A dynamic gesture recognition method, device and storage medium

Through deep learning algorithms, the spatial and temporal key points of 3D hands are modeled, which solves the problem of difficulty in identifying changes in timing of dynamic gesture recognition in the existing technology, improves the recognition accuracy and real-timeness, and achieves a more efficient gesture recognition effect.

CN118470793BActive Publication Date: 2024-11-15GUANGZHOU BAOLUN ELECTRONICS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410614636.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-17
Publication Date
2024-11-15
Estimated Expiration
2044-05-17

AI Technical Summary

Technical Problem

When the existing gesture recognition method recognizes dynamic gestures, it is difficult to accurately identify the timing trend of gestures, resulting in misidentification problems, and the sensor acquisition method is costly and has limited accuracy.

Method used

Through deep learning algorithms, model the space-time key points of 3D hand, use the camera to obtain video frame groups, perform hand detection and key point extraction, calculate spatial and temporal representation, input gesture recognition model for feature extraction and mapping, and realize dynamic gestures Identify.

Benefits of technology

It improves the accuracy and real-timeness of dynamic gesture recognition, eliminates the influence of the speed of hand movement, and enhances the reliability of recognition results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118470793B_ABST
    Figure CN118470793B_ABST
Patent Text Reader

Abstract

The present invention discloses a dynamic gesture recognition method, device and storage medium, which belongs to the field of deep learning technology. The method comprises: obtaining a video frame group through a camera; the video frame group comprises a plurality of continuous video frames; inputting the video frame group into a hand detection model to judge whether a hand appears in the video frame group, and if so, determining the number of frames in which the hand appears; judging whether the number of frames in which the hand appears is greater than a preset number; if the number of frames in which the hand appears is greater than the preset number, inputting the video frame group into a key point extraction model to extract 3D hand key points of the video frame group; calculating the 3D hand key points to obtain a spatial representation and a temporal representation; inputting the spatial representation and the temporal representation into a gesture recognition model, extracting features from the spatial representation and the temporal representation, and mapping to obtain a dynamic gesture recognition result, so as to realize modeling of 3D hand spatiotemporal key points by using a deep learning algorithm, and improve the accuracy and real-time performance of dynamic gesture recognition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a dynamic gesture recognition method, device and storage medium. Background Art

[0002] Existing gesture recognition methods include static gesture image recognition and gesture recognition methods that use sensors to collect human electromyographic signals.

[0003] Among them, static gesture image recognition determines the dynamic gesture category by calculating the finger bending angle and average displacement of each hand key point between image frame groups. It only uses the spatial position characteristics of the hand, but does not consider the temporal change trend of the gesture context. It is difficult to accurately identify dynamic gestures. When the boundaries of gestures are not clear, misrecognition problems will occur, resulting in low gesture recognition accuracy. Using sensors to collect human electromyography signals will incur high equipment expenditures and labor costs. At the same time, the accuracy of gesture recognition will also be limited by the accuracy of the sensor, resulting in low gesture recognition accuracy. Summary of the invention

[0004] The present invention provides a dynamic gesture recognition method, device and storage medium to achieve modeling of 3D hand spatiotemporal key points using a deep learning algorithm, thereby improving the accuracy and real-time performance of dynamic gesture recognition.

[0005] The present invention provides a dynamic gesture recognition method, comprising:

[0006] Acquire a video frame group through a camera; the video frame group includes a plurality of consecutive video frames;

[0007] Input the video frame group into a hand detection model to determine whether a hand appears in the video frame group, and if so, determine the number of frames in which the hand appears;

[0008] Determine whether the number of frames where the hand appears is greater than a preset number;

[0009] If the number of hand frames is greater than a preset number, inputting the video frame group into a key point extraction model to extract 3D hand key points of the video frame group;

[0010] The 3D hand key points are calculated to obtain a spatial representation and a temporal representation; the spatial representation and the temporal representation are input into a gesture recognition model, features are extracted from the spatial representation and the temporal representation, and mapped to obtain a dynamic gesture recognition result.

[0011] Furthermore, the step of inputting the video frame group into a hand detection model further includes:

[0012] If no hands appear in the video frame group, or the number of hand frames is less than or equal to a preset number, a new video frame group is obtained to input into the hand detection model.

[0013] Furthermore, the hand detection model is used to determine whether a hand appears in the input video frame group and determine the number of frames in which a hand appears in the input video frame group.

[0014] Furthermore, the key point extraction model is used to extract 3D hand key points of the input video frame group.

[0015] Furthermore, the gesture recognition model is constructed by a convolutional neural network; the gesture recognition model includes several layers of one-dimensional convolution modules and several fully connected modules.

[0016] Furthermore, the 3D hand key points are calculated to obtain spatial representation and temporal representation, including:

[0017] Arrange the coordinates of the 3D hand key points in each video frame in spatial order;

[0018] Take any two 3D hand key points of each video frame as a 3D hand key point pair;

[0019] Calculating the first Euclidean distance between the coordinates of the 3D hand key point pairs and the second Euclidean distance between the coordinates of all 3D hand key points and the origin;

[0020] Generate a spatial distance matrix according to the first Euclidean distance and the second Euclidean distance;

[0021] The spatial distance matrix is ​​used as a spatial representation.

[0022] Furthermore, the 3D hand key points are calculated to obtain spatial representation and temporal representation, further comprising:

[0023] Arrange the coordinates of the 3D hand key points in each video frame in chronological order and record them as Slow trajectories;

[0024] The video frames in the Slow track are equally divided into half of the number of video frames in the Slow track by the slicing method, which is recorded as the Fast track;

[0025] Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Slow trajectory to obtain the motion trajectory information of the Slow scale;

[0026] Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Fast trajectory to obtain the motion trajectory information of the Fast scale;

[0027] The Slow-scale motion trajectory information and the Fast-scale motion trajectory information are used as time representation.

[0028] Furthermore, the spatial representation and the time representation are input into a gesture recognition model, feature extraction is performed on the spatial representation and the time representation, and mapping is performed to obtain a dynamic gesture recognition result, specifically:

[0029] Several layers of one-dimensional convolution modules are used to extract features of the spatial distance matrix and the motion trajectory information corresponding to several video frame subgroups, respectively, to obtain spatial variation features and temporal trend features;

[0030] Fusing the spatial variation feature and the temporal trend feature to obtain a spatiotemporal feature;

[0031] The spatiotemporal features are feature mapped through a plurality of fully connected modules and mapped to dynamic gesture recognition results to obtain dynamic gesture recognition results.

[0032] As a preferred solution, the present invention uses a deep learning algorithm to construct a hand detection model to detect hands in the video, a key point extraction model to obtain 3D hand spatiotemporal key points, and a gesture recognition model to model the 3D hand spatiotemporal key points using a deep learning algorithm. Dynamic gesture recognition is performed through the spatiotemporal change characteristics of the 3D hand key points and the deep learning model. The spatiotemporal change characteristics can better reflect the distance changes and movement trends of the hand key points during the movement process, and can also eliminate the influence of the speed of hand movement, thereby improving the accuracy and real-time performance of dynamic gesture recognition.

[0033] Other features and advantages of the present invention will be described in detail in the following detailed description.

[0034] Accordingly, the present invention also provides a dynamic gesture recognition device, comprising: an acquisition module, a judgment module, an extraction module and a recognition module;

[0035] The acquisition module is used to acquire a video frame group through a camera; the video frame group includes a plurality of continuous video frames;

[0036] The judgment module is used to input the video frame group into the hand detection model, judge whether the hand appears in the video frame group, and if so, determine the number of frames in which the hand appears;

[0037] The extraction module is used to determine whether the number of frames in which the hand appears is greater than a preset number;

[0038] If the number of hand frames is greater than a preset number, inputting the video frame group into a key point extraction model to extract 3D hand key points of the video frame group;

[0039] The recognition module is used to calculate the 3D hand key points to obtain spatial representation and temporal representation; input the spatial representation and the temporal representation into a gesture recognition model, perform feature extraction on the spatial representation and the temporal representation, and map them to obtain dynamic gesture recognition results.

[0040] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute a dynamic gesture recognition method as described in the content of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of an embodiment of a dynamic gesture recognition method provided by the present invention;

[0042] Figure 2 It is a structural schematic diagram of an embodiment of a dynamic gesture recognition device provided by the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Embodiment 1

[0045] Please refer to Figure 1 , a dynamic gesture recognition method provided by an embodiment of the present invention, comprising steps S101-S103:

[0046] Step S101: obtaining a video frame group through a camera; the video frame group includes a plurality of consecutive video frames;

[0047] In this embodiment, a continuous video frame group with a fixed frame number m is obtained each time, where m is a positive integer.

[0048] Step S102: inputting the video frame group into a hand detection model to determine whether a hand appears in the video frame group, and if so, determining the number of frames where the hand appears;

[0049] Furthermore, the hand detection model is used to determine whether a hand appears in the input video frame group and determine the number of frames in which a hand appears in the input video frame group.

[0050] Step S103: determining whether the number of frames in which the hand appears is greater than a preset number;

[0051] If the number of hand frames is greater than a preset number, inputting the video frame group into a key point extraction model to extract 3D hand key points of the video frame group;

[0052] Furthermore, the key point extraction model is used to extract 3D hand key points of the input video frame group.

[0053] Furthermore, the step of inputting the video frame group into a hand detection model further includes:

[0054] If no hands appear in the video frame group, or the number of hand frames is less than or equal to a preset number, a new video frame group is obtained to input into the hand detection model.

[0055] In this embodiment, the preset number n is less than or equal to m, and n is a positive integer.

[0056] A new video frame group of the next time period is obtained in chronological order and input into the hand detection model, while keeping the fixed number of input frames m.

[0057] Step S104: Calculate the 3D hand key points to obtain spatial representation and temporal representation; input the spatial representation and the temporal representation into a gesture recognition model, extract features from the spatial representation and the temporal representation, and map them to obtain dynamic gesture recognition results.

[0058] Furthermore, the gesture recognition model is constructed by a convolutional neural network; the gesture recognition model includes several layers of one-dimensional convolution modules and several fully connected modules.

[0059] The gesture recognition model is a deep learning model.

[0060] Furthermore, the 3D hand key points are calculated to obtain spatial representation and temporal representation, including:

[0061] Arrange the coordinates of the 3D hand key points in each video frame in spatial order;

[0062] Take any two 3D hand key points of each video frame as a 3D hand key point pair;

[0063] Calculating the first Euclidean distance between the coordinates of the 3D hand key point pairs and the second Euclidean distance between the coordinates of all 3D hand key points and the origin;

[0064] Generate a spatial distance matrix according to the first Euclidean distance and the second Euclidean distance;

[0065] The spatial distance matrix is ​​used as a spatial representation.

[0066] In this embodiment, in order to avoid duplication of information, only the upper triangular elements in the spatial distance matrix are used to construct feature vectors for expressing the spatial variation of hand motion.

[0067] Furthermore, the 3D hand key points are calculated to obtain spatial representation and temporal representation, further comprising:

[0068] Arrange the coordinates of the 3D hand key points in each video frame in chronological order and record them as Slow trajectories;

[0069] The video frames in the Slow track are equally divided into half of the number of video frames in the Slow track by the slicing method, which is recorded as the Fast track;

[0070] Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Slow trajectory to obtain the motion trajectory information of the Slow scale;

[0071] Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Fast trajectory to obtain the motion trajectory information of the Fast scale;

[0072] The Slow-scale motion trajectory information and the Fast-scale motion trajectory information are used as time representation.

[0073] In this embodiment, it is assumed that each gesture action will generate a video frame group of T frames, which is recorded as a Slow video frame group. In this embodiment, the video frame group is divided into a video frame subgroup of T / 2 frames by a segmentation method, which is recorded as a Fast video frame group.

[0074] By calculating the coordinate difference between two adjacent key points in each frame of the Slow video frame group and the Fast video frame group respectively, we can obtain motion trajectory information of two scales, Fast motion trajectory information and Slow motion trajectory information, which are used to describe the time trend of hand movement and eliminate the influence of movement speed.

[0075] Furthermore, the spatial representation and the time representation are input into a gesture recognition model, feature extraction is performed on the spatial representation and the time representation, and mapping is performed to obtain a dynamic gesture recognition result, specifically:

[0076] Several layers of one-dimensional convolution modules are used to extract features of the spatial distance matrix and the motion trajectory information corresponding to several video frame subgroups, respectively, to obtain spatial variation features and temporal trend features;

[0077] Fusing the spatial variation feature and the temporal trend feature to obtain a spatiotemporal feature;

[0078] The spatiotemporal features are feature mapped through a plurality of fully connected modules and mapped to dynamic gesture recognition results to obtain dynamic gesture recognition results.

[0079] In this embodiment, the gesture recognition model includes three layers of one-dimensional convolution modules.

[0080] Among them, the three-layer one-dimensional convolution module extracts features of the spatial distance matrix, Fast motion trajectory information and Slow motion trajectory information respectively to obtain spatial change features, first time trend features and second time trend features.

[0081] In this embodiment, in order to improve the response speed of the recognition model in the actual measurement scenario, a lightweight convolution is used to construct the recognition model to make it more real-time.

[0082] Implementing the embodiments of the present invention has the following effects:

[0083] The present invention uses a deep learning algorithm to construct a hand detection model to detect hands in the video, a key point extraction model to obtain 3D hand spatiotemporal key points, and a gesture recognition model to model the 3D hand spatiotemporal key points using a deep learning algorithm. Dynamic gesture recognition is performed through the spatiotemporal change characteristics of the 3D hand key points and the deep learning model. The spatiotemporal change characteristics can better reflect the distance changes and movement trends of the hand key points during the movement process, and can also eliminate the influence of the speed of hand movement, thereby improving the accuracy and real-time performance of dynamic gesture recognition.

[0084] Embodiment 2

[0085] Please refer to Figure 2 , a dynamic gesture recognition device provided by an embodiment of the present invention, comprising: an acquisition module 201, a judgment module 202, an extraction module 203 and a recognition module 204;

[0086] The acquisition module is used to acquire a video frame group through a camera; the video frame group includes a plurality of continuous video frames;

[0087] The judgment module is used to input the video frame group into the hand detection model, judge whether the hand appears in the video frame group, and if so, determine the number of frames in which the hand appears;

[0088] The extraction module is used to determine whether the number of frames in which the hand appears is greater than a preset number;

[0089] If the number of hand frames is greater than a preset number, inputting the video frame group into a key point extraction model to extract 3D hand key points of the video frame group;

[0090] The recognition module is used to calculate the 3D hand key points to obtain spatial representation and temporal representation; input the spatial representation and the temporal representation into a gesture recognition model, perform feature extraction on the spatial representation and the temporal representation, and map them to obtain dynamic gesture recognition results.

[0091] Further, the judgment module includes: an updating unit;

[0092] The updating unit is used to obtain a new video frame group to input into the hand detection model if no hand appears in the video frame group or the number of hand frames is less than or equal to a preset number.

[0093] Furthermore, the hand detection model is used to determine whether a hand appears in the input video frame group and determine the number of frames in which a hand appears in the input video frame group.

[0094] Furthermore, the key point extraction model is used to extract 3D hand key points of the input video frame group.

[0095] Furthermore, the gesture recognition model is constructed by a convolutional neural network; the gesture recognition model includes several layers of one-dimensional convolution modules and several fully connected modules.

[0096] Further, the recognition module includes: a spatial representation calculation unit, a temporal representation calculation unit and a recognition unit;

[0097] The spatial representation calculation unit is used to arrange the coordinates of the 3D hand key points in each video frame in a spatial order;

[0098] Take any two 3D hand key points of each video frame as a 3D hand key point pair;

[0099] Calculating the first Euclidean distance between the coordinates of the 3D hand key point pairs and the second Euclidean distance between the coordinates of all 3D hand key points and the origin;

[0100] Generate a spatial distance matrix according to the first Euclidean distance and the second Euclidean distance;

[0101] The spatial distance matrix is ​​used as a spatial representation.

[0102] The time representation calculation unit is used to arrange the coordinates of the 3D hand key points in each video frame in time sequence, which is recorded as a Slow trajectory;

[0103] The video frames in the Slow track are equally divided into half of the number of video frames in the Slow track by the slicing method, which is recorded as the Fast track;

[0104] Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Slow trajectory to obtain the motion trajectory information of the Slow scale;

[0105] Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Fast trajectory to obtain the motion trajectory information of the Fast scale;

[0106] The Slow-scale motion trajectory information and the Fast-scale motion trajectory information are used as time representation.

[0107] The recognition unit is used to extract features of the spatial distance matrix and the motion trajectory information corresponding to the subgroups of several video frames by using several layers of one-dimensional convolution modules, so as to obtain spatial variation features and time trend features;

[0108] Fusing the spatial variation feature and the temporal trend feature to obtain a spatiotemporal feature;

[0109] The spatiotemporal features are feature mapped through a plurality of fully connected modules and mapped to dynamic gesture recognition results to obtain dynamic gesture recognition results.

[0110] The above-mentioned dynamic gesture recognition device can implement the dynamic gesture recognition method of the above-mentioned method embodiment. The options in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The rest of the contents of the embodiment of this application can refer to the contents of the above-mentioned method embodiment, and will not be repeated in this embodiment.

[0111] Embodiment 3

[0112] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the dynamic gesture recognition method described in any one of the above embodiments.

[0113] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the terminal device.

[0114] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0115] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0116] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile terminal, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0117] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0118] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dynamic gesture recognition method, characterized in that: include: Acquire a video frame group through a camera; the video frame group includes a plurality of consecutive video frames; Input the video frame group into a hand detection model to determine whether a hand appears in the video frame group, and if so, determine the number of frames in which the hand appears; Determine whether the number of frames where the hand appears is greater than a preset number; If the number of hand frames is greater than a preset number, inputting the video frame group into a key point extraction model to extract 3D hand key points of the video frame group; The 3D hand key points are calculated to obtain a spatial representation and a temporal representation; the spatial representation and the temporal representation are input into a gesture recognition model, features are extracted from the spatial representation and the temporal representation, and mapped to obtain a dynamic gesture recognition result; The calculating of the 3D hand key points to obtain a spatial representation and a temporal representation includes: Arrange the coordinates of the 3D hand key points in each video frame in spatial order; Take any two 3D hand key points of each video frame as a 3D hand key point pair; Calculating a first Euclidean distance between the coordinates of the 3D hand key point pairs and a second Euclidean distance between the coordinates of all 3D hand key points and the origin; Generate a spatial distance matrix according to the first Euclidean distance and the second Euclidean distance; Using the spatial distance matrix as a spatial representation; Arrange the coordinates of the 3D hand key points in each video frame in chronological order and record them as Slow trajectories; The video frames in the Slow track are equally divided into half of the number of video frames in the Slow track by the slicing method, which is recorded as the Fast track; Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Slow trajectory to obtain the motion trajectory information of the Slow scale; Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Fast trajectory to obtain the motion trajectory information of the Fast scale; The Slow-scale motion trajectory information and the Fast-scale motion trajectory information are used as time representation.

2. A dynamic gesture recognition method as claimed in claim 1, characterized in that: The step of inputting the video frame group into the hand detection model further includes: If no hands appear in the video frame group, or the number of hand frames is less than or equal to a preset number, a new video frame group is obtained to input into the hand detection model.

3. A dynamic gesture recognition method as claimed in claim 1, characterized in that: The hand detection model is used to determine whether a hand appears in an input video frame group and determine the number of frames in which a hand appears in the input video frame group.

4. A dynamic gesture recognition method as claimed in claim 1, characterized in that: The key point extraction model is used to extract 3D hand key points of an input video frame group.

5. A dynamic gesture recognition method as claimed in claim 1, characterized in that: The gesture recognition model is constructed by a convolutional neural network; the gesture recognition model includes several layers of one-dimensional convolution modules and several fully connected modules.

6. A dynamic gesture recognition method as claimed in claim 5, characterized in that: The step of inputting the spatial representation and the temporal representation into a gesture recognition model, extracting features from the spatial representation and the temporal representation, and mapping to obtain a dynamic gesture recognition result is specifically as follows: Several layers of one-dimensional convolution modules are used to extract features of the spatial distance matrix and the motion trajectory information corresponding to several video frame subgroups, respectively, to obtain spatial variation features and temporal trend features; Fusing the spatial variation feature and the temporal trend feature to obtain a spatiotemporal feature; The spatiotemporal features are feature mapped through a plurality of fully connected modules and mapped to dynamic gesture recognition results to obtain dynamic gesture recognition results.

7. A dynamic gesture recognition device, characterized in that: include: Acquisition module, judgment module, extraction module and recognition module; The acquisition module is used to acquire a video frame group through a camera; the video frame group includes a plurality of continuous video frames; The judgment module is used to input the video frame group into the hand detection model, judge whether the hand appears in the video frame group, and if so, determine the number of frames in which the hand appears; The extraction module is used to determine whether the number of frames in which the hand appears is greater than a preset number; If the number of hand frames is greater than a preset number, inputting the video frame group into a key point extraction model to extract 3D hand key points of the video frame group; The recognition module is used to calculate the 3D hand key points to obtain a spatial representation and a temporal representation; input the spatial representation and the temporal representation into a gesture recognition model, perform feature extraction on the spatial representation and the temporal representation, and map them to obtain a dynamic gesture recognition result; The calculating of the 3D hand key points to obtain a spatial representation and a temporal representation includes: Arrange the coordinates of the 3D hand key points in each video frame in spatial order; Take any two 3D hand key points of each video frame as a 3D hand key point pair; Calculating the first Euclidean distance between the coordinates of the 3D hand key point pairs and the second Euclidean distance between the coordinates of all 3D hand key points and the origin; Generate a spatial distance matrix according to the first Euclidean distance and the second Euclidean distance; Using the spatial distance matrix as a spatial representation; Arrange the coordinates of the 3D hand key points in each video frame in chronological order and record them as Slow trajectories; The video frames in the Slow track are equally divided into half of the number of video frames in the Slow track by the slicing method, which is recorded as the Fast track; Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Slow trajectory to obtain the motion trajectory information of the Slow scale; Calculate the coordinate difference between adjacent 3D hand key points in each video frame in the Fast trajectory to obtain the motion trajectory information of the Fast scale; The Slow-scale motion trajectory information and the Fast-scale motion trajectory information are used as time representation.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program; wherein, when the computer program is run, it controls the device where the computer-readable storage medium is located to execute a dynamic gesture recognition method as claimed in any one of claims 1 to 6.