WO2020140798A1 - 手势识别方法、装置、电子设备及存储介质 - Google Patents

手势识别方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2020140798A1
WO2020140798A1 PCT/CN2019/128029 CN2019128029W WO2020140798A1 WO 2020140798 A1 WO2020140798 A1 WO 2020140798A1 CN 2019128029 W CN2019128029 W CN 2019128029W WO 2020140798 A1 WO2020140798 A1 WO 2020140798A1
Authority
WO
WIPO (PCT)
Prior art keywords
hand
coordinate
feature
coordinates
depth
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/128029
Other languages
English (en)
French (fr)
Inventor
刘裕峰
郑文
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Dajia Internet Information Technology Co Ltd
Original Assignee
Beijing Dajia Internet Information Technology Co Ltd
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 Beijing Dajia Internet Information Technology Co Ltd filed Critical Beijing Dajia Internet Information Technology Co Ltd
Publication of WO2020140798A1 publication Critical patent/WO2020140798A1/zh
Priority to US17/139,369 priority Critical patent/US11163373B2/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/017—Gesture based interaction, e.g. based on a set of recognized hand gestures
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G06T7/74—Determining position or orientation of objects or cameras using feature-based methods involving reference images or patches
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/64—Three-dimensional [3D] objects
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/20—Movements or behaviour, e.g. gesture recognition
    • G06V40/28—Recognition of hand or arm movements, e.g. recognition of deaf sign language
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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

Definitions

  • This application relates to the field of computer technology, and in particular, to a gesture recognition method, device, electronic device, and storage medium.
  • Gesture recognition is the estimation of human hand pose, which refers to predicting the three-dimensional coordinates of the key points of the human hand in the image containing the human hand.
  • the existing human hand pose estimation technology due to depth ambiguity (multiple three-dimensional coordinates can be mapped to the same two-dimensional coordinate) and the flexible and changeable hand pose, the self-occlusion phenomenon is very serious, which makes the three-dimensional directly on the image. It is very difficult to label key points, so the real-time human data set often does not have three-dimensional coordinate labeling.
  • the image of the human hand synthesized using the renderer is often sufficient and well-marked, but the gap between the synthesized hand and the real scene is relatively large on the skin and background, so only the gestures trained based on the synthetic data set
  • the recognition model has poor generalization ability.
  • weakly supervised human pose estimation methods that is, to use synthetic data and real data to jointly train to improve the accuracy of pose estimation in real scenes.
  • Existing weak supervision methods will learn the two-dimensional to three-dimensional mapping relationship through synthetic data and apply this relationship to real scenes.
  • use the easily obtained depth map to constrain the depth prediction range and reduce the impact of depth ambiguity.
  • the inventor realized that since the synthesized data and the real data have a large difference, only the depth map constraint is not enough to eliminate this difference, and thus the accuracy of the three-dimensional pose estimation is also limited.
  • the present application provides a gesture recognition method, device, electronic device, and storage medium.
  • a gesture recognition method including:
  • a feature mapping model is used to extract the first standard feature of the hand in the hand image; the feature mapping model is pre-trained according to the second standard feature of the hand in the synthetic image sample and the real image sample;
  • the posture of the hand is determined according to the three-dimensional coordinates of each of the key points.
  • a gesture recognition device comprising:
  • the first acquisition module is configured to acquire the hand image
  • the feature extraction module is configured to use a feature mapping model to extract the first standard feature of the hand in the hand image; the feature mapping model is based on the second standard feature of the hand in the composite image sample and the real image sample in advance Obtained by training;
  • a coordinate acquisition module configured to process the first standard feature and obtain three-dimensional coordinates of multiple key points of the hand
  • the gesture determination module is configured to determine the posture of the hand according to the three-dimensional coordinates of each of the key points.
  • an electronic device comprising:
  • Memory for storing processor executable instructions
  • the processor is configured to execute the gesture recognition method as described in the first aspect.
  • a non-transitory computer-readable storage medium when instructions in the storage medium are executed by a processor of an electronic device, enabling the electronic device to perform the gesture as described in the first aspect recognition methods.
  • a computer program product which, when instructions in the computer program product are executed by a processor of an electronic device, enables the electronic device to perform the gesture recognition method as described in the first aspect.
  • the gesture recognition method provided in this application uses the common features of the hands in the composite image and the real image for model training, eliminating the difference between the composite image and the real image at the standard feature level, which can improve gesture recognition in the case of weak supervision
  • the accuracy of the method and the method does not require synthetic image samples and real image samples with completely identical gestures, which relaxes the requirements for training data sets; the method is also suitable for gesture recognition of color images, and the application is more extensive.
  • FIG. 3 is a flowchart of steps for obtaining three-dimensional coordinates of key points shown in this application;
  • FIG. 6 is a block diagram of a gesture recognition device shown in this application.
  • FIG. 7 is a block diagram of the first electronic device shown in this application.
  • FIG. 8 is a block diagram of a second type of electronic device shown in this application.
  • FIG. 1 is a flowchart of a gesture recognition method shown in this application. As shown in FIG. 1, the method is applied to an electronic device. The method includes the following steps.
  • step S101 obtain a hand image
  • the hand image may be a color image or a grayscale image including the hand.
  • step S102 a feature mapping model is used to extract the first standard feature of the hand in the hand image; the feature mapping model is pre-trained according to the second standard feature of the hand in the synthetic image sample and the real image sample;
  • the image of the hand is input into the feature map model pre-trained to output the first standard feature of the hand.
  • the first standard feature may include at least one of the common features of the hand, such as the skeleton structure of the hand and the relative length of the hand bone.
  • the training of the feature mapping model can first extract the first feature of the synthesized image sample and the second feature of the real image sample, determine the second standard feature according to the distance between the first feature and the second feature, and according to the second standard feature, the second feature A feature and a second feature are trained to obtain a feature mapping model.
  • the second standard feature may include at least one of the common features of the hand, such as the skeleton structure of the hand and the relative length of the hand bone.
  • step S103 the first standard feature is processed to obtain three-dimensional coordinates of multiple key points of the hand;
  • the key points of the hand include palms, wrists, fingertips and so on.
  • the first standard feature can be processed using a model obtained in advance to determine the three-dimensional coordinates of multiple key points of the hand.
  • a coordinate regression model obtained in advance can be used to process the first standard feature to obtain 2.5D coordinates of multiple hand key points in the hand image, where the 2.5D coordinates include pixels of the key points Coordinates and normalized relative depth coordinates; based on the 2.5D coordinates and known camera parameters and pre-specified root node coordinates, the 3D coordinates of each key point of the hand in the hand image can be restored. Subsequent embodiments will introduce this implementation in detail.
  • step S104 the posture of the hand is determined based on the three-dimensional coordinates of each key point.
  • the hand posture can be determined.
  • the gesture recognition method learns the second standard feature common to the hands in the synthesized image sample and the real image sample through the network to obtain a feature mapping model, and then uses the feature mapping model to extract the first Standard feature, the first standard feature is processed to obtain the three-dimensional coordinates of key points, and then the hand pose is determined; this method uses the common features of the hand in the composite image and the real image for model training, eliminating the composite image at the standard feature level
  • This difference between the real image and the real image can improve the accuracy of gesture recognition under weak supervision; and the method does not require synthetic image samples and real image samples with completely consistent gestures, which relaxes the requirements on the training data set; the method also It is suitable for gesture recognition of color images and has a wider range of applications.
  • the above gesture recognition method may further include the step of obtaining a feature mapping model.
  • this step may further include:
  • step S201 synthetic image samples and real image samples are acquired.
  • the synthesized image sample may be a color image or a grayscale image containing hands, which may be synthesized using a renderer; a 3D hand model can also be downloaded from the Internet, and then the posture of the hand model can be adjusted through software to generate Different hands pictures.
  • the real image sample may be a color image or a grayscale image containing hands, which can be obtained by shooting with a camera.
  • the data packet of the training sample may include half of the synthesized image and half of the real image.
  • step S202 the first feature of the synthesized image sample and the second feature of the real image sample are extracted.
  • the first feature of the synthesized image sample and the second feature of the real image sample can be extracted from the encoder in the encoder.
  • an encoder may be used to encode the extracted first and second features into a low-dimensional space, thereby filtering out high-dimensional noise.
  • step S203 the distance between the first feature and the second feature is calculated.
  • a domain classifier can be added to the output of the encoder to narrow the characteristics of the synthesized image samples and real image samples.
  • the purpose is to let the network pay attention to and learn two data samples (synthesized image samples and real image samples )
  • Common feature information such as standard features such as the skeleton structure of the human hand and the relative length of the human skeleton.
  • the domain classifier may calculate the iganstein distance between the first feature and the second feature, and determine whether the first feature and the second feature are similar.
  • step S204 when the distance is less than or equal to a preset threshold, it is determined that the first feature and the second feature are second standard features.
  • the domain classifier cannot distinguish between the first feature and the second feature.
  • the first feature and the second feature can be confirmed as the second standard feature, so as to achieve the extraction of the synthetic image sample and the real image
  • the purpose of the common feature information in the sample can be specifically set according to the actual situation, which is not limited in this application.
  • step S205 model training is performed according to the first feature, the second feature, and the second standard feature to obtain a feature mapping model.
  • the domain classifier is added to the output part of the encoder, and the distance between the first feature and the second feature is minimized by minimizing the loss function.
  • the loss function can update the weight of the encoder through back propagation to obtain a feature mapping model.
  • the gesture recognition method provided by the embodiment of the present application can first extract the features of the synthesized image sample and the real image sample through the encoder, and at the same time, can encode these features from the high-dimensional space to the low-dimensional space to filter out noise; and then classify by domain
  • the decoder determines the common features of the two image samples; in the end, the common features (second standard features) can also be input into the decoder of the autoencoder for decoding, and the features of the low-dimensional space can be decoded into the high-dimensional space to facilitate subsequent Module calculations.
  • step S103 may specifically include:
  • step S301 the coordinate regression model obtained in advance is used to process the first standard feature to obtain the first pixel coordinates and the first depth coordinates of multiple key points of the hand, the first depth coordinates are normalized Relative depth.
  • K represents the number of key points
  • x i , y i represent the pixel coordinates of the i-th key point
  • Z i r represents the relative depth coordinates after the i-th key point is normalized, as shown in formula (2) :
  • Z i represents the depth coordinate of the i-th key point, that is, the depth coordinate determined with the camera lens as the origin;
  • Z root is the coordinate of the root node or reference point selected by man.
  • C in the above formula is a constant, usually set to 1, and S is the length of the joint that is artificially selected for normalization.
  • the training of the coordinate regression model can, for example, obtain the 2.5D coordinates of the key points of the hands in each synthetic image sample and the 2.5D coordinates of the key points of the hands in each real image sample according to the second standard feature of the synthetic image sample and the real image sample Coordinates; then according to the 2.5D coordinates of the hand key points in the composite image sample, the 2.5D coordinates of the hand key points in the real image sample, the three-dimensional coordinates of the hand key points marked in the composite image sample, and the The two-dimensional coordinates of key points of the hand are used to train the neural network model, etc. to obtain the coordinate regression model. Subsequent embodiments will introduce the training process of the coordinate regression model in detail.
  • step S302 the three-dimensional coordinates of each key point are obtained according to the first pixel coordinates, the first depth coordinates of each key point, the preset camera parameters and the root node coordinates.
  • the first pixel coordinates of each key point, the first depth coordinate, and preset camera parameters and root node coordinates may be input into the camera formula (3) to obtain the three-dimensional space coordinates of each key point.
  • the embodiments of the present application make full use of the relationship between the two-dimensional coordinates and the three-dimensional coordinates, reduce the uncertainty of the three-dimensional coordinate prediction in the case of weak supervision, and further improve the accuracy of gesture recognition.
  • the gesture recognition methods of the above embodiments may further include the step of obtaining a coordinate regression model.
  • this step may further include:
  • step S401 the three-dimensional coordinates of the key points of the hand of the synthesized image sample and the two-dimensional coordinates of the key points of the hand of the real image sample are obtained.
  • the three-dimensional coordinates of multiple key points of the hand marked in the synthesized image sample and the two-dimensional coordinates of multiple key points of the hand marked in the real image sample are obtained.
  • step S402 the second standard feature is processed to obtain the second pixel coordinate and the second depth coordinate of the hand key point of each composite image sample, and the third pixel coordinate and the third pixel coordinate of the hand key point of each real image sample The third depth coordinate.
  • a two-stack hourglass network model with skip connection is used for processing, and each composite image sample is obtained through a 2.5D heat map of the hidden variable space output by the network
  • the 2.5D coordinates of the key points of the hand and the 2.5D coordinates of the key points of the hand of each real image sample include the second pixel coordinates and the second depth coordinates (normalized relative depth)
  • the 2.5D coordinates of the hand key points of each real image sample include the first Three-pixel coordinates and third depth coordinates (normalized relative depth).
  • the expressions of 2.5D coordinates and normalized relative depth coordinates can refer to formula (1) and formula (2).
  • step S403 the model is constructed according to the second pixel coordinate, the second depth coordinate, the third pixel coordinate, the third depth coordinate, the three-dimensional coordinate of the hand key point of the synthesized image sample and the two-dimensional coordinate of the hand key point of the real image sample Training to get the coordinate regression model.
  • step S403 it may further include: using a pre-trained coordinate optimization model, the second pixel coordinates, the second depth coordinates, the third Steps for optimization of pixel coordinates and third depth coordinates.
  • the steps of obtaining the coordinate optimization model may specifically include:
  • step S501 the first depth map of the synthesized image sample and the second depth map of the real image sample are acquired.
  • the first depth map of the synthesized image sample can be generated using the manual model information of the synthesized data when generating the synthesized image, and the second depth map of the real image sample can be obtained by using a depth camera.
  • step S502 the second pixel coordinates and the second depth coordinates are processed to obtain a third depth map.
  • an upsampling algorithm may be used to process the 2.5D coordinates (second pixel coordinates and second depth coordinates) of multiple key points in the synthesized image sample to obtain a third depth map.
  • step S503 the third pixel coordinate and the third depth coordinate are processed to obtain a fourth depth map.
  • an upsampling algorithm may be used to process the 2.5D coordinates (third pixel coordinates and the third depth coordinates) of multiple key points in a real image sample to obtain a fourth depth map.
  • step S504 model training is performed according to the first depth map, the second depth map, the third depth map, and the fourth depth map to obtain a coordinate optimization model.
  • the embodiment of the present application adds the constraint of accurate depth map, that is, input the 2.5D coordinates of key points of the hand to a decoder network to predict the depth map Then, the loss function is optimized according to the gap between the accurate depth map and the predicted depth map, which can restrict the range of depth prediction to a certain extent and solve the problem of depth ambiguity.
  • the coordinate optimization model provided in this embodiment can improve the accuracy of 2.5D coordinates of key points of the hand, thereby improving the accuracy of gesture recognition.
  • the gesture recognition method provided by the embodiment of the present application by adding a domain classifier to the output part of the encoder, and narrowing the distribution of the two image samples at the feature level, can better eliminate the difference between the synthesized image sample and the real image sample, and promote The network learns the common feature information of the two data sets, which improves the accuracy of pose estimation compared with the existing weak supervision methods.
  • FIG. 6 is a block diagram of a gesture recognition device shown in this application. 6, the device includes:
  • the first acquiring module 601 is configured to acquire the hand image
  • the feature extraction module 602 is configured to use a feature mapping model to extract the first standard feature of the hand in the hand image; the feature mapping model is based on the second standard feature of the hand in the composite image sample and the real image sample Pre-trained;
  • the coordinate acquisition module 603 is configured to process the first standard feature to obtain three-dimensional coordinates of multiple key points of the hand;
  • the gesture determination module 604 is configured to determine the posture of the hand according to the three-dimensional coordinates of each of the key points.
  • the first standard feature and the second standard feature include at least one of a hand skeleton structure and a relative length of the hand bone.
  • the device further includes: a second obtaining module configured to obtain the feature mapping model;
  • the second obtaining module includes:
  • a first acquisition unit configured to acquire the synthesized image sample and the real image sample
  • a feature extraction unit configured to extract the first feature of the synthesized image sample and the second feature of the real image sample
  • a distance calculation unit configured to calculate the distance between the first feature and the second feature
  • a standard feature unit configured to determine that the first feature and the second feature are second standard features when the distance is less than or equal to a preset threshold
  • the first training unit is configured to perform model training based on the first feature, the second feature, and the second standard feature to obtain the feature mapping model.
  • the coordinate acquisition module 603 includes:
  • the second obtaining unit is configured to process the first standard feature using a coordinate regression model obtained in advance to obtain first pixel coordinates and first depth coordinates of the plurality of key points of the hand.
  • a depth coordinate is the normalized relative depth
  • the third obtaining unit is configured to obtain the three-dimensional coordinates of each key point according to the first pixel coordinates, the first depth coordinates, the preset camera parameters and the root node coordinates of each key point.
  • the device further includes: a third obtaining module configured to obtain the coordinate regression model;
  • the third obtaining module includes:
  • a fourth acquisition unit configured to acquire the three-dimensional coordinates of the key points of the hand of the composite image sample and the two-dimensional coordinates of the key points of the hand of the real image sample;
  • a fifth acquisition unit configured to process the second standard feature to obtain second pixel coordinates and second depth coordinates of key points of the hand of each of the synthesized image samples, and hands of each of the real image samples The third pixel coordinate and the third depth coordinate of the key point;
  • the second training unit is configured to determine the three-dimensional coordinates of key points of the hand of the composite image sample according to the second pixel coordinates, the second depth coordinates, the third pixel coordinates, the third depth coordinates, and The two-dimensional coordinates of key points of the hands of the real image samples are used for model training to obtain the coordinate regression model.
  • the third obtaining module further includes:
  • the coordinate optimization unit is configured to optimize the second pixel coordinate, the second depth coordinate, the third pixel coordinate, and the third depth coordinate using a pre-trained coordinate optimization model.
  • the device further includes: a fourth acquisition module configured to obtain the coordinate optimization model;
  • the fourth obtaining module includes:
  • a sixth acquisition unit configured to acquire a first depth map of the synthesized image sample and a second depth map of the real image sample
  • the first processing unit is configured to perform upsampling processing on the second pixel coordinates and the second depth coordinates to obtain a third depth map;
  • a second processing unit configured to perform upsampling processing on the third pixel coordinates and the third depth coordinates to obtain a fourth depth map
  • the third training unit is configured to perform model training according to the first depth map, the second depth map, the third depth map, and the fourth depth map to obtain the coordinate optimization model.
  • FIG. 7 is a block diagram of the first electronic device 800 shown in this application.
  • the electronic device 800 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness device, personal digital assistant, or the like.
  • the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input/output (Input/Output, I/O) interface 812 , The sensor component 814, and the communication component 816.
  • the processing component 802 generally controls the overall operations of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations.
  • the processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps in the above method.
  • the processing component 802 may include one or more modules to facilitate interaction between the processing component 802 and other components.
  • the processing component 802 may include a multimedia module to facilitate interaction between the multimedia component 808 and the processing component 802.
  • the memory 804 is configured to store various types of data to support operation at the device 800. Examples of these data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, images, videos, and so on.
  • the memory 804 may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random-Access Memory, SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable read only memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-only Memory (PROM), Read-Only Memory (Read-Only Memory , ROM), magnetic memory, flash memory, magnetic disk or optical disk.
  • SRAM static random access memory
  • EEPROM Electrically erasable programmable read-only memory
  • EPROM Erasable Programmable Read Only Memory
  • PROM Programmable Read-only Memory
  • Read-Only Memory Read-Only Memory
  • magnetic memory flash memory, magnetic disk or optical
  • the power supply component 806 provides power to various components of the electronic device 800.
  • the power component 806 may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for the electronic device 800.
  • the multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
  • the screen may include a liquid crystal display (Liquid Crystal) (LCD) and a touch panel (TouchPanel, TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user.
  • the touch panel includes one or more touch sensors to sense touch, swipe, and gestures on the touch panel. The touch sensor may not only sense the boundary of the touch or sliding action, but also detect the duration and pressure related to the touch or sliding operation.
  • the multimedia component 808 includes a front camera and/or a rear camera. When the device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera may receive external multimedia data.
  • Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.
  • the audio component 810 is configured to output and/or input audio signals.
  • the audio component 810 includes a microphone (MIC).
  • the microphone is configured to receive an external audio signal.
  • the received audio signal may be further stored in the memory 804 or transmitted via the communication component 816.
  • the audio component 810 further includes a speaker for outputting audio signals.
  • the I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module.
  • the peripheral interface module may be a keyboard, a click wheel, or a button. These buttons may include, but are not limited to: home button, volume button, start button, and lock button.
  • the sensor component 814 includes one or more sensors for providing the electronic device 800 with various aspects of status assessment.
  • the sensor component 814 can detect the on/off state of the device 800 and the relative positioning of the components, for example, the component is the display and keypad of the electronic device 800, and the sensor component 814 can also detect the electronic device 800 or a component of the electronic device 800 ,
  • the location of the user changes, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration/deceleration of the electronic device 800, and the temperature of the electronic device 800 change.
  • the sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact.
  • the sensor component 814 may also include a light sensor, such as a complementary metal oxide semiconductor (Complementary Metal Oxide Semiconductor (CMOS) or charge-coupled device (CCD) image sensor, for use in imaging applications.
  • CMOS Complementary Metal Oxide Semiconductor
  • CCD charge-coupled device
  • the sensor component 814 may further include an acceleration sensor, a gyro sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
  • the communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices.
  • the electronic device 800 may access a wireless network based on a communication standard, such as WiFi, an operator network (such as 2G, 3G, 4G, or 5G), or a combination thereof.
  • the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel.
  • the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communication.
  • NFC Near Field Communication
  • the NFC module can be based on radio frequency identification (Radio Frequency Identification, RFID) technology, infrared data association (Infrared Data Association (IrDA) technology, ultra wideband (Ultra Wide Band, UWB) technology, Bluetooth (Bluetooth, BT) technology and other Technology to achieve.
  • RFID Radio Frequency Identification
  • IrDA Infrared Data Association
  • UWB Ultra Wide Band
  • Bluetooth Bluetooth, BT
  • the electronic device 800 may be one or more application specific integrated circuits (Application Specific Integrated Circuit (ASIC), digital signal processor (Digital Signal Processing, DSP), digital signal processing device (DSPD), Programmable logic device (programmable logic device, PLD), field programmable gate array (Field Programmable Gate Array, FPGA), controller, microcontroller, microprocessor, or other electronic components are used to implement the above method.
  • ASIC Application Specific Integrated Circuit
  • DSP Digital Signal Processing
  • DSPD digital signal processing device
  • PLD Programmable logic device
  • FPGA field programmable gate array
  • controller microcontroller, microprocessor, or other electronic components are used to implement the above method.
  • a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, which can be executed by the processor 820 of the electronic device 800 to complete the above method.
  • the non-transitory computer-readable storage medium may be ROM, Random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device.
  • FIG. 8 is a block diagram of a second type of electronic device 1900 shown in this application.
  • the electronic device 1900 may be provided as a server.
  • the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by the memory 1932 for storing instructions executable by the processing component 1922, such as application programs.
  • the application programs stored in the memory 1932 may include one or more modules each corresponding to a set of instructions.
  • the processing component 1922 is configured to execute instructions to perform the above method.
  • the electronic device 1900 may also include a power component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to the network, and an input output (I/O) interface 1958 .
  • the electronic device 1900 can operate an operating system based on the memory 1932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • An embodiment of the present application further provides a computer program product, which when the instructions in the computer program product are executed by the processor of the mobile terminal, enables the mobile terminal to execute any one of the keypoint matching methods described above.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Human Computer Interaction (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Software Systems (AREA)
  • Evolutionary Biology (AREA)
  • Computing Systems (AREA)
  • Psychiatry (AREA)
  • Social Psychology (AREA)
  • Medical Informatics (AREA)
  • User Interface Of Digital Computer (AREA)
  • Image Analysis (AREA)

Abstract

一种手势识别方法、装置、电子设备及存储介质，其中手势识别方法包括：获取手部图像（S101）；采用特征映射模型，提取手部的第一标准特征；特征映射模型是根据合成图像样本和真实图像样本中手部的第二标准特征预先训练得到的（S102）；对第一标准特征进行处理，获得手部多个关键点的三维坐标（S103）；根据各关键点的三维坐标，确定手部的姿势（S104）。该方法由于采用合成图像与真实图像中手部的共同特征进行模型训练，在标准特征层面消除了合成图像与真实图像之间的差异，从而可以提升弱监督情形下手势识别的准确率；并且该方法不需要手势完全一致的合成图像样本与真实图像样本，放宽了对训练数据集的要求；该方法还适用于彩色图像的手势识别，应用面更广泛。

Description

手势识别方法、装置、电子设备及存储介质
相关申请的交叉引用
本申请要求在2019年01月04日提交中国专利局、申请号为201910009529.2、申请名称为“手势识别方法、装置、电子设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，尤其涉及一种手势识别方法、装置、电子设备及存储介质。
背景技术
手势识别即人手姿态估计，是指在包含人手的图像中预测人手关键点的三维坐标。现有的人手姿态估计技术中，由于深度歧义性(多个三维坐标可以映射到同一二维坐标)，以及手部姿态灵活多变导致的自遮挡现象十分严重，使得直接在图像上进行三维关键点标注十分困难，因此真实情形下的人手数据集往往没有三维坐标标注。随着渲染技术的发展，使用渲染器合成的人手图像往往数量充足而且标注完善，但是合成的人手与真实场景的人手在皮肤和背景上差距都比较大，因此仅根据合成数据集训练得到的手势识别模型泛化能力较差。
为了缓解真实图像标注困难这个问题，通常做法是采用弱监督人手姿态估计方法，即利用合成数据与真实数据联合训练来提升真实场景姿态估计的准确率。现有的弱监督方法会通过合成数据学习二维到三维的映射关系，并将这种关系应用于真实场景，同时使用易于获得的深度图来约束深度预测的范围，降低深度歧义性的影响。然而，发明人意识到由于合成数据与真实数据具有较大的差异，仅仅深度图约束不足以消除这种差异，因而对三维姿态估计准确率的提升也有限。
发明内容
为克服相关技术中存在的由于合成数据与真实数据具有较大的差异，从而导致三维姿态估计准确率较低的问题，本申请提供一种手势识别方法、装置、电子设备及存储介质。
根据本申请的第一方面，提供一种手势识别方法，所述方法包括：
获取手部图像；
采用特征映射模型，提取所述手部图像中手部的第一标准特征；所述特征映射模型是根据合成图像样本和真实图像样本中手部的第二标准特征预先训练得到的；
对所述第一标准特征进行处理，获得所述手部多个关键点的三维坐标；
根据各所述关键点的三维坐标，确定所述手部的姿势。
根据本申请的第二方面，提供一种手势识别装置，所述装置包括：
第一获取模块，被配置为获取手部图像；
特征提取模块，被配置为采用特征映射模型，提取所述手部图像中手部的第一标准特征；所述特征映射模型是根据合成图像样本和真实图像样本中手部的第二标准特征预先训练得到的；
坐标获取模块，被配置为对所述第一标准特征进行处理，获得所述手部多个关键点的三维坐标；
手势确定模块，被配置为根据各所述关键点的三维坐标，确定所述手部的姿势。
根据本申请的第三方面，提供一种电子设备，所述电子设备包括：
处理器；
用于存储处理器可执行指令的存储器；
其中，所述处理器被配置为执行如第一方面所述的手势识别方法。
根据本申请的第四方面，提供一种非临时性计算机可读存储介质，当所述存储介质中的指令由电子设备的处理器执行时，使得电子设备能够执行如第一方面所述的手势识别方法。
根据本申请的第五方面，提供一种计算机程序产品，当所述计算机程序产品中的指令由电子设备的处理器执行时，使得电子设备能够执行如第一方面所述的手势识别方法。
本申请提供的技术方案可以包括以下有益效果：
本申请提供的手势识别方法，由于采用合成图像与真实图像中手部的共同特征进行模型训练，在标准特征层面消除了合成图像与真实图像之间的差异，从而可以提升弱监督情形下手势识别的准确率；并且该方法不需要手势完全一致的合成图像样本与真实图像样本，放宽了对训练数据集的要求；该方法还适用于彩色图像的手势识别，应用面更广泛。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，并不能限制本申请。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，示出了符合本申请的实施例，并与说明书一起用于解释本申请的原理。
图1是本申请示出的一种手势识别方法的步骤流程图；
图2是本申请示出的一种获得特征映射模型的步骤流程图；
图3是本申请示出的一种获得关键点三维坐标的步骤流程图；
图4是本申请示出的一种获得坐标回归模型的步骤流程图；
图5是本申请示出的一种获得坐标优化模型的步骤流程图；
图6是本申请示出的一种手势识别装置的框图；
图7是本申请示出的第一种电子设备的框图；
图8是本申请示出的第二种电子设备的框图。
具体实施方式
这里将详细地对示例性实施例进行说明，其示例表示在附图中。下面的描述涉及附图时，除非另有表示，不同附图中的相同数字表示相同或相似的 要素。以下示例性实施例中所描述的实施方式并不代表与本申请相一致的所有实施方式。相反，它们仅是与如所附权利要求书中所详述的、本申请的一些方面相一致的装置和方法的例子。
图1是本申请示出的一种手势识别方法的流程图，如图1所示，该方法应用于电子设备中，该方法包括以下步骤。
在步骤S101中，获取手部图像；
其中，该手部图像可以是包含手部的彩色图像或者灰度图像等。
在步骤S102中，采用特征映射模型，提取手部图像中手部的第一标准特征；特征映射模型是根据合成图像样本和真实图像样本中手部的第二标准特征预先训练得到的；
将手部图像输入预先训练得到的特征映射模型，可以输出手部的第一标准特征。其中，第一标准特征可以包括手部骨架结构、手部骨骼相对长度等手部共同特征中的至少一种。
特征映射模型的训练例如可以首先提取合成图像样本的第一特征以及真实图像样本的第二特征，根据第一特征和第二特征之间的距离确定第二标准特征，根据第二标准特征、第一特征和第二特征进行模型训练得到特征映射模型。其中，第二标准特征可以包括手部骨架结构、手部骨骼相对长度等手部共同特征中的至少一种。具体的训练过程在后续实施例会详细介绍。
在步骤S103中，对第一标准特征进行处理，获得手部多个关键点的三维坐标；
其中，手部关键点包括手掌、手腕、指尖等。
可以采用预先训练得到的模型对第一标准特征进行处理，确定多个手部关键点的三维坐标。在一种实现方式中，例如可以采用预先训练得到的坐标回归模型对第一标准特征进行处理，得到手部图像中多个手部关键点的2.5D坐标，其中2.5D坐标包括关键点的像素坐标和归一化后的相对深度坐标；根据该2.5D坐标和已知的相机参数以及预先指定的根节点坐标等可以还原得到手部图像中手部各关键点的三维坐标。后续实施例会对该实现方式进行详细 介绍。
在步骤S104中，根据各关键点的三维坐标，确定手部的姿势。
根据各关键点如手掌、手腕、指尖等三维空间位置信息，可以确定手部姿势。
本申请实施例提供的手势识别方法，预先通过网络学习合成图像样本和真实图像样本中手部共同的第二标准特征得到特征映射模型，然后再采用该特征映射模型提取手部图像中的第一标准特征，对第一标准特征进行处理得到关键点的三维坐标，进而确定手部姿势；该方法由于采用合成图像与真实图像中手部的共同特征进行模型训练，在标准特征层面消除了合成图像与真实图像之间的差异，从而可以提升弱监督情形下手势识别的准确率；并且该方法不需要手势完全一致的合成图像样本与真实图像样本，放宽了对训练数据集的要求；该方法还适用于彩色图像的手势识别，应用面更广泛。
一种实现方式中，在步骤S102之前，上述的手势识别方法还可以包括：获得特征映射模型的步骤。参照图2，该步骤可以进一步包括：
在步骤S201中，获取合成图像样本以及真实图像样本。
具体地，合成图像样本可以是包含手部的彩色图像或灰度图像，可以采用渲染器合成得到；还可以首先从网上下载一个3D手部模型，然后通过软件调整这个手部模型的姿态，生成不同的手部图片。真实图像样本可以是包含手部的彩色图像或灰度图像，可以通过相机拍摄得到。在训练样本的数据包中可以包括一半合成图像，一半真实图像。
在步骤S202中，提取合成图像样本的第一特征以及真实图像样本的第二特征。
具体地，可以采用自编码器中的编码器提取合成图像样本的第一特征以及真实图像样本的第二特征。同时为了降低噪声的影响，还可以采用编码器对提取得到的第一特征和第二特征编码至低维空间，从而滤除高维噪声。
在步骤S203中，计算第一特征与第二特征之间的距离。
在实际应用中，可以在编码器的输出部分加入一个领域分类器来拉近合 成图像样本和真实图像样本的特征，目的是让网络关注并学习到两个数据样本(合成图像样本和真实图像样本)的共同特征信息，比如人手的骨架结构、人手的骨骼相对长度等标准特征。
具体地，领域分类器可以计算第一特征与第二特征之间的wasserstein距离，并判断第一特征和第二特征是否相近。
在步骤S204中，当距离小于或等于预设阈值时，确定第一特征和第二特征为第二标准特征。
当wasserstein距离小于或等于预设阈值时，领域分类器无法区分第一特征和第二特征，此时可以确认第一特征和第二特征为第二标准特征，从而达到提取合成图像样本和真实图像样本中共同特征信息的目的。其中预设阈值的大小可以根据实际情况具体设定，本申请对此不作限定。
在步骤S205中，根据第一特征、第二特征以及第二标准特征进行模型训练，得到特征映射模型。
将领域分类器加在编码器输出部分，通过最小化损失函数使得第一特征和第二特征之间的距离最小，损失函数可以通过反向传播更新编码器的权重，从而得到特征映射模型。
本申请实施例提供的手势识别方法，可以首先通过编码器提取合成图像样本和真实图像样本的特征，同时还可以将这些特征从高维空间编码到低维空间以滤除噪声；然后通过领域分类器确定出两个图像样本的共同特征；最后还可以将共同特征(第二标准特征)输入到自编码器的解码器中进行解码，将低维空间的特征解码到高维空间，以便于后续模块的计算。
一种实现方式中，参照图3，在步骤S103中具体可以包括：
在步骤S301中，采用预先训练得到的坐标回归模型，对第一标准特征进行处理，得到手部多个关键点的第一像素坐标和第一深度坐标，第一深度坐标为归一化后的相对深度。
将第一标准特征输入预先训练得到的坐标回归模型，可以输出如手掌、手腕、指尖等多个手部关键点的2.5D坐标P2.5D，包括各关键点的第一像素 坐标和第一深度坐标，如公式(1)所示：
Figure PCTCN2019128029-appb-000001
其中，K代表关键点个数，x i，y i表示第i个关键点的像素坐标，Z i r表示第i个关键点归一化后的相对深度坐标，具体如公式(2)所示：
Figure PCTCN2019128029-appb-000002
其中，Z i表示第i个关键点的深度坐标，即以相机镜头为原点确定的深度坐标；Z root是人为选定的根节点坐标或参照点坐标。上述公式中的C为常数，通常设置为1，S是为了归一化而人为选定的关节长度。
其中，坐标回归模型的训练例如可以根据合成图像样本和真实图像样本的第二标准特征，得到各合成图像样本中手部关键点的2.5D坐标以及各真实图像样本中手部关键点的2.5D坐标；然后再根据合成图像样本中手部关键点的2.5D坐标、真实图像样本中手部关键点的2.5D坐标、合成图像样本中标注的手部关键点三维坐标以及真实图像样本中标注的手部关键点二维坐标对神经网络模型等进行训练，得到坐标回归模型。后续实施例会对坐标回归模型的训练过程进行详细介绍。
在步骤S302中，根据各关键点的第一像素坐标、第一深度坐标以及预先设定的相机参数和根节点坐标，得到各关键点的三维坐标。
例如，可以将各关键点的第一像素坐标、第一深度坐标以及预先设定的相机参数和根节点坐标输入相机公式(3)，得到各关键点的三维空间坐标。
Figure PCTCN2019128029-appb-000003
其中，(X i，Y i，Z i)代表相机空间第i个关键点的三维坐标，x i，y i表示第i个关键点的像素坐标。
本申请实施例充分利用了二维坐标与三维坐标之间的关系，减少了弱监督情形下三维坐标预测的不确定性，进一步提升了手势识别的准确率。
在步骤S301之前，上述各实施例的手势识别方法还可以包括：获得坐标 回归模型的步骤。参照图4，该步骤可以进一步包括：
在步骤S401中，获取合成图像样本的手部关键点三维坐标，以及真实图像样本的手部关键点二维坐标。
具体地，获取在合成图像样本中标注的多个手部关键点的三维坐标，以及在真实图像样本中标注的多个手部关键点的二维坐标。
在步骤S402中，对第二标准特征进行处理，得到各合成图像样本的手部关键点的第二像素坐标和第二深度坐标，以及各真实图像样本的手部关键点的第三像素坐标和第三深度坐标。
根据在步骤S204中获得的第二标准特征，采用带有skip connection(残差连接)的two-stack hourglass网络模型进行处理，通过网络输出的隐变量空间的2.5D热点图，得到各合成图像样本的手部关键点的2.5D坐标以及各真实图像样本的手部关键点的2.5D坐标。其中，各合成图像样本的手部关键点的2.5D坐标包括第二像素坐标和第二深度坐标(归一化后的相对深度)，各真实图像样本的手部关键点的2.5D坐标包括第三像素坐标和第三深度坐标(归一化后的相对深度)。2.5D坐标和归一化后的相对深度坐标的表达形式可以参照公式(1)和公式(2)所示。
在步骤S403中，根据第二像素坐标、第二深度坐标、第三像素坐标、第三深度坐标、合成图像样本的手部关键点三维坐标以及真实图像样本的手部关键点二维坐标进行模型训练，得到坐标回归模型。
可以将合成图像样本的手部关键点的2.5D坐标与在该合成图像样本中标注的手部关键点的三维坐标进行对比，并将真实图像样本的手部关键点的2.5D坐标与在该真实图像样本中标注的二维坐标进行对比，通过最小化损失函数使计算得到的2.5D坐标与标注坐标之间的差距最小，通过反向传播确定模型中各参数，训练得到坐标回归模型。
为了缓解弱监督情形下通过RGB图像预测三维关键点的深度歧义性问题，在上述步骤S403之前还可以包括：采用预先训练好的坐标优化模型，对第二像素坐标、第二深度坐标、第三像素坐标以及第三深度坐标进行优化的步骤。
其中，参照图5，获得坐标优化模型的步骤具体可以包括：
在步骤S501中，获取合成图像样本的第一深度图以及真实图像样本的第二深度图。
其中，合成图像样本的第一深度图可以在生成合成图像的时候，利用合成数据的人手模型信息生成，真实图像样本的第二深度图可以采用深度相机拍摄得到。
在步骤S502中，对第二像素坐标和第二深度坐标进行处理，得到第三深度图。
例如，可以采用上采样算法，对合成图像样本中多个关键点的2.5D坐标(第二像素坐标和第二深度坐标)进行处理，得到第三深度图。
在步骤S503中，对第三像素坐标和第三深度坐标进行处理，得到第四深度图。
例如，可以采用上采样算法，对真实图像样本中多个关键点的2.5D坐标(第三像素坐标和所述第三深度坐标)进行处理，得到第四深度图。
在步骤S504中，根据第一深度图、第二深度图、第三深度图、第四深度图进行模型训练，得到坐标优化模型。
可以将合成图像样本的精确的第一深度图与预测得到的第三深度图进行比对(如各像素点的灰度值)，并将真实图像样本的精确的第二深度图与预测得到的第四深度图进行比对(如各像素点的灰度值)，通过最小化损失函数使预测深度图与精确深度图趋于一致，通过反向传播确定上采样算法中各参数，得到坐标优化模型。
本申请实施例为了缓解弱监督情形下通过RGB图像预测三维关键点的深度歧义性问题，加入了精确深度图约束，即将手部关键点的2.5D坐标输入到一个解码器网络，预测得到深度图，再根据精确深度图与预测深度图之间的差距优化损失函数，这样能够在一定程度上约束深度预测的范围，解决深度歧义性问题。采用本实施例提供的坐标优化模型，可以提高手部关键点的2.5D坐标的准确性，进而提高手势识别的准确性。
本申请实施例提供的手势识别方法，通过在编码器输出部分加入领域分类器，在特征层面拉近两图像样本的分布，可以较好地消除合成图像样本和真实图像样本之间的差异，促使网络学习到两个数据集的共同特征信息，相比现有的弱监督方法，提升了姿态估计的准确率。
图6是本申请示出的一种手势识别装置的框图。参照图6，该装置包括：
第一获取模块601，被配置为获取手部图像；
特征提取模块602，被配置为采用特征映射模型，提取所述手部图像中手部的第一标准特征；所述特征映射模型是根据合成图像样本和真实图像样本中手部的第二标准特征预先训练得到的；
坐标获取模块603，被配置为对所述第一标准特征进行处理，获得所述手部多个关键点的三维坐标；
手势确定模块604，被配置为根据各所述关键点的三维坐标，确定所述手部的姿势。
在一个可选的实现方式中，所述第一标准特征和所述第二标准特征包括手部骨架结构和手部骨骼相对长度中至少一种。
在一个可选的实现方式中，所述装置还包括：第二获取模块，被配置为获得所述特征映射模型；
所述第二获取模块包括：
第一获取单元，被配置为获取所述合成图像样本以及所述真实图像样本；
特征提取单元，被配置为提取所述合成图像样本的第一特征以及所述真实图像样本的第二特征；
距离计算单元，被配置为计算所述第一特征与所述第二特征之间的距离；
标准特征单元，被配置为当所述距离小于或等于预设阈值时，确定所述第一特征和所述第二特征为第二标准特征；
第一训练单元，被配置为根据所述第一特征、所述第二特征以及所述第二标准特征进行模型训练，得到所述特征映射模型。
在一个可选的实现方式中，所述坐标获取模块603包括：
第二获取单元，被配置为采用预先训练得到的坐标回归模型，对所述第一标准特征进行处理，得到所述手部多个关键点的第一像素坐标和第一深度坐标，所述第一深度坐标为归一化后的相对深度；
第三获取单元，被配置为根据各所述关键点的第一像素坐标、第一深度坐标以及预先设定的相机参数和根节点坐标，得到各所述关键点的三维坐标。
在一个可选的实现方式中，所述装置还包括：第三获取模块，被配置为获得所述坐标回归模型；
所述第三获取模块包括：
第四获取单元，被配置为获取所述合成图像样本的手部关键点三维坐标，以及所述真实图像样本的手部关键点二维坐标；
第五获取单元，被配置为对所述第二标准特征进行处理，得到各所述合成图像样本的手部关键点的第二像素坐标和第二深度坐标，以及各所述真实图像样本的手部关键点的第三像素坐标和第三深度坐标；
第二训练单元，被配置为根据所述第二像素坐标、所述第二深度坐标、所述第三像素坐标、所述第三深度坐标、所述合成图像样本的手部关键点三维坐标以及所述真实图像样本的手部关键点二维坐标进行模型训练，得到所述坐标回归模型。
在一个可选的实现方式中，所述第三获取模块还包括：
坐标优化单元，被配置为采用预先训练好的坐标优化模型，对所述第二像素坐标、所述第二深度坐标、所述第三像素坐标以及所述第三深度坐标进行优化。
在一个可选的实现方式中，所述装置还包括：第四获取模块，被配置为获得所述坐标优化模型；
所述第四获取模块包括：
第六获取单元，被配置为获取所述合成图像样本的第一深度图以及所述真实图像样本的第二深度图；
第一处理单元，被配置为对所述第二像素坐标和所述第二深度坐标进行 上采样处理，得到第三深度图；
第二处理单元，被配置为对所述第三像素坐标和所述第三深度坐标进行上采样处理，得到第四深度图；
第三训练单元，被配置为根据所述第一深度图、所述第二深度图、所述第三深度图、所述第四深度图进行模型训练，得到所述坐标优化模型。
关于上述实施例中的装置，其中各个模块执行操作的具体方式以及有益效果已经在有关该方法的实施例中进行了详细描述，此处将不做详细阐述说明。
图7是本申请示出的第一种电子设备800的框图。例如，电子设备800可以是移动电话，计算机，数字广播终端，消息收发设备，游戏控制台，平板设备，医疗设备，健身设备，个人数字助理等。
参照图7，电子设备800可以包括以下一个或多个组件：处理组件802，存储器804，电源组件806，多媒体组件808，音频组件810，输入/输出(Input/Output，I/O)的接口812，传感器组件814，以及通信组件816。
处理组件802通常控制电子设备800的整体操作，诸如与显示，电话呼叫，数据通信，相机操作和记录操作相关联的操作。处理组件802可以包括一个或多个处理器820来执行指令，以完成上述方法的全部或部分步骤。此外，处理组件802可以包括一个或多个模块，便于处理组件802和其他组件之间的交互。例如，处理组件802可以包括多媒体模块，以方便多媒体组件808和处理组件802之间的交互。
存储器804被配置为存储各种类型的数据以支持在设备800的操作。这些数据的示例包括用于在电子设备800上操作的任何应用程序或方法的指令，联系人数据，电话簿数据，消息，图像，视频等。存储器804可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，如静态随机存取存储器(Static Random-Access Memory，SRAM)，电可擦除可编程只读存储器(Electrically Erasable Programmable read only memory，EEPROM)，可擦除可编程只读存储器(Erasable Programmable Read Only Memory，EPROM)，可编 程只读存储器(Programmable read-only memory，PROM)，只读存储器(Read-Only Memory，ROM)，磁存储器，快闪存储器，磁盘或光盘。
电源组件806为电子设备800的各种组件提供电力。电源组件806可以包括电源管理系统，一个或多个电源，及其他与为电子设备800生成、管理和分配电力相关联的组件。
多媒体组件808包括在所述电子设备800和用户之间的提供一个输出接口的屏幕。在一些实施例中，屏幕可以包括液晶显示器(Liquid Crystal Display，LCD)和触摸面板(TouchPanel，TP)。如果屏幕包括触摸面板，屏幕可以被实现为触摸屏，以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界，而且还检测与所述触摸或滑动操作相关的持续时间和压力。在一些实施例中，多媒体组件808包括一个前置摄像头和/或后置摄像头。当设备800处于操作模式，如拍摄模式或视频模式时，前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜系统或具有焦距和光学变焦能力。
音频组件810被配置为输出和/或输入音频信号。例如，音频组件810包括一个麦克风(MIC)，当电子设备800处于操作模式，如呼叫模式、记录模式和语音识别模式时，麦克风被配置为接收外部音频信号。所接收的音频信号可以被进一步存储在存储器804或经由通信组件816发送。在一些实施例中，音频组件810还包括一个扬声器，用于输出音频信号。
I/O接口812为处理组件802和外围接口模块之间提供接口，上述外围接口模块可以是键盘，点击轮，按钮等。这些按钮可包括但不限于：主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件814包括一个或多个传感器，用于为电子设备800提供各个方面的状态评估。例如，传感器组件814可以检测到设备800的打开/关闭状态，组件的相对定位，例如所述组件为电子设备800的显示器和小键盘，传感器组件814还可以检测电子设备800或电子设备800一个组件的位置改变， 用户与电子设备800接触的存在或不存在，电子设备800方位或加速/减速和电子设备800的温度变化。传感器组件814可以包括接近传感器，被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件814还可以包括光传感器，如互补金属氧化物半导体(Complementary Metal Oxide Semiconductor，CMOS)或电荷耦合元件(Charge-coupled Device，CCD)图像传感器，用于在成像应用中使用。在一些实施例中，该传感器组件814还可以包括加速度传感器，陀螺仪传感器，磁传感器，压力传感器或温度传感器。
通信组件816被配置为便于电子设备800和其他设备之间有线或无线方式的通信。电子设备800可以接入基于通信标准的无线网络，如WiFi，运营商网络(如2G、3G、4G或5G)，或它们的组合。在一个示例性实施例中，通信组件816经由广播信道接收来自外部广播管理系统的广播信号或广播相关信息。在一个示例性实施例中，所述通信组件816还包括近场通信(Near Field Communication，NFC)模块，以促进短程通信。例如，在NFC模块可基于射频识别(Radio Frequency Identification，RFID)技术，红外数据协会(Infrared Data Association，IrDA)技术，超宽带(Ultra Wide Band，UWB)技术，蓝牙(Bluetooth，BT)技术和其他技术来实现。
在示例性实施例中，电子设备800可以被一个或多个应用专用集成电路(Application Specific Integrated Circuit，ASIC)、数字信号处理器(Digital Signal Processing，DSP)、数字信号处理设备(DSPD)、可编程逻辑器件(programmable logic device，PLD)、现场可编程门阵列(Field Programmable Gate Array，FPGA)、控制器、微控制器、微处理器或其他电子元件实现，用于执行上述方法。
在示例性实施例中，还提供了一种包括指令的非临时性计算机可读存储介质，例如包括指令的存储器804，上述指令可由电子设备800的处理器820执行以完成上述方法。例如，所述非临时性计算机可读存储介质可以是ROM、随机存取存储器(Random Access Memory，RAM)、CD-ROM、磁带、软盘和 光数据存储设备等。
图8是本申请示出的第二种电子设备1900的框图。例如，电子设备1900可以被提供为一服务器。
参照图8，电子设备1900包括处理组件1922，其进一步包括一个或多个处理器，以及由存储器1932所代表的存储器资源，用于存储可由处理组件1922的执行的指令，例如应用程序。存储器1932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件1922被配置为执行指令，以执行上述方法。
电子设备1900还可以包括一个电源组件1926被配置为执行电子设备1900的电源管理，一个有线或无线网络接口1950被配置为将电子设备1900连接到网络，和一个输入输出(I/O)接口1958。电子设备1900可以操作基于存储在存储器1932的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM,LinuxTM，FreeBSDTM或类似。
本申请实施例还提供一种计算机程序产品，当所述计算机程序产品中的指令由移动终端的处理器执行时，使得移动终端能够执行上述任意一项所述的关键点匹配方法。
本领域技术人员在考虑说明书及实践这里公开的发明后，将容易想到本申请的其它实施方案。本申请旨在涵盖本申请的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本申请的一般性原理并包括本申请未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的，本申请的真正范围和精神由下面的权利要求指出。
应当理解的是，本申请并不局限于上面已经描述并在附图中示出的精确结构，并且可以在不脱离其范围进行各种修改和改变。本申请的范围仅由所附的权利要求来限制。

Claims (16)

  1. 一种手势识别方法，所述方法包括：
    获取手部图像；
    采用特征映射模型，提取所述手部图像中手部的第一标准特征；所述特征映射模型是根据合成图像样本和真实图像样本中手部的第二标准特征预先训练得到的；
    对所述第一标准特征进行处理，获得所述手部多个关键点的三维坐标；
    根据各所述关键点的三维坐标，确定所述手部的姿势。
  2. 根据权利要求1所述的手势识别方法，所述第一标准特征和所述第二标准特征包括手部骨架结构和手部骨骼相对长度中至少一种。
  3. 根据权利要求1所述的手势识别方法，在所述采用特征映射模型，提取所述手部的第一标准特征的步骤之前，还包括：获得所述特征映射模型；
    所述获得所述特征映射模型的步骤，包括：
    获取所述合成图像样本以及所述真实图像样本；
    提取所述合成图像样本的第一特征以及所述真实图像样本的第二特征；
    计算所述第一特征与所述第二特征之间的距离；
    当所述距离小于或等于预设阈值时，确定所述第一特征和所述第二特征为第二标准特征；
    根据所述第一特征、所述第二特征以及所述第二标准特征进行模型训练，得到所述特征映射模型。
  4. 根据权利要求3所述的手势识别方法，所述对所述第一标准特征进行处理，获得所述手部多个关键点的三维坐标的步骤，包括：
    采用预先训练得到的坐标回归模型，对所述第一标准特征进行处理，得到所述手部多个关键点的第一像素坐标和第一深度坐标，所述第一深度坐标为归一化后的相对深度；
    根据各所述关键点的第一像素坐标、第一深度坐标以及预先设定的相机 参数和根节点坐标，得到各所述关键点的三维坐标。
  5. 根据权利要求4所述的手势识别方法，在所述采用预先训练得到的坐标回归模型，对所述第一标准特征进行处理，得到所述手部多个关键点的第一像素坐标和第一深度坐标的步骤之前，还包括：获得所述坐标回归模型；
    所述获得所述坐标回归模型的步骤，包括：
    获取所述合成图像样本的手部关键点三维坐标，以及所述真实图像样本的手部关键点二维坐标；
    对所述第二标准特征进行处理，得到各所述合成图像样本的手部关键点的第二像素坐标和第二深度坐标，以及各所述真实图像样本的手部关键点的第三像素坐标和第三深度坐标；
    根据所述第二像素坐标、所述第二深度坐标、所述第三像素坐标、所述第三深度坐标、所述合成图像样本的手部关键点三维坐标以及所述真实图像样本的手部关键点二维坐标进行模型训练，得到所述坐标回归模型。
  6. 根据权利要求5所述的手势识别方法，在所述根据所述第二像素坐标、所述第二深度坐标、所述第三像素坐标、所述第三深度坐标、所述合成图像样本的手部关键点三维坐标以及所述真实图像样本的手部关键点二维坐标进行模型训练，得到所述坐标回归模型的步骤之前，还包括：
    采用预先训练好的坐标优化模型，对所述第二像素坐标、所述第二深度坐标、所述第三像素坐标以及所述第三深度坐标进行优化。
  7. 根据权利要求6所述的手势识别方法，在所述采用预先训练好的坐标优化模型，对所述第二像素坐标、所述第二深度坐标、所述第三像素坐标以及所述第三深度坐标进行优化的步骤之前，还包括：获得所述坐标优化模型；
    所述获得所述坐标优化模型的步骤，包括：
    获取所述合成图像样本的第一深度图以及所述真实图像样本的第二深度图；
    对所述第二像素坐标和所述第二深度坐标进行上采样处理，得到第三深度图；
    对所述第三像素坐标和所述第三深度坐标进行上采样处理，得到第四深度图；
    根据所述第一深度图、所述第二深度图、所述第三深度图、所述第四深度图进行模型训练，得到所述坐标优化模型。
  8. 一种手势识别装置，所述装置包括：
    第一获取模块，被配置为获取手部图像；
    特征提取模块，被配置为采用特征映射模型，提取所述手部图像中手部的第一标准特征；所述特征映射模型是根据合成图像样本和真实图像样本中手部的第二标准特征预先训练得到的；
    坐标获取模块，被配置为对所述第一标准特征进行处理，获得所述手部多个关键点的三维坐标；
    手势确定模块，被配置为根据各所述关键点的三维坐标，确定所述手部的姿势。
  9. 一种电子设备，所述电子设备包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    其中，所述处理器被配置为当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现下列过程：
    获取手部图像；
    采用特征映射模型，提取所述手部图像中手部的第一标准特征；所述特征映射模型是根据合成图像样本和真实图像样本中手部的第二标准特征预先训练得到的；
    对所述第一标准特征进行处理，获得所述手部多个关键点的三维坐标；
    根据各所述关键点的三维坐标，确定所述手部的姿势。
  10. 根据权利要求9所述的电子设备，所述第一标准特征和所述第二标准特征包括手部骨架结构和手部骨骼相对长度中至少一种。
  11. 根据权利要求9所述的电子设备，在所述采用特征映射模型，提取 所述手部的第一标准特征的步骤之前，所述一个或多个处理器还用于：获得所述特征映射模型；
    所述获得所述特征映射模型时，所述一个或多个处理器具体用于：
    获取所述合成图像样本以及所述真实图像样本；
    提取所述合成图像样本的第一特征以及所述真实图像样本的第二特征；
    计算所述第一特征与所述第二特征之间的距离；
    当所述距离小于或等于预设阈值时，确定所述第一特征和所述第二特征为第二标准特征；
    根据所述第一特征、所述第二特征以及所述第二标准特征进行模型训练，得到所述特征映射模型。
  12. 根据权利要求11所述的电子设备，所述对所述第一标准特征进行处理，获得所述手部多个关键点的三维坐标时，所述一个或多个处理器具体用于：
    采用预先训练得到的坐标回归模型，对所述第一标准特征进行处理，得到所述手部多个关键点的第一像素坐标和第一深度坐标，所述第一深度坐标为归一化后的相对深度；
    根据各所述关键点的第一像素坐标、第一深度坐标以及预先设定的相机参数和根节点坐标，得到各所述关键点的三维坐标。
  13. 根据权利要求12所述的电子设备，在所述采用预先训练得到的坐标回归模型，对所述第一标准特征进行处理，得到所述手部多个关键点的第一像素坐标和第一深度坐标的步骤之前，所述一个或多个处理器还用于：获得所述坐标回归模型；
    所述获得所述坐标回归模型时，所述一个或多个处理器具体用于：
    获取所述合成图像样本的手部关键点三维坐标，以及所述真实图像样本的手部关键点二维坐标；
    对所述第二标准特征进行处理，得到各所述合成图像样本的手部关键点的第二像素坐标和第二深度坐标，以及各所述真实图像样本的手部关键点的 第三像素坐标和第三深度坐标；
    根据所述第二像素坐标、所述第二深度坐标、所述第三像素坐标、所述第三深度坐标、所述合成图像样本的手部关键点三维坐标以及所述真实图像样本的手部关键点二维坐标进行模型训练，得到所述坐标回归模型。
  14. 根据权利要求13所述的电子设备，在所述根据所述第二像素坐标、所述第二深度坐标、所述第三像素坐标、所述第三深度坐标、所述合成图像样本的手部关键点三维坐标以及所述真实图像样本的手部关键点二维坐标进行模型训练，得到所述坐标回归模型之前，所述一个或多个处理器还用于：
    采用预先训练好的坐标优化模型，对所述第二像素坐标、所述第二深度坐标、所述第三像素坐标以及所述第三深度坐标进行优化。
  15. 根据权利要求14所述的电子设备，在所述采用预先训练好的坐标优化模型，对所述第二像素坐标、所述第二深度坐标、所述第三像素坐标以及所述第三深度坐标进行优化之前，所述一个或多个处理器还用于：获得所述坐标优化模型；
    所述获得所述坐标优化模型时，所述一个或多个处理器具体用于：
    获取所述合成图像样本的第一深度图以及所述真实图像样本的第二深度图；
    对所述第二像素坐标和所述第二深度坐标进行上采样处理，得到第三深度图；
    对所述第三像素坐标和所述第三深度坐标进行上采样处理，得到第四深度图；
    根据所述第一深度图、所述第二深度图、所述第三深度图、所述第四深度图进行模型训练，得到所述坐标优化模型。
  16. 一种非临时性计算机可读存储介质，当所述存储介质中的指令由电子设备的处理器执行时，使得电子设备能够执行如权利要求1-7任一项所述的手势识别方法。
PCT/CN2019/128029 2019-01-04 2019-12-24 手势识别方法、装置、电子设备及存储介质 Ceased WO2020140798A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/139,369 US11163373B2 (en) 2019-01-04 2020-12-31 Method and electronic device of gesture recognition

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910009529.2 2019-01-04
CN201910009529.2A CN109858524B (zh) 2019-01-04 2019-01-04 手势识别方法、装置、电子设备及存储介质

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/139,369 Continuation US11163373B2 (en) 2019-01-04 2020-12-31 Method and electronic device of gesture recognition

Publications (1)

Publication Number Publication Date
WO2020140798A1 true WO2020140798A1 (zh) 2020-07-09

Family

ID=66893985

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/128029 Ceased WO2020140798A1 (zh) 2019-01-04 2019-12-24 手势识别方法、装置、电子设备及存储介质

Country Status (3)

Country Link
US (1) US11163373B2 (zh)
CN (1) CN109858524B (zh)
WO (1) WO2020140798A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112115799A (zh) * 2020-08-24 2020-12-22 青岛小鸟看看科技有限公司 基于标记点的三维手势的识别方法、装置及设备
CN113158911A (zh) * 2021-04-25 2021-07-23 北京华捷艾米科技有限公司 数据生成方法及装置
CN113723187A (zh) * 2021-07-27 2021-11-30 武汉光庭信息技术股份有限公司 手势关键点的半自动标注方法及系统
CN114004264A (zh) * 2020-07-13 2022-02-01 珠海格力电器股份有限公司 手势动作检测方法、装置、设备及存储介质
CN114972618A (zh) * 2021-02-22 2022-08-30 广州视源电子科技股份有限公司 三维关键点采集系统、方法、装置及存储介质
CN116758634A (zh) * 2023-06-15 2023-09-15 安徽大学 规范化洗手自动评估方法及系统
US12219239B2 (en) 2020-11-18 2025-02-04 Samsung Electronics Co., Ltd. Method of controlling electronic device by recognizing movement in peripheral zone of field of view of camera, and electronic device therefor

Families Citing this family (35)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109858524B (zh) * 2019-01-04 2020-10-16 北京达佳互联信息技术有限公司 手势识别方法、装置、电子设备及存储介质
US11176699B2 (en) * 2019-05-24 2021-11-16 Tencent America LLC Augmenting reliable training data with CycleGAN for hand pose estimation
CN110443154B (zh) * 2019-07-15 2022-06-03 北京达佳互联信息技术有限公司 关键点的三维坐标定位方法、装置、电子设备和存储介质
CN110458494A (zh) * 2019-07-19 2019-11-15 暨南大学 一种无人机物流送货方法及系统
CN110751097B (zh) * 2019-10-22 2023-05-02 中山大学 一种半监督的三维点云手势关键点检测方法
CN110796096B (zh) * 2019-10-30 2023-01-24 北京达佳互联信息技术有限公司 一种手势识别模型的训练方法、装置、设备及介质
WO2021098587A1 (zh) * 2019-11-20 2021-05-27 Oppo广东移动通信有限公司 手势分析方法、装置、设备及计算机可读存储介质
CN110991319B (zh) * 2019-11-29 2021-10-19 广州市百果园信息技术有限公司 手部关键点检测方法、手势识别方法及相关装置
CN111145865A (zh) * 2019-12-26 2020-05-12 中国科学院合肥物质科学研究院 一种基于视觉的手部精细动作训练指导系统及方法
CN111260774B (zh) * 2020-01-20 2023-06-23 北京百度网讯科技有限公司 生成3d关节点回归模型的方法和装置
CN111368668B (zh) * 2020-02-25 2023-05-23 北京字节跳动网络技术有限公司 三维手部识别方法、装置、电子设备及存储介质
CN113448429B (zh) * 2020-03-25 2024-11-01 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
CN111462234B (zh) * 2020-03-27 2023-07-18 北京华捷艾米科技有限公司 一种位置确定方法及装置
CN113553877B (zh) * 2020-04-07 2023-05-30 舜宇光学（浙江）研究院有限公司 深度手势识别方法及其系统和电子设备
CN111488824B (zh) * 2020-04-09 2023-08-08 北京百度网讯科技有限公司 运动提示方法、装置、电子设备和存储介质
CN111401318B (zh) * 2020-04-14 2022-10-04 支付宝(杭州)信息技术有限公司 动作识别方法及装置
US12518653B2 (en) * 2020-05-11 2026-01-06 Sign-Speak Inc. Realtime AI sign language recognition
US12430833B2 (en) 2020-05-11 2025-09-30 Sign-Speak Inc. Realtime AI sign language recognition with avatar
EP3965071B1 (en) * 2020-09-08 2025-01-15 Samsung Electronics Co., Ltd. Method and apparatus for pose identification
CN112949587B (zh) * 2021-03-31 2023-05-02 上海电机学院 基于关键点的手部握姿矫正方法、系统及计算机可读介质
CN113238705A (zh) * 2021-05-10 2021-08-10 青岛小鸟看看科技有限公司 虚拟键盘交互方法及系统
CN113326751B (zh) * 2021-05-19 2024-02-13 中国科学院上海微系统与信息技术研究所 一种手部3d关键点的标注方法
CN115393889A (zh) * 2021-05-24 2022-11-25 阿里巴巴新加坡控股有限公司 图像处理方法、图像变形方法和健身评测方法
CN113408443B (zh) * 2021-06-24 2022-07-05 齐鲁工业大学 基于多视角图像的手势姿态预测方法及系统
CN114185429B (zh) * 2021-11-11 2024-03-26 杭州易现先进科技有限公司 手势关键点定位或姿态估计的方法、电子装置和存储介质
CN114299609B (zh) * 2021-12-17 2025-08-01 云从科技集团股份有限公司 行为识别模型训练及行为识别方法、装置及计算机存储介质
CN114332933A (zh) * 2021-12-29 2022-04-12 苏州臻迪智能科技有限公司 一种模型训练、手势识别方法、装置、设备及介质
CN114693779B (zh) * 2022-04-02 2025-04-15 蔚来汽车科技(安徽)有限公司 用于确定手部三维关键点的方法及装置
CN116071822A (zh) * 2023-01-03 2023-05-05 科大讯飞股份有限公司 一种手语识别方法及相关装置、设备、存储介质
CN116051926B (zh) * 2023-01-12 2024-04-16 北京百度网讯科技有限公司 图像识别模型的训练方法、图像识别方法和装置
CN116486476A (zh) * 2023-03-15 2023-07-25 业成科技（成都）有限公司 基于3d技术的手势构建方法、装置和计算机设备
CN116185205B (zh) * 2023-04-25 2023-07-18 中国电子科技集团公司第十五研究所 非接触手势交互方法和装置
CN116665245B (zh) * 2023-05-12 2026-01-09 上海数迹智能科技有限公司 多视角手势识别方法、装置、计算机设备和存储介质
US20250341897A1 (en) * 2024-05-02 2025-11-06 Dell Products L.P. Gesture-based processing of digital human responses
CN118838534A (zh) * 2024-09-12 2024-10-25 深圳市奥盛通科技有限公司 一种基于加固计算机的人机交互方法及装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104598915A (zh) * 2014-01-24 2015-05-06 深圳奥比中光科技有限公司 一种手势识别方法与装置
US20150378444A1 (en) * 2011-08-12 2015-12-31 The Research Foundation For The State University Of New York Hand pointing estimation for human computer interaction
CN107103613A (zh) * 2017-03-28 2017-08-29 深圳市未来媒体技术研究院 一种三维手势姿态估计方法
CN108062526A (zh) * 2017-12-15 2018-05-22 厦门美图之家科技有限公司 一种人体姿态估计方法及移动终端
CN108229318A (zh) * 2017-11-28 2018-06-29 北京市商汤科技开发有限公司 手势识别和手势识别网络的训练方法及装置、设备、介质
CN109858524A (zh) * 2019-01-04 2019-06-07 北京达佳互联信息技术有限公司 手势识别方法、装置、电子设备及存储介质

Family Cites Families (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6546309B1 (en) * 2000-06-29 2003-04-08 Kinney & Lange, P.A. Virtual fitting room
US7152051B1 (en) * 2002-09-30 2006-12-19 Michael Lamport Commons Intelligent control with hierarchical stacked neural networks
US8600166B2 (en) * 2009-11-06 2013-12-03 Sony Corporation Real time hand tracking, pose classification and interface control
US20110298897A1 (en) * 2010-06-08 2011-12-08 Iva Sareen System and method for 3d virtual try-on of apparel on an avatar
US9323325B2 (en) * 2011-08-30 2016-04-26 Microsoft Technology Licensing, Llc Enhancing an object of interest in a see-through, mixed reality display device
CN104067295B (zh) * 2011-12-01 2018-06-22 诺基亚技术有限公司 姿势识别方法、用于该方法的装置和计算机程序
US8934675B2 (en) * 2012-06-25 2015-01-13 Aquifi, Inc. Systems and methods for tracking human hands by performing parts based template matching using images from multiple viewpoints
CN103577793B (zh) * 2012-07-27 2017-04-05 中兴通讯股份有限公司 手势识别方法及装置
US10430985B2 (en) * 2014-03-14 2019-10-01 Magic Leap, Inc. Augmented reality systems and methods utilizing reflections
US9418319B2 (en) * 2014-11-21 2016-08-16 Adobe Systems Incorporated Object detection using cascaded convolutional neural networks
CN105425964B (zh) * 2015-11-30 2018-07-13 青岛海信电器股份有限公司 一种手势识别方法和系统
US10157309B2 (en) * 2016-01-14 2018-12-18 Nvidia Corporation Online detection and classification of dynamic gestures with recurrent convolutional neural networks
CN106997236B (zh) * 2016-01-25 2018-07-13 亮风台（上海）信息科技有限公司 基于多模态输入进行交互的方法和设备
US9933855B2 (en) * 2016-03-31 2018-04-03 Intel Corporation Augmented reality in a field of view including a reflection
WO2018014018A1 (en) * 2016-07-15 2018-01-18 University Of Central Florida Research Foundation, Inc. Synthetic data generation of time series data
EP3500911B1 (en) * 2016-08-22 2023-09-27 Magic Leap, Inc. Augmented reality display device with deep learning sensors
CN107273871A (zh) * 2017-07-11 2017-10-20 夏立 一种人脸特征模型的训练方法及装置
CN107463326A (zh) * 2017-07-17 2017-12-12 努比亚技术有限公司 一种移动终端触控手势识别方法、移动终端及存储介质
IL273991B2 (en) * 2017-10-26 2023-11-01 Magic Leap Inc Gradual normalization systems and methods for adaptive loss balancing in deep multitasking networks
CN107766842B (zh) * 2017-11-10 2020-07-28 济南大学 一种手势识别方法及其应用
US20190244062A1 (en) * 2018-02-04 2019-08-08 KaiKuTek Inc. Gesture recognition method, gesture recognition system, and performing device therefore
CN108256504A (zh) * 2018-02-11 2018-07-06 苏州笛卡测试技术有限公司 一种基于深度学习的三维动态手势识别方法
US10606364B2 (en) * 2018-06-14 2020-03-31 Dell Products, L.P. Two-handed gesture sequences in virtual, augmented, and mixed reality (xR) applications
US10592002B2 (en) * 2018-06-14 2020-03-17 Dell Products, L.P. Gesture sequence recognition using simultaneous localization and mapping (SLAM) components in virtual, augmented, and mixed reality (xR) applications
US11047691B2 (en) * 2018-10-31 2021-06-29 Dell Products, L.P. Simultaneous localization and mapping (SLAM) compensation for gesture recognition in virtual, augmented, and mixed reality (xR) applications
US10999088B2 (en) * 2018-11-20 2021-05-04 Dell Products, L.P. Proximity and context-based telepresence in collaborative environments
US10928899B2 (en) * 2019-01-11 2021-02-23 Dell Products, L.P. Handling multiple eye-gaze tracking (EGT) sources in virtual, augmented, or mixed reality (XR) applications
US11501495B2 (en) * 2019-05-10 2022-11-15 Qualcomm Incorporated Virtual models for communications between autonomous vehicles and external observers
EP3973468A4 (en) * 2019-05-21 2022-09-14 Magic Leap, Inc. Hand pose estimation

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150378444A1 (en) * 2011-08-12 2015-12-31 The Research Foundation For The State University Of New York Hand pointing estimation for human computer interaction
CN104598915A (zh) * 2014-01-24 2015-05-06 深圳奥比中光科技有限公司 一种手势识别方法与装置
CN107103613A (zh) * 2017-03-28 2017-08-29 深圳市未来媒体技术研究院 一种三维手势姿态估计方法
CN108229318A (zh) * 2017-11-28 2018-06-29 北京市商汤科技开发有限公司 手势识别和手势识别网络的训练方法及装置、设备、介质
CN108062526A (zh) * 2017-12-15 2018-05-22 厦门美图之家科技有限公司 一种人体姿态估计方法及移动终端
CN109858524A (zh) * 2019-01-04 2019-06-07 北京达佳互联信息技术有限公司 手势识别方法、装置、电子设备及存储介质

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114004264A (zh) * 2020-07-13 2022-02-01 珠海格力电器股份有限公司 手势动作检测方法、装置、设备及存储介质
CN112115799A (zh) * 2020-08-24 2020-12-22 青岛小鸟看看科技有限公司 基于标记点的三维手势的识别方法、装置及设备
CN112115799B (zh) * 2020-08-24 2023-12-26 青岛小鸟看看科技有限公司 基于标记点的三维手势的识别方法、装置及设备
US12217544B2 (en) 2020-08-24 2025-02-04 Qingdao Pico Technology Co., Ltd. Method, apparatus and device for recognizing three-dimensional gesture based on mark points
US12219239B2 (en) 2020-11-18 2025-02-04 Samsung Electronics Co., Ltd. Method of controlling electronic device by recognizing movement in peripheral zone of field of view of camera, and electronic device therefor
CN114972618A (zh) * 2021-02-22 2022-08-30 广州视源电子科技股份有限公司 三维关键点采集系统、方法、装置及存储介质
CN113158911A (zh) * 2021-04-25 2021-07-23 北京华捷艾米科技有限公司 数据生成方法及装置
CN113723187A (zh) * 2021-07-27 2021-11-30 武汉光庭信息技术股份有限公司 手势关键点的半自动标注方法及系统
CN116758634A (zh) * 2023-06-15 2023-09-15 安徽大学 规范化洗手自动评估方法及系统

Also Published As

Publication number Publication date
US20210124425A1 (en) 2021-04-29
US11163373B2 (en) 2021-11-02
CN109858524B (zh) 2020-10-16
CN109858524A (zh) 2019-06-07

Similar Documents

Publication Publication Date Title
WO2020140798A1 (zh) 手势识别方法、装置、电子设备及存储介质
US11308692B2 (en) Method and device for processing image, and storage medium
CN106651955B (zh) 图片中目标物的定位方法及装置
JP6134446B2 (ja) 画像分割方法、画像分割装置、画像分割デバイス、プログラム及び記録媒体
CN106355573B (zh) 图片中目标物的定位方法及装置
CN106339680B (zh) 人脸关键点定位方法及装置
WO2021008158A1 (zh) 一种人体关键点检测方法及装置、电子设备和存储介质
US20210256672A1 (en) Method, electronic device and storage medium for processing image
WO2020192252A1 (zh) 图像生成方法及装置、电子设备和存储介质
CN112115894B (zh) 手部关键点检测模型的训练方法、装置及电子设备
CN109410276B (zh) 关键点位置确定方法、装置及电子设备
TWI718631B (zh) 人臉圖像的處理方法及裝置、電子設備和儲存介質
CN110889382A (zh) 虚拟形象渲染方法及装置、电子设备和存储介质
WO2020114236A1 (zh) 关键点检测方法、装置、电子设备及存储介质
CN107424130B (zh) 图片美颜方法和装置
CN114581525B (zh) 姿态确定方法及装置、电子设备和存储介质
CN111462238B (zh) 姿态估计优化方法、装置及存储介质
CN107967459B (zh) 卷积处理方法、装置及存储介质
TW202029062A (zh) 網路優化方法及裝置、圖像處理方法及裝置、儲存媒體
CN111626183A (zh) 一种目标对象展示方法及装置、电子设备和存储介质
CN107480785B (zh) 卷积神经网络的训练方法及装置
CN106327473A (zh) 前景图像的获取方法及装置
CN109325908B (zh) 图像处理方法及装置、电子设备和存储介质
CN107133354A (zh) 图像描述信息的获取方法及装置
CN110619325A (zh) 一种文本识别方法及装置

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: 19907800

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: 19907800

Country of ref document: EP

Kind code of ref document: A1