位置来定位出当前位置标识。
在本实施例中，如图3所示，将每一当前图像的人脸区域在参与者图像库中进行匹配，获取每一当前图像的匹配参与者标识，具体包括以下步骤：
S21：根据每一当前图像的当前位置标识在参与者图像库中查询对应的基准位置标识，并将当前图像的人脸区域与对应的基准位置标识的基准人脸图像进行匹配。
根据每一当前图像的当前位置标识在参与者图像库中查询对应的基准位置标识，其中，对应的基准位置标识是指基准位置标识和当前位置标识相同。例如，若当前位置标识为007，则在参与者图像库中也为007的基准位置标识是和该当前位置标识对应的。在获取到与当前位置标识对应的基准位置标识之后，将当前图像的人脸区域和该基准位置标识的基准人脸图像进行匹配。具体地，可以分别将当前图像的人脸区域和该基准位置标识的基准人脸图像转化为特征向量，再通过计算两个特征向量的相似度来判断是否匹配。可选地，可以设置一相似度阈值，然后根据计算出来的相似度和该相似度阈值进行比较，得到匹配成功或者匹配失败的结果。
S22：将匹配成功的参与者图像库的基准人脸图像的基准参与者标识作为对应的当前图像的匹配参与者标识。
在步骤S21中对两幅图像进行匹配之后，若匹配成功，则将匹配成功的参与者图像库 的基准人脸图像的基准参与者标识作为对应的当前图像的匹配参与者标识。
S23：获取所有匹配失败的基准位置标识对应的基准人脸图像和对应的基准参与者标识，作为基准人脸图像库，将每一匹配失败的当前图像的人脸区域和基准人脸图像库中的每一基准人脸图像进行匹配。
在该步骤中，将所有匹配失败的基准位置标识对应的基准人脸图像和对应的基准参与者标识，作为基准人脸图像库。并将每一匹配失败的当前图像的人脸区域和基准人脸图像库中的每一基准人脸图像进行匹配。具体地，将每一当前图像的人脸区域转化的特征向量，和基准人脸图像库中每一基准人脸图像的特征向量计算特征向量相似度，特征向量相似度数值最高的基准人脸图像即与对应的当前图像匹配成功。
S24：将匹配成功的基准人脸图像库的基准人脸图像的基准参与者标识作为对应的当前图像的匹配参与者标识。
在步骤S23中进行匹配之后，将匹配成功的基准人脸图像库的基准人脸图像的基准参与者标识作为对应的当前图像的匹配参与者标识。
在本实施例中，首先根据当前位置标识和基准位置标识来查询到对应的人脸图像进行匹配，避免了每一当前图像在匹配时都需要和参与者图像库中的所有基准人脸图像进行匹配，提高了人脸匹配的效率。在出现匹配失败的情况之后再通过一一比对的方式来和剩余的基准人脸图像进行匹配，也保证了当前图像匹配的完整性。
在一实施例中，如图4所示，将当前图像的所述人脸区域与对应的基准位置标识的基准人脸图像进行匹配，具体包括以下步骤：
S221：将当前图像的人脸区域转化成匹配特征向量。
其中，匹配特征向量是指当前图像的人脸区域的特征向量，是用于表征当前图像的人脸区域的图像信息特征的向量，例如：基于投影的特征向量(如PCA(Principal Component Analysis，主成分分析)特征向量)、基于方向的特征向量(如HOG(Histogram of Oriented Gradient，梯度方向直方图)特征向量)和基于深度学习的特征向量(如卷积神经网络特征向量)等。特征向量能够以简单的数据表征图像信息，通过提取人脸图像的特征向量可以简化后续的比对过程。
优选地，本实施例中匹配特征向量可以为基于深度学习的特征向量。采用深度卷积神经网络进行特征提取，由于深度学习能够自动从人脸图像的数据中学习，因此能够适用多种环境，并且省去了复杂的预处理操作，而基于投影、方向和重心的特征向量往往只能提取一种特征如颜色特征或形状特征等，这些特征很难应用到现实复杂环境。因此，匹配特征向量为基于深度学习的特征向量能够提高后续人脸匹配的准确率。
S222：获取基准人脸特征向量，其中基准人脸特征向量是将对应的基准位置标识的基准人脸图像进行特征向量转化得到的。
其中，基准人脸特征向量是基准人脸图像的特征向量，用于表征基准人脸图像的图像信息特征的向量。该基准人脸特征向量的特征向量转化方式和步骤S221中匹配特征向量 的特征向量转化方式相同，在此不再赘述。
S223：计算匹配特征向量和对应的基准人脸特征向量的特征向量相似度。
在获取到匹配特征向量和对应的基准人脸特征向量之后，计算两者之间的特征向量相似度。可选地，特征向量相似度可以通过欧几里得距离算法、曼哈顿距离算法、明可夫斯基距离算法或者余弦相似度算法来计算。
具体地，可以采用欧几里得距离算法计算匹配特征向量和基准人脸特征向量之间的特征向量相似度：
Figure PCTCN2018106432-appb-000001
特征向量，而x i为匹配特征向量中的向量元素，而y i为基准人脸特征向量中的向量元素，i为正整数，且0＜i≤n。特征向量相似度sim(U,V) EDG越大说明两个向量距离越相近，则对应的两幅图像之间的相似程度越高。
通过上述公式，计算匹配特征向量和基准人脸特征向量的特征向量相似度。
在一个具体实施方式中，在计算得到匹配特征向量和基准人脸特征向量的特征向量相似度之后，通过比较特征向量相似度和预设相似度阈值之间的大小来判断是否匹配成功。若特征向量相似度大于或者等于预设相似度阈值，则匹配成功。若若特征向量相似度小于预设相似度阈值，则匹配失败。
在本实施例中，通过将当前图像的人脸区域和与对应的基准位置标识的基准人脸图像对应的特征向量进行计算特征向量相似度，保证了图像匹配的精度和效率。
在一实施例中，如图5所示，获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息，具体包括以下步骤：
S41：获取决策者的手势图像，从手势图像中截取手势区域图像。
获取决策者的手势图像，由于获取的是决策者的手势图像，该手势图像可能囊括的区域比较大，不利于后续的识别处理。因此，需要将手势区域从手势图像中截取出来，得到手势区域图像。具体地，可以采用边缘检测算法获取手势图像的边缘图像，再根据边缘图像的曲率找到手势图像中的各个指关节点和两个手腕点。可选地，边缘检测算法可以采用差分边缘检测算法、Sobel边缘检测算法或者Reborts边缘检测算法等。在找到手势图像中的各个指关节点和两个手腕点之后，根据各个指关节点和两个手腕点来获取到手势区域，最后根据该手势区域得到手势区域图像。具体地，可以根据各个指关节点和两个手腕点来确定一个固定区域，该固定区域可以为一矩形或者圆形，该固定区域包括了手势区域。可选地，将该固定区域设置为一圆形，根据两个手腕点和其中的一个指关节点确定圆形的圆心，然后再根据手腕点和其中的一个指关节点的距离来乘以一定的比例来确定该圆形的半径，该比例可以根据实际需要来设定，在此不做具体限定。在获取到手势区域之后，可以 使用imageclipper工具截取手势区域图像，imageclipper工具能快速的从手势图像中截取手势区域图像。
S42：将手势区域图像输入手势图像识别模型进行识别，得到决策者的决策指令信息。
将手势区域图像输入到一个手势图像识别模型中，就可以通过该手势识别模型的输出结果获取到与手势区域图像对应的决策指令信息。其中，手势图像识别模型是预先训练好的一个识别模型，通过该手势图像识别模型，可以快速地对手势区域图像的类型进行识别，输出决策者的决策指令信息，以利于后续的匹配。
在本实施例中，通过截取包含手势区域的手势区域图像，再将手势区域图像输入手势图像识别模型中识别出当前手势类型，并据此得到决策者的决策指令信息，提高了手势识别的准确性和效率。
在一实施例中，如图6所示，在识别每一当前图像的所述手势区域，得到参与者指令信息，将参与者指令信息和对应的匹配参与者标识进行关联的步骤之前，该用户指令匹配方法还包括以下步骤：
S421：获取原始图像，为每一原始图像进行分类标注，组成手势训练图像。
其中，原始图像是预先采集的包含不同手势的图像。不同手势的图像可以通过摄像头获取得到，也可以通过网络上现有的数据集来获取。通过获取大量不同手势的原始图像，并为每一原始图像进行分类标注，例如：“石头”、“剪刀”或“布”。对原始图像进行分类标注完成之后，使得每一原始图像都包含了对应的标注数据，并将包含了标注数据的原始图像作为手势训练图像。
S422：采用手势训练图像对卷积神经网络模型进行训练，获得手势图像识别模型。
通过将手势训练图像输入到卷积神经网络模型中进行训练，即可得到手势图像识别模型。其中，卷积神经网络(Convolutional Neural Network,CNN)模型，是一种前馈神经网络，它的人工神经元可以响应一部分覆盖范围内的周围单元，常应用于大型图像的处理。卷积神经网络通常包括至少两个非线性可训练的卷积层，至少两个非线性的池化层和至少一个全连接层，即包括至少五个隐含层，此外还包括输入层和输出层。通过手势训练图像对卷积神经网络模型进行训练而获得的手势图像识别模型，可以更为准确地对手势训练图像进行分类。
通过获取不同手势的原始图像，并对每一手势的原始图像进行分类标注，获取当前手势的原始训练图像，再采用手势训练图像对卷积神经网络模型进行训练，得到手势图像识别模型。将当前手势图像输入到手势图像识别模型中进行识别，获取当前手势图像对应的当前手势类型。
在本实施例中，通过获取原始图像，每一原始图像进行分类标注，组成手势训练图像。并采用手势训练图像对卷积神经网络模型进行训练，获得手势图像识别模型，更好地保证了后续手势区域图像的识别效率和精度。
在一个实施例中，如图7所示，将每一匹配参与者标识对应的参与者指令信息与决策 者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果，具体包括以下步骤：
S51：将每一匹配参与者标识对应的参与者指令信息转化为参与者标签。
其中，参与者标签是指将参与者的参与者指令信息进行转化得到的数值化或符号化的信息。例如，参与者的参与者指令信息包括石头、剪刀和布，则对应地将这三个指令信息转化为数值化或符号化的信息，以方便后续的处理，例如：将“石头”指令映射为“1”，“剪刀”指令映射为“2”，“布”指令映射为“3”。
S52：将决策者的决策指令信息转化为决策者标签。
其中，决策者标签是指将决策者的决策指令信息进行转化得到的数值化或符号化的信息。可以理解地，决策者的决策指令信息和参与者的目标指令信息或者随机指令信息是对应的，例如：石头、剪刀和布。具体转化方式也是和步骤S51相同的，这里不再赘述。
S53：将每一参与者标签和决策者标签通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
在得到参与者标签和决策者标签之后，通过映射关系表进行匹配，就可以得到每一参与者的指令匹配结果。具体地，可以预先设置一参与者标签和决策者标签的映射关系表，根据不同的参与者标签和不同的决策者标签就可以在映射关系表中查询到对应的指令匹配结果。例如：若指令匹配结果包括赢、平和负三种情况，则可以分别采用X、Y和Z代表这三种情况。
在这个实施方式中，通过将对应的指令信息转化为标签，然后通过预设的映射关系表获取到每一参与者的指令匹配结果，提高了指令匹配结果获取的效率。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
在一实施例中，提供一种用户指令匹配装置，该用户指令匹配装置与上述实施例中用户指令匹配方法一一对应。如图8所示，该用户指令匹配装置包括当前图像获取模块10、匹配参与者标识获取模块20、参与者指令信息获取模块30、手势图像识别模块40和指令匹配结果获取模块50。各功能模块详细说明如下：
当前图像获取模块10，用于获取每一参与者的当前图像，其中，当前图像包括人脸区域和手势区域。
匹配参与者标识获取模块20，用于将每一当前图像的人脸区域在参与者图像库中进行匹配，获取每一当前图像的匹配参与者标识。
参与者指令信息获取模块30，用于识别每一当前图像的手势区域，得到参与者指令信息，将参与者指令信息和对应的匹配参与者标识进行关联。
手势图像识别模块40，用于获取决策者的手势图像，并识别决策者的手势图像，得到决策者的决策指令信息。
指令匹配结果获取模块50，用于将每一匹配参与者标识对应的参与者指令信息与决策 者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
优选地，匹配参与者标识获取模块20包括第一匹配单元、第一匹配参与者标识获取单元、第二匹配单元和第二匹配参与者标识获取单元。
第一匹配单元，用于根据每一当前图像的当前位置标识在参与者图像库中查询对应的基准位置标识，并将当前图像的人脸区域与对应的基准位置标识的基准人脸图像进行匹配。
第一匹配参与者标识获取单元，用于将匹配成功的参与者图像库的基准人脸图像的基准参与者标识作为对应的当前图像的匹配参与者标识。
第二匹配单元，用于获取所有匹配失败的基准位置标识对应的基准人脸图像和对应的基准参与者标识，作为基准人脸图像库，将每一匹配失败的当前图像的人脸区域和基准人脸图像库中的每一基准人脸图像进行匹配。
第二匹配参与者标识获取单元，用于将匹配成功的基准人脸图像库的基准人脸图像的基准参与者标识作为对应的当前图像的匹配参与者标识。
优选地，第一匹配参与者标识获取单元包括匹配特征向量转化子单元、基准人脸特征向量获取子单元、特征向量相似度计算子单元。
匹配特征向量转化子单元，用于将当前图像的人脸区域转化成匹配特征向量。
基准人脸特征向量获取子单元，用于获取基准人脸特征向量，其中基准人脸特征向量是将对应的基准位置标识的基准人脸图像进行特征向量转化得到的。
特征向量相似度计算子单元，用于计算匹配特征向量和对应的基准人脸特征向量的特征向量相似度。
优选地，手势图像识别模块40包括手势区域图像获取单元和决策指令信息获取单元。
手势区域图像获取单元，用于获取决策者的手势图像，从手势图像中截取手势区域图像。
决策指令信息获取单元，用于将手势区域图像输入手势图像识别模型进行识别，得到决策者的决策指令信息。
优选地，该用户指令匹配装置还包括手势训练图像获取模块和手势图像识别模型获取模块。
手势训练图像获取模块，用于获取原始图像，为每一原始图像进行分类标注，组成手势训练图像。
手势图像识别模型获取模块，用于采用手势训练图像对卷积神经网络模型进行训练，获得手势图像识别模型。
优选地，指令匹配结果获取模块50包括参与者标签转化单元、决策者标签转化单元和指令匹配结果获取单元。
参与者标签转化单元，用于将每一匹配参与者标识对应的参与者指令信息转化为参与者标签。
决策者标签转化单元，用于将决策者的决策指令信息转化为决策者标签。
指令匹配结果获取单元，用于将每一参与者标签和决策者标签通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
关于用户指令匹配装置的具体限定可以参见上文中对于用户指令匹配方法的限定，在此不再赘述。上述用户指令匹配装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图9所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储参与者的当前图像和参与者指令信息，决策者的手势图像和决策指令信息，以及手势图片识别模型。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种用户指令匹配方法。
在一个实施例中，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机可读指令，处理器执行计算机可读指令时实现如上述实施例中用户指令匹配方法的步骤。
在一个实施例中，提供了一个或多个存储有计算机可读指令的非易失性可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如上述实施例中用户指令匹配方法的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种用户指令匹配方法，其特征在于，包括：
    获取每一参与者的当前图像，其中，所述当前图像包括人脸区域和手势区域；
    将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识；
    识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信息和对应的所述匹配参与者标识进行关联；
    获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息；
    将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
  2. 如权利要求1所述的用户指令匹配方法，其特征在于，所述参与者图像库包括基准参与者标识、基准位置标识和基准人脸图像；
    其中，所述基准人脸图像为预先采集的参与者的人脸图像，在采集所述基准人脸图像后为每一参与者的所述基准人脸图像分配对应的基准参与者标识和基准位置标识。
  3. 如权利要求2所述的用户指令匹配方法，其特征在于，所述当前图像包括当前位置标识；
    所述将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识，包括：
    根据每一当前图像的所述当前位置标识在参与者图像库中查询对应的基准位置标识，并将所述当前图像的所述人脸区域与对应的基准位置标识的基准人脸图像进行匹配；
    将匹配成功的所述参与者图像库的基准人脸图像的基准参与者标识作为对应的所述当前图像的匹配参与者标识；
    获取所有匹配失败的基准位置标识对应的基准人脸图像和对应的所述基准参与者标识，作为基准人脸图像库，将每一匹配失败的当前图像的人脸区域和所述基准人脸图像库中的每一基准人脸图像进行匹配；
    将匹配成功的所述基准人脸图像库的基准人脸图像的基准参与者标识作为对应的所述当前图像的匹配参与者标识。
  4. 如权利要求3所述的用户指令匹配方法，其特征在于，将所述当前图像的所述人脸区域与对应的基准位置标识的基准人脸图像进行匹配，包括：
    将所述当前图像的所述人脸区域转化成匹配特征向量；
    获取基准人脸特征向量，其中所述基准人脸特征向量是将对应的基准位置标识的基准人脸图像进行特征向量转化得到的；
    计算所述匹配特征向量和对应的所述基准人脸特征向量的特征向量相似度。
  5. 如权利要求1所述的用户指令匹配方法，其特征在于，所述获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息，包括：
    获取决策者的手势图像，从所述手势图像中截取手势区域图像；
    将所述手势区域图像输入手势图像识别模型进行识别，得到所述决策者的决策指令信息。
  6. 如权利要求1所述的用户指令匹配方法，其特征在于，在所述识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信息和对应的所述匹配参与者标识进行关联的步骤之前，所述用户指令匹配方法还包括：
    获取原始图像，为每一所述原始图像进行分类标注，组成手势训练图像；
    采用所述手势训练图像对卷积神经网络模型进行训练，获得手势图像识别模型。
  7. 如权利要求1所述的用户指令匹配方法，其特征在于，所述将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果，包括：
    将每一所述匹配参与者标识对应的参与者指令信息转化为参与者标签；
    将所述决策者的决策指令信息转化为决策者标签；
    将每一所述参与者标签和所述决策者标签通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
  8. 一种用户指令匹配装置，其特征在于，包括：
    当前图像获取模块，用于获取每一参与者的当前图像，其中，所述当前图像包括人脸区域和手势区域；
    匹配参与者标识获取模块，用于将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识；
    参与者指令信息获取模块，用于识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信息和对应的所述匹配参与者标识进行关联；
    手势图像识别模块，用于获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息；
    指令匹配结果获取模块，用于将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
  9. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，其特征在于，所述处理器执行所述计算机可读指令时实现如下步骤：
    获取每一参与者的当前图像，其中，所述当前图像包括人脸区域和手势区域；
    将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识；
    识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信 息和对应的所述匹配参与者标识进行关联；
    获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息；
    将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
  10. 如权利要求9所述的计算机设备，其特征在于，所述参与者图像库包括基准参与者标识、基准位置标识和基准人脸图像；
    其中，所述基准人脸图像为预先采集的参与者的人脸图像，在采集所述基准人脸图像后为每一参与者的所述基准人脸图像分配对应的基准参与者标识和基准位置标识。
  11. 如权利要求10所述的计算机设备，其特征在于，所述当前图像包括当前位置标识；
    所述将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识，包括：
    根据每一当前图像的所述当前位置标识在参与者图像库中查询对应的基准位置标识，并将所述当前图像的所述人脸区域与对应的基准位置标识的基准人脸图像进行匹配；
    将匹配成功的所述参与者图像库的基准人脸图像的基准参与者标识作为对应的所述当前图像的匹配参与者标识；
    获取所有匹配失败的基准位置标识对应的基准人脸图像和对应的所述基准参与者标识，作为基准人脸图像库，将每一匹配失败的当前图像的人脸区域和所述基准人脸图像库中的每一基准人脸图像进行匹配；
    将匹配成功的所述基准人脸图像库的基准人脸图像的基准参与者标识作为对应的所述当前图像的匹配参与者标识。
  12. 如权利要求11所述的计算机设备，其特征在于，将所述当前图像的所述人脸区域与对应的基准位置标识的基准人脸图像进行匹配，包括：
    将所述当前图像的所述人脸区域转化成匹配特征向量；
    获取基准人脸特征向量，其中所述基准人脸特征向量是将对应的基准位置标识的基准人脸图像进行特征向量转化得到的；
    计算所述匹配特征向量和对应的所述基准人脸特征向量的特征向量相似度。
  13. 如权利要求9所述的计算机设备，其特征在于，所述获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息，包括：
    获取决策者的手势图像，从所述手势图像中截取手势区域图像；
    将所述手势区域图像输入手势图像识别模型进行识别，得到所述决策者的决策指令信息。
  14. 如权利要求9所述的计算机设备，其特征在于，所述将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果，包括：
    将每一所述匹配参与者标识对应的参与者指令信息转化为参与者标签；
    将所述决策者的决策指令信息转化为决策者标签；
    将每一所述参与者标签和所述决策者标签通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
  15. 一个或多个存储有计算机可读指令的非易失性可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
    获取每一参与者的当前图像，其中，所述当前图像包括人脸区域和手势区域；
    将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识；
    识别每一所述当前图像的所述手势区域，得到参与者指令信息，将所述参与者指令信息和对应的所述匹配参与者标识进行关联；
    获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息；
    将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
  16. 如权利要求15所述的非易失性可读存储介质，其特征在于，所述参与者图像库包括基准参与者标识、基准位置标识和基准人脸图像；
    其中，所述基准人脸图像为预先采集的参与者的人脸图像，在采集所述基准人脸图像后为每一参与者的所述基准人脸图像分配对应的基准参与者标识和基准位置标识。
  17. 如权利要求16所述的非易失性可读存储介质，其特征在于，所述当前图像包括当前位置标识；
    所述将每一所述当前图像的所述人脸区域在参与者图像库中进行匹配，获取每一所述当前图像的匹配参与者标识，包括：
    根据每一当前图像的所述当前位置标识在参与者图像库中查询对应的基准位置标识，并将所述当前图像的所述人脸区域与对应的基准位置标识的基准人脸图像进行匹配；
    将匹配成功的所述参与者图像库的基准人脸图像的基准参与者标识作为对应的所述当前图像的匹配参与者标识；
    获取所有匹配失败的基准位置标识对应的基准人脸图像和对应的所述基准参与者标识，作为基准人脸图像库，将每一匹配失败的当前图像的人脸区域和所述基准人脸图像库中的每一基准人脸图像进行匹配；
    将匹配成功的所述基准人脸图像库的基准人脸图像的基准参与者标识作为对应的所述当前图像的匹配参与者标识。
  18. 如权利要求16所述的非易失性可读存储介质，其特征在于，将所述当前图像的所述人脸区域与对应的基准位置标识的基准人脸图像进行匹配，包括：
    将所述当前图像的所述人脸区域转化成匹配特征向量；
    获取基准人脸特征向量，其中所述基准人脸特征向量是将对应的基准位置标识的基准人脸图像进行特征向量转化得到的；
    计算所述匹配特征向量和对应的所述基准人脸特征向量的特征向量相似度。
  19. 如权利要求15所述的非易失性可读存储介质，其特征在于，所述获取决策者的手势图像，并识别所述决策者的手势图像，得到所述决策者的决策指令信息，包括：
    获取决策者的手势图像，从所述手势图像中截取手势区域图像；
    将所述手势区域图像输入手势图像识别模型进行识别，得到所述决策者的决策指令信息。
  20. 如权利要求15所述的非易失性可读存储介质，其特征在于，所述将每一所述匹配参与者标识对应的参与者指令信息与所述决策者的决策指令信息通过映射关系表进行匹配，得到每一参与者的指令匹配结果，包括：
    将每一所述匹配参与者标识对应的参与者指令信息转化为参与者标签；
    将所述决策者的决策指令信息转化为决策者标签；
    将每一所述参与者标签和所述决策者标签通过映射关系表进行匹配，得到每一参与者的指令匹配结果。
PCT/CN2018/106432 2018-07-27 2018-09-19 用户指令匹配方法、装置、计算机设备及存储介质 WO2020019457A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810841703.5 2018-07-27
CN201810841703.5A CN109214278B (zh) 2018-07-27 2018-07-27 用户指令匹配方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020019457A1 true WO2020019457A1 (zh) 2020-01-30

Family

ID=64990273

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/106432 WO2020019457A1 (zh) 2018-07-27 2018-09-19 用户指令匹配方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN109214278B (zh)
WO (1) WO2020019457A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111738782A (zh) * 2020-05-29 2020-10-02 北京立思辰新技术有限公司 一种基于图像双特征识别的电子凭据信息录入方法和系统
CN112488650A (zh) * 2020-11-26 2021-03-12 万翼科技有限公司 会议气氛调节方法、电子设备及相关产品

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113448429A (zh) * 2020-03-25 2021-09-28 南京人工智能高等研究院有限公司 基于手势控制电子设备的方法及装置、存储介质和电子设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103201710A (zh) * 2010-11-10 2013-07-10 日本电气株式会社 图像处理系统、图像处理方法以及存储图像处理程序的存储介质
US20150235079A1 (en) * 2011-05-23 2015-08-20 Sony Corporation Learning device, learning method, and program
CN105468249A (zh) * 2014-09-09 2016-04-06 联胜（中国）科技有限公司 智能互动系统及其控制方法
CN106407436A (zh) * 2016-09-27 2017-02-15 维沃移动通信有限公司 一种通信账号头像处理方法及移动终端

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5605725B2 (ja) * 2010-11-10 2014-10-15 日本電気株式会社 情報報知システム、情報報知方法、情報処理装置及びその制御方法と制御プログラム
CN108229277B (zh) * 2017-03-31 2020-05-01 北京市商汤科技开发有限公司 手势识别、手势控制及多层神经网络训练方法、装置及电子设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103201710A (zh) * 2010-11-10 2013-07-10 日本电气株式会社 图像处理系统、图像处理方法以及存储图像处理程序的存储介质
US20150235079A1 (en) * 2011-05-23 2015-08-20 Sony Corporation Learning device, learning method, and program
CN105468249A (zh) * 2014-09-09 2016-04-06 联胜（中国）科技有限公司 智能互动系统及其控制方法
CN106407436A (zh) * 2016-09-27 2017-02-15 维沃移动通信有限公司 一种通信账号头像处理方法及移动终端

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111738782A (zh) * 2020-05-29 2020-10-02 北京立思辰新技术有限公司 一种基于图像双特征识别的电子凭据信息录入方法和系统
CN112488650A (zh) * 2020-11-26 2021-03-12 万翼科技有限公司 会议气氛调节方法、电子设备及相关产品

Also Published As

Publication number Publication date
CN109214278B (zh) 2023-04-18
CN109214278A (zh) 2019-01-15

Similar Documents

Publication Publication Date Title
WO2017088432A1 (zh) 图像识别方法和装置
WO2019033572A1 (zh) 人脸遮挡检测方法、装置及存储介质
WO2019033571A1 (zh) 面部特征点检测方法、装置及存储介质
WO2019033525A1 (zh) Au特征识别方法、装置及存储介质
WO2019033569A1 (zh) 眼球动作分析方法、装置及存储介质
WO2014036350A2 (en) Facilitating photo sharing
WO2020019457A1 (zh) 用户指令匹配方法、装置、计算机设备及存储介质
Vazquez-Fernandez et al. Built-in face recognition for smart photo sharing in mobile devices
WO2019033570A1 (zh) 嘴唇动作分析方法、装置及存储介质
CN110852704B (zh) 基于密集微小人脸识别的考勤方法、系统、设备及介质
JP2022133378A (ja) 顔生体検出方法、装置、電子機器、及び記憶媒体
WO2020019451A1 (zh) 人脸识别方法、装置、计算机设备及存储介质
WO2019033567A1 (zh) 眼球动作捕捉方法、装置及存储介质
CN113298158B (zh) 数据检测方法、装置、设备及存储介质
WO2021082045A1 (zh) 微笑表情检测方法、装置、计算机设备及存储介质
WO2020172870A1 (zh) 一种目标对象的移动轨迹确定方法和装置
Jha et al. An Automated Attendance System Using Facial Detection and Recognition Technology
Baytamouny et al. AI-based home security system with face recognition
TW201828156A (zh) 圖像識別方法、度量學習方法、圖像來源識別方法及裝置
Gupta et al. HaarCascade and LBPH Algorithms in Face Recognition Analysis
US10621419B2 (en) Method and system for increasing biometric acceptance rates and reducing false accept rates and false rates
Kesharwani et al. Automated Attendance System Using Computer Vision
Su et al. Complete Region of Interest for Unconstrained Palmprint Recognition
Vivek et al. A Way to Mark Attentance using Face Recognition using PL
US20140111431A1 (en) Optimizing photos

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: 18927250

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: 18927250

Country of ref document: EP

Kind code of ref document: A1