WO2023083030A1

WO2023083030A1 - Posture recognition method and related device

Info

Publication number: WO2023083030A1
Application number: PCT/CN2022/128504
Authority: WO
Inventors: 李志豪; 张子霄; 许松岑
Original assignee: 华为技术有限公司
Priority date: 2021-11-15
Filing date: 2022-10-31
Publication date: 2023-05-19
Also published as: CN114241597A

Abstract

Provided in the present application are a posture recognition method and a related device. The posture of a target object, which is obtained after an image including the target object is processed, has relatively high accuracy, thereby facilitating the improvement of the accuracy of motion information of an acquired target image. The method in the present application comprises: acquiring a target image; and processing, by means of a posture recognition model, a target area of the target image and position information of the target area in the target image, so as to obtain a posture parameter, wherein the target area is the area where the target object is located, and the posture parameter is used for acquiring the posture of the target object.

Description

A gesture recognition method and related equipment

This application claims the priority of the Chinese patent application with the application number 202111350981.9 filed on November 15, 2021, and the title of the invention is "a gesture recognition method and related equipment", the entire contents of which are incorporated herein by reference. Applying.

technical field

This application relates to the technical field of artificial intelligence (AI), in particular to a gesture recognition method and related equipment.

Background technique

Human body motion capture technology is a commonly used technology in the film and television and game industries. This technology can process the input video stream to capture the posture of the human body in each frame of image, so as to obtain the motion information of the human body. Based on this technology, the posture data of human body movement can be extracted from massive videos, which has a wide range of application scenarios.

At present, human body motion capture can be realized based on AI technology. Specifically, for a certain image, the trained neural network can be used to process the area where the human body is located in the image, so as to obtain posture parameters, and the posture parameters can be used to determine the posture of the human body.

If there are multiple images showing the same environment, and the human body is located in different positions in the environment in these multiple images, and the postures of the human body are extremely similar, in this case, the neural network will determine the The posture parameters of the human body in the image are consistent, but in the actual environment, once the human body moves, although the postures of the human body in different positions are similar, there must be subtle differences between the postures of the human body. This is the current neural network. If it cannot be identified, the captured posture of the human body is not accurate enough, which in turn affects the accuracy of the motion information of the human body.

Contents of the invention

The embodiment of the present application provides a posture recognition method and related equipment, which can obtain the posture of the target object after processing the image containing the target object, which has high accuracy, and is conducive to improving the accuracy of the acquired target image. Accuracy of exercise information.

The first aspect of the embodiments of the present application provides a gesture recognition method, the method comprising:

When it is necessary to perform gesture recognition on the target image, the target image can be obtained first. The target image usually presents the target object and the environment in which the target object is located. It can be understood that the goal of gesture recognition is to obtain the pose of the target object in the target image .

In order to reduce the computational load of the pose recognition model, the target image can be preprocessed. Specifically, the target image may be detected to determine an area where the target object is located in the target image, which may be called a target area, and position information of the target area in the target image may be acquired.

After obtaining the target area and the position information of the target area in the target image, the target area and the position information of the target area in the target image can be input to the gesture recognition model, so that the target area of the target image and the target area in the target image can be identified by the gesture recognition model. The position information in the target image is processed to obtain the attitude parameters. Then, the pose of the target object in the target image can be obtained based on the pose parameters.

It can be seen from the above method that after the target image is acquired, the target area where the target object is located in the target image and the position information of the target area in the target image can be input into the gesture recognition model, so that the target image of the target image can be identified by the gesture recognition model. The position information of the area and the target area in the target image is processed to obtain the pose parameters, so the pose of the target object can be obtained based on the pose parameters. In the aforementioned process, the input of the gesture recognition model not only includes the cropped target area, but also includes the position information of the target area in the target image, so the gesture recognition model not only considers the image information of the target area itself when performing image processing The impact on the posture of the target object also takes into account the influence of the position information of the target area in the target image on the posture of the target object. The factors considered are relatively comprehensive, so the posture of the target object obtained based on this method , has higher accuracy, which is beneficial to improve the accuracy of the acquired motion information of the target image.

In a possible implementation manner, the location information includes the coordinates of the center point of the target area in the image coordinate system and the size of the target area, and the image coordinate system is constructed based on the target image. In the aforementioned implementation manner, a certain vertex of the target image can be used as the origin of the image coordinate system, then, the coordinates of the center point of the target area in the image coordinate system can be determined, as well as the length and width of the target area (which is equivalent to obtaining The size of the target area), this information can be used to indicate the position of the target area in the target image, so that after inputting the position information of the target area into the gesture recognition model, the gesture recognition model will perform image processing on the target area , can effectively take into account the influence of the position information of the target area in the target image on the pose of the target object, so the pose of the target object obtained based on these information has high accuracy.

In a possible implementation manner, the location information includes coordinates of vertices of the target area in an image coordinate system, and the image coordinate system is constructed based on the target image. In the aforementioned implementation, a certain vertex of the target image can be used as the origin of the image coordinate system, then the coordinates of all vertices of the target area in the image coordinate system can be determined, and this information can be used to indicate the position of the target area in the target image In this way, after inputting the position information of the target area into the gesture recognition model, the gesture recognition model can effectively take into account the influence of the position information of the target area in the target image on the pose of the target object when processing the image of the target area. Therefore, the posture of the target object obtained based on these information has a high accuracy.

In a possible implementation, the pose recognition model is trained based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real projection result of the real pose of the target object on the image to be processed to obtain . In the foregoing process, during the training process of the gesture recognition model, the gesture recognition model is made to perceive the position of the target object in the whole image, and learn the relationship between the position and the pose of the target object. Then, in the application process of the gesture recognition model, the model can accurately capture the pose of the target object from the whole picture based on the position information of the target object in the whole picture.

In a possible implementation, the pose of the target object includes the orientation of the target object in the camera coordinate system (that is, the orientation of the target object relative to the camera) and the body behavior of the target object in the camera coordinate system (that is, the orientation of the target object in the three-dimensional action in space), the origin of the camera coordinate system is the camera that captures the target image. In the aforementioned implementation, even if multiple images of the target object in different positions in the same environment are obtained, and the poses of the target object in these multiple images are similar, the gesture recognition model can also recognize the difference between the poses of the target object in these multiple images. The difference between objects, that is, the difference between the orientation of the target object relative to the camera, and the difference between the motion of the target object in three-dimensional space.

In a possible implementation, the method further includes: processing the target area of the target image and the position information of the target area in the target image through the gesture recognition model to obtain shape parameters and displacement parameters, and the pose parameters, shape parameters and The displacement parameters are collectively used to obtain the pose of the target object. In the aforementioned implementation manners, the output of the gesture recognition model may include gesture parameters, shape parameters and displacement parameters. Wherein, the attitude parameter is used to indicate the rotation angle of the target object relative to the camera and the angle between each joint of the target object itself. The shape parameter is used to indicate the three-dimensional shape of the target object. The displacement parameter is used to indicate the range occupied by the target object in the target area, and the offset of the target object in the target area (for example, taking the center point of the target area as a reference point, the degree to which the target object shifts to the left or to degree of right offset). The attitude of the target object can be accurately obtained by calculating based on the attitude parameters, shape parameters and displacement parameters.

In a possible implementation, the method further includes: performing normalization processing on the position information of the target area in the target image to obtain the normalized position information; Processing the position information of the target area in the target image to obtain the pose parameters includes: processing the target area of the target image and the normalized position information through a pose recognition model to obtain the pose parameters. In the aforementioned implementation, the gesture recognition model can be implemented based on the normalized position information when performing the gesture recognition operation. Since the processing of the normalized position information is less difficult, it is beneficial to reduce the calculation amount of the gesture recognition model and reduce The design cost of the model.

In a possible implementation manner, the target object is a human body.

The second aspect of the embodiment of the present application provides a gesture recognition method, the method comprising:

After the target image is obtained, the position information of the pixels in the target image can be obtained, and the target image and the position information of the pixels in the target image can be input into the gesture recognition model, so that the target image and the pixel points in the target image can be detected by the gesture recognition model. The position information is processed to obtain attitude parameters, and the attitude parameters are used to obtain the attitude of the target object included in the target image.

It can be seen from the above method that after the target image is acquired, the target image and the position information of the pixels in the target image can be input to the gesture recognition model, so as to perform the target image and the position information of the pixels in the target image through the gesture recognition model. After processing, the posture parameters are obtained, so the posture of the target object presented by the target image can be obtained based on the posture parameters. In the aforementioned process, the input of the gesture recognition model not only includes the target image, but also includes the position information of the pixels in the target image. Therefore, when the gesture recognition model performs image processing, it not only considers the image information of the target image itself to the pose of the target object. The impact caused by this method also takes into account the influence of the position information of the pixel points in the target image on the attitude of the target object. The factors considered are relatively comprehensive, so the attitude of the target object obtained based on this method has high accuracy. degree, which in turn helps to improve the accuracy of the acquired motion information of the target image.

In a possible implementation manner, the position information includes the coordinates of the pixel points in the image coordinate system, and the image coordinate system is constructed based on the target image. In the aforementioned implementation, a certain vertex of the target image can be used as the origin of the image coordinate system, then the coordinates of all pixels in the target image in the image coordinate system can be determined, and this information can be used as the input of the gesture recognition model, so that , the input of the gesture recognition model not only includes each pixel of the target image, but also includes the coordinates of each pixel in the target image in the image coordinate system. Then, after inputting the information of the target area into the pose recognition model, the pose recognition model can effectively consider the influence of the position information of all pixels in the target area on the pose of the target object when performing image processing on the target area, Therefore, the posture of the target object obtained based on these information has high accuracy.

In a possible implementation, the pose recognition model is trained based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real projection result of the real pose of the target object on the image to be processed to obtain .

In a possible implementation manner, the pose of the target object includes the orientation of the target object in the camera coordinate system and the body behavior of the target object in the camera coordinate system, and the camera coordinate system is constructed based on the camera that captures the target image.

In a possible implementation, the method further includes: processing the target image and the position information of the pixels in the target image through the gesture recognition model to obtain shape parameters and displacement parameters, and the posture parameters, shape parameters and displacement parameters are used in common to obtain the pose of the target object.

In a possible implementation, the method further includes: performing normalization processing on the position information of the pixels in the target image to obtain the normalized position information; Processing the position information of the point to obtain the attitude parameters includes: processing the target image and the normalized position information through the attitude recognition model to obtain the attitude parameters.

In a possible implementation manner, the target object is a human body.

The third aspect of the embodiment of the present application provides a model training method, the method includes: acquiring the image to be processed; processing the target area of the image to be processed by the model to be trained and the position information of the target area in the image to be processed, to obtain Attitude parameters, the target area is the area where the target object is located; based on the attitude parameters, the predicted attitude of the target object is obtained; based on the predicted attitude of the target object and the real attitude of the target object, the training model is trained to obtain the attitude recognition model.

The gesture recognition model obtained by the above method can make the gesture recognition model perceive the position of the target object in the whole image, and learn the relationship between the position and the pose of the target object, so as to capture the pose of the target object more accurately. When the gesture recognition model is in image processing, the input of the gesture recognition model not only includes the cropped target area, but also includes the position information of the target area in the target image, so the gesture recognition model not only considers the image information of the target area itself. The influence caused by the posture of the target object also takes into account the influence of the position information of the target area in the target image on the posture of the target object. The factors considered are relatively comprehensive, so the posture of the target object obtained based on this method, It has higher accuracy, which is beneficial to improve the accuracy of the acquired motion information of the target image.

In a possible implementation, the model to be trained is trained based on the predicted pose of the target object and the real pose of the target object, and the gesture recognition model obtained includes: a predicted projection result based on the predicted pose of the target object on the image to be processed and The real projection result of the real pose of the target object on the image to be processed is used to train the model to be trained to obtain a pose recognition model.

In a possible implementation manner, the location information includes the coordinates of the center point of the target area in the image coordinate system and the size of the target area, and the image coordinate system is constructed based on the image to be processed.

In a possible implementation manner, the location information includes coordinates of vertices of the target area in an image coordinate system, and the image coordinate system is constructed based on the image to be processed.

In a possible implementation, the predicted pose of the target object includes the predicted orientation of the target object in the camera coordinate system and the predicted body behavior of the target object in the camera coordinate system, and the camera coordinate system is constructed based on the camera that captures the image to be processed.

In a possible implementation, the method includes: processing the target area of the image to be processed by the model to be trained and the position information of the target area in the image to be processed to obtain shape parameters and displacement parameters, attitude parameters, shape parameters and The displacement parameters are collectively used to obtain the pose of the target object.

In a possible implementation, the method further includes: performing normalization processing on the position information of the target area in the image to be processed to obtain the normalized position information; And processing the position information of the target area in the image to be processed to obtain the attitude parameters includes: processing the target area of the image to be processed by the model to be trained and the normalized position information to obtain the attitude parameters.

In a possible implementation manner, the target object is a human body.

The fourth aspect of the embodiments of the present application provides a model training method, the method comprising: acquiring an image to be processed;

The attitude parameters are obtained by processing the image to be processed and the position information of the pixels in the image to be processed by the model to be trained; based on the attitude parameters, the predicted attitude of the target object is obtained; based on the predicted attitude of the target object and the real attitude of the target object, the attitude parameter The model is trained to obtain a gesture recognition model.

The gesture recognition model obtained by the above method can make the gesture recognition model perceive the position of the target object in the whole image, and learn the relationship between the position and the pose of the target object, so as to capture the pose of the target object more accurately. When the pose recognition model is in image processing, its input not only includes the target image, but also includes the position information of the pixels in the target image, so the pose recognition model not only considers the influence of the image information of the target image itself on the pose of the target object , it also takes into account the impact of the position information of the pixel points in the target image on the attitude of the target object, and the factors considered are relatively comprehensive, so the attitude of the target object obtained based on this method has high accuracy, and further has It is beneficial to improve the accuracy of the acquired motion information of the target image.

In a possible implementation, based on the predicted pose of the target object and the real pose of the target object, the model to be trained is trained to obtain a pose recognition model: based on the predicted projection result of the predicted pose of the target object on the image to be processed and the target The real projection result of the real pose of the object on the image to be processed is used to train the model to be trained to obtain a pose recognition model.

In a possible implementation manner, the location information includes the coordinates of the pixel points in the image coordinate system, and the image coordinate system is constructed based on the image to be processed.

In a possible implementation, the method further includes: processing the target area of the image to be processed by the model to be trained and the position information of the pixels in the image to be processed to obtain attitude parameters, shape parameters and displacement parameters, attitude parameters, The shape parameter and the displacement parameter are jointly used to obtain the pose of the target object.

In a possible implementation, the method further includes: performing normalization processing on the position information of the pixels in the image to be processed to obtain the normalized position information; Processing the position information of the pixels to obtain the attitude parameters includes: processing the image to be processed and the normalized position information by the model to be trained to obtain the attitude parameters.

In a possible implementation manner, the target object is a human body.

The fifth aspect of the embodiment of the present application provides a gesture recognition device, the device includes: an acquisition module, used to acquire a target image; The position information in is processed to obtain the attitude parameters, the target area is the area where the target object is located, and the attitude parameters are used to obtain the attitude of the target object.

It can be seen from the above device that after the target image is acquired, the target area where the target object is located in the target image and the position information of the target area in the target image can be input into the gesture recognition model, so that the target image of the target image can be detected by the gesture recognition model. The position information of the area and the target area in the target image is processed to obtain the pose parameters, so the pose of the target object can be obtained based on the pose parameters. In the aforementioned process, the input of the gesture recognition model not only includes the cropped target area, but also includes the position information of the target area in the target image, so the gesture recognition model not only considers the image information of the target area itself when performing image processing The impact on the posture of the target object also takes into account the influence of the position information of the target area in the target image on the posture of the target object. The factors considered are relatively comprehensive, so the posture of the target object obtained based on this method , has higher accuracy, which is beneficial to improve the accuracy of the acquired motion information of the target image.

In a possible implementation manner, the location information includes the coordinates of the center point of the target area in the image coordinate system and the size of the target area, and the image coordinate system is constructed based on the target image.

In a possible implementation manner, the location information includes coordinates of vertices of the target area in an image coordinate system, and the image coordinate system is constructed based on the target image.

In a possible implementation manner, the pose recognition model is obtained by training based on a predicted projection result of the predicted pose of the target object on the image to be processed and a real projection result of the real pose of the target object on the image to be processed.

In a possible implementation, the processing module is also used to process the target area of the target image and the position information of the target area in the target image through the pose recognition model to obtain shape parameters and displacement parameters, pose parameters, shape parameters Together with the displacement parameter, it is used to obtain the pose of the target object.

In a possible implementation manner, the device further includes: a normalization module, configured to perform normalization processing on the position information of the target area in the target image to obtain normalized position information; The target area of the target image and the normalized position information are processed by the gesture recognition model to obtain the gesture parameters.

In a possible implementation manner, the target object is a human body.

The sixth aspect of the embodiment of the present application provides a gesture recognition device, which includes: an acquisition module, used to acquire a target image; a processing module, used to use a gesture recognition model to process the target image and the position information of the pixels in the target image Processing is performed to obtain attitude parameters, and the attitude parameters are used to obtain the attitude of the target object included in the target image.

It can be seen from the above device that after the target image is acquired, the target image and the position information of the pixels in the target image can be input to the gesture recognition model, so as to perform the target image and the position information of the pixels in the target image through the gesture recognition model. After processing, the posture parameters are obtained, so the posture of the target object presented by the target image can be obtained based on the posture parameters. In the aforementioned process, the input of the gesture recognition model not only includes the target image, but also includes the position information of the pixels in the target image. Therefore, when the gesture recognition model performs image processing, it not only considers the image information of the target image itself to the pose of the target object. The impact caused by this method also takes into account the influence of the position information of the pixel points in the target image on the attitude of the target object. The factors considered are relatively comprehensive, so the attitude of the target object obtained based on this method has high accuracy. degree, which in turn helps to improve the accuracy of the acquired motion information of the target image.

In a possible implementation manner, the position information includes the coordinates of the pixel points in the image coordinate system, and the image coordinate system is constructed based on the target image.

In a possible implementation, the processing module is also used to process the target image and the position information of the pixels in the target image through the gesture recognition model to obtain the shape parameter and the displacement parameter, and the gesture parameter, the shape parameter and the displacement parameter are jointly Used to get the pose of the target object.

In a possible implementation manner, the device further includes: a normalization module, configured to perform normalization processing on the position information of pixels in the target image to obtain normalized position information; a processing module, configured to The target image and the normalized position information are processed by the gesture recognition model to obtain the gesture parameters.

In a possible implementation manner, the target object is a human body.

The seventh aspect of the embodiment of the present application provides a model training device, the device includes: a first acquisition module, used to acquire the image to be processed; a processing module, used to use the model to be trained to target the target area of the image to be processed and the target area Process the position information in the image to be processed to obtain attitude parameters, and the target area is the area where the target object is located;

The second acquisition module is used to acquire the predicted pose of the target object based on the pose parameters; the training module is used to train the model to be trained based on the predicted pose of the target object and the real pose of the target object to obtain a pose recognition model.

The gesture recognition model obtained by the above-mentioned device can make the gesture recognition model perceive the position of the target object in the whole picture, and learn the relationship between the position and the posture of the target object, so as to capture the posture of the target object more accurately. When the gesture recognition model is in image processing, the input of the gesture recognition model not only includes the cropped target area, but also includes the position information of the target area in the target image, so the gesture recognition model not only considers the image information of the target area itself. The influence caused by the posture of the target object also takes into account the influence of the position information of the target area in the target image on the posture of the target object. The factors considered are relatively comprehensive, so the posture of the target object obtained based on this method, It has higher accuracy, which is beneficial to improve the accuracy of the acquired motion information of the target image.

In a possible implementation, the training module is configured to train the model to be trained based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real projection result of the real pose of the target object on the image to be processed , to get the gesture recognition model.

In a possible implementation, the processing module is also used to process the target area of the image to be processed and the position information of the target area in the image to be processed through the pose recognition model to obtain shape parameters and displacement parameters, pose parameters, shape The parameters and the displacement parameters are used together to obtain the pose of the target object.

In a possible implementation manner, the device further includes: a normalization module, configured to perform normalization processing on the position information of the target area in the image to be processed to obtain normalized position information; a processing module, It is used to process the target area of the image to be processed by the model to be trained and the normalized position information to obtain the pose parameters.

In a possible implementation manner, the target object is a human body.

The eighth aspect of the embodiment of the present application provides a model training device, the device includes: a first acquisition module, used to acquire the image to be processed; a processing module, used to use the model to be trained to process the image to be processed and the pixels in the image to be processed The position information of the point is processed to obtain the attitude parameters; the second acquisition module is used to obtain the predicted attitude of the target object based on the attitude parameters; the training module is used to treat the training model based on the predicted attitude of the target object and the real attitude of the target object Perform training to obtain a gesture recognition model.

The gesture recognition model obtained by the above-mentioned device can make the gesture recognition model perceive the position of the target object in the whole picture, and learn the relationship between the position and the posture of the target object, so as to capture the posture of the target object more accurately. When the pose recognition model is in image processing, its input not only includes the target image, but also includes the position information of the pixels in the target image, so the pose recognition model not only considers the influence of the image information of the target image itself on the pose of the target object , it also takes into account the influence of the position information of the pixel points in the target image on the attitude of the target object, and the factors considered are relatively comprehensive. Therefore, the attitude of the target object obtained based on this method has high accuracy, and further has It is beneficial to improve the accuracy of the acquired motion information of the target image.

In a possible implementation, the processing module is also used to process the target area of the image to be processed and the position information of the pixels in the image to be processed through the pose recognition model to obtain shape parameters and displacement parameters, pose parameters, shape parameters Together with the displacement parameter, it is used to obtain the pose of the target object.

In a possible implementation manner, the device further includes: a normalization module, configured to perform normalization processing on the position information of pixels in the image to be processed, to obtain normalized position information; a processing module, configured to The attitude parameters are obtained by processing the image to be processed and the normalized position information by the model to be trained.

In a possible implementation manner, the target object is a human body.

The ninth aspect of the embodiment of the present application provides a gesture recognition device, the device includes a memory and a processor; the memory stores codes, the processor is configured to execute the codes, and when the codes are executed, the gesture recognition device performs the first aspect, any one possible implementation manner of the first aspect, the second aspect, or the method described in any one possible implementation manner of the second aspect.

The tenth aspect of the embodiments of the present application provides a model training device, the device includes a memory and a processor; the memory stores codes, the processor is configured to execute the codes, and when the codes are executed, the model training device executes the third aspect, any one possible implementation manner of the third aspect, the fourth aspect, or the method described in any one possible implementation manner of the fourth aspect.

An eleventh aspect of the embodiments of the present application provides a circuit system, where the circuit system includes a processing circuit configured to perform any one of the possible implementations in the first aspect, the first aspect, the second aspect, Any one of the possible implementations of the second aspect, the third aspect, any one of the possible implementations of the third aspect, the fourth aspect, or the method described in any one of the possible implementations of the fourth aspect.

The twelfth aspect of the embodiments of the present application provides a chip system, the chip system includes a processor, used to call the computer program or computer instruction stored in the memory, so that the processor executes the first aspect, the first aspect Any possible implementation of the second aspect, any possible implementation of the second aspect, the third aspect, any possible implementation of the third aspect, the fourth aspect or any of the fourth aspects The method described in one possible implementation.

In a possible implementation manner, the processor is coupled to the memory through an interface.

In a possible implementation manner, the chip system further includes a memory, where computer programs or computer instructions are stored.

The thirteenth aspect of the embodiments of the present application provides a computer storage medium, the computer storage medium stores a computer program, and when the program is executed by a computer, the computer implements any one of the possibilities in the first aspect and the first aspect. implementation of the second aspect, any possible implementation of the second aspect, the third aspect, any possible implementation of the third aspect, the fourth aspect, or any possible implementation of the fourth aspect method described in the method.

The fourteenth aspect of the embodiments of the present application provides a computer program product, the computer program product stores instructions, and when the instructions are executed by a computer, the computer implements any one of the possible functions of the first aspect and the first aspect. Implementation, the second aspect, any possible implementation of the second aspect, the third aspect, any possible implementation of the third aspect, the fourth aspect, or any possible implementation of the fourth aspect the method described.

In the embodiment of the present application, after the target image is acquired, the target area where the target object is located in the target image and the position information of the target area in the target image can be input into the gesture recognition model, so that the target area of the target image can be identified by the gesture recognition model. And the position information of the target area in the target image is processed to obtain the pose parameters, so the pose of the target object can be obtained based on the pose parameters. In the aforementioned process, the input of the gesture recognition model not only includes the cropped target area, but also includes the position information of the target area in the target image, so the gesture recognition model not only considers the image information of the target area itself when performing image processing The impact on the posture of the target object also takes into account the influence of the position information of the target area in the target image on the posture of the target object. The factors considered are relatively comprehensive, so the posture of the target object obtained based on this method , has higher accuracy, which is beneficial to improve the accuracy of the acquired motion information of the target image.

Description of drawings

Fig. 1 is a schematic diagram of related technology;

Fig. 2 is another schematic diagram of related technology;

Fig. 3 is a kind of structural schematic diagram of main frame of artificial intelligence;

FIG. 4a is a schematic structural diagram of an image processing system provided by an embodiment of the present application;

Fig. 4b is another schematic structural diagram of the image processing system provided by the embodiment of the present application;

FIG. 4c is a schematic diagram of related equipment for image processing provided by the embodiment of the present application;

FIG. 5 is a schematic diagram of the architecture of the system 100 provided by the embodiment of the present application;

FIG. 6 is a schematic flowchart of a gesture recognition method provided in an embodiment of the present application;

FIG. 7 is a schematic diagram of an application example of gesture recognition provided by the embodiment of the present application;

FIG. 8 is another schematic flowchart of the gesture recognition method provided by the embodiment of the present application;

Fig. 9a is a schematic diagram of another application example of gesture recognition provided by the embodiment of the present application;

Fig. 9b is a schematic structural diagram of the gesture recognition model provided by the embodiment of the present application;

Fig. 10 is a schematic diagram of the model training method provided by the embodiment of the present application;

Fig. 11 is a schematic diagram of the projection result provided by the embodiment of the present application;

FIG. 12 is another schematic diagram of the model training method provided by the embodiment of the present application;

FIG. 13 is a schematic structural diagram of a gesture recognition device provided by an embodiment of the present application;

Fig. 14 is another schematic structural diagram of the gesture recognition device provided by the embodiment of the present application;

Fig. 15 is another structural schematic diagram of the model training device provided by the embodiment of the present application;

Fig. 16 is another structural schematic diagram of the model training device provided by the embodiment of the present application;

Fig. 17 is a schematic structural diagram of the execution device provided by the embodiment of the present application;

FIG. 18 is a schematic structural diagram of a training device provided in an embodiment of the present application;

FIG. 19 is a schematic structural diagram of a chip provided by an embodiment of the present application.

Detailed ways

The terms "first", "second" and the like in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific sequence or sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances, and this is merely a description of the manner in which objects with the same attribute are described in the embodiments of the present application. Furthermore, the terms "comprising" and "having", and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, product, or apparatus comprising a series of elements is not necessarily limited to those elements, but may include elements not expressly included. Other elements listed explicitly or inherent to the process, method, product, or apparatus.

Human body motion capture technology is a commonly used technology in the film and television and game industries. This technology can process the input video stream to capture the posture of the human body in each frame of image, so as to obtain the motion information of the human body. Based on this technology, the posture data of human body movement can be extracted from massive videos, and has a wide range of application scenarios, for example, virtual characters in augmented reality (augmented reality, AR) scenes and virtual reality (virtual reality, VR) scenes drives, teleconferencing, and the metaverse, to name a few.

As shown in FIG. 1 (FIG. 1 is a schematic diagram of the related art), in the related art, if there are multiple images presenting the same environment, they are Image 1, Image 2, and Image 3 respectively. In the three images, the human body is located in three positions in the environment, and when they are located in different positions, the postures of the human body are extremely similar. In this case, the neural network will determine that the posture parameters of the human body in the three images are consistent, that is, it is determined that the posture 1 of the human body in image 1, the posture 2 of the human body in image 2, and the posture 3 of the human body in image 3 are the same of. But in the actual environment, once the human body moves, although the postures of the human body in different positions are similar, there must be subtle differences between the postures of the human body, as shown in Figure 2 (Figure 2 is another schematic diagram of related technology) , observe the three positions of the human body in the environment from the perspective of looking down. For the posture 1 of the human body, the orientation of the human body is facing the camera sideways (tilted to the right). For the posture 2 of the human body, the orientation of the human body is It is directly facing the camera. For the posture 3 of the human body, the orientation of the human body is facing the camera sideways (leaning to the left). It can be seen that the difference between these three postures cannot be recognized by the neural network in the related technology, which leads to the inaccurate captured posture of the human body, which in turn affects the accuracy of the motion information of the human body.

In order to solve the above problems, an embodiment of the present application provides a gesture recognition method, which can be implemented in combination with artificial intelligence (AI) technology. AI technology is a technical discipline that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence. AI technology obtains the best results by perceiving the environment, acquiring knowledge and using knowledge. In other words, artificial intelligence technology is a branch of computer science that attempts to understand the nature of intelligence and produce a new kind of intelligent machine that can respond in a similar way to human intelligence. Image processing using artificial intelligence is a common application of artificial intelligence.

First, describe the overall workflow of the artificial intelligence system. Please refer to Figure 3. Figure 3 is a schematic structural diagram of the main framework of artificial intelligence. The following is from the "intelligent information chain" (horizontal axis) and "IT value chain" (vertical axis) Two dimensions are used to elaborate the above artificial intelligence theme framework. Among them, the "intelligent information chain" reflects a series of processes from data acquisition to processing. For example, it can be the general process of intelligent information perception, intelligent information representation and formation, intelligent reasoning, intelligent decision-making, intelligent execution and output. In this process, data has undergone a condensed process of "data-information-knowledge-wisdom". The "IT value chain" reflects the value brought by artificial intelligence to the information technology industry from the underlying infrastructure of artificial intelligence, information (provided and processed by technology) to the industrial ecological process of the system.

(1) Infrastructure

The infrastructure provides computing power support for the artificial intelligence system, realizes communication with the outside world, and realizes support through the basic platform. Communicate with the outside through sensors; computing power is provided by smart chips (CPU, NPU, GPU, ASIC, FPGA and other hardware acceleration chips); the basic platform includes distributed computing framework and network and other related platform guarantees and supports, which can include cloud storage and Computing, interconnection network, etc. For example, sensors communicate with the outside to obtain data, and these data are provided to the smart chips in the distributed computing system provided by the basic platform for calculation.

(2) data

Data from the upper layer of the infrastructure is used to represent data sources in the field of artificial intelligence. The data involves graphics, images, voice, text, and IoT data of traditional equipment, including business data of existing systems and sensory data such as force, displacement, liquid level, temperature, and humidity.

(3) Data processing

Data processing usually includes data training, machine learning, deep learning, search, reasoning, decision-making, etc.

Among them, machine learning and deep learning can symbolize and formalize intelligent information modeling, extraction, preprocessing, training, etc. of data.

Reasoning refers to the process of simulating human intelligent reasoning in a computer or intelligent system, and using formalized information to carry out machine thinking and solve problems according to reasoning control strategies. The typical functions are search and matching.

Decision-making refers to the process of decision-making after intelligent information is reasoned, and usually provides functions such as classification, sorting, and prediction.

(4) General ability

After the above-mentioned data processing is performed on the data, some general capabilities can be formed based on the results of data processing, such as algorithms or a general system, such as translation, text analysis, computer vision processing, speech recognition, image processing identification, etc.

(5) Smart products and industry applications

Intelligent products and industry applications refer to the products and applications of artificial intelligence systems in various fields. It is the packaging of the overall solution of artificial intelligence, which commercializes intelligent information decision-making and realizes landing applications. Its application fields mainly include: intelligent terminals, intelligent transportation, Smart healthcare, autonomous driving, smart cities, etc.

Next, several application scenarios of this application are introduced.

Fig. 4a is a schematic structural diagram of an image processing system provided by an embodiment of the present application, and the image processing system includes a user device and a data processing device. Wherein, the user equipment includes smart terminals such as a mobile phone, a personal computer, or an information processing center. The user equipment is the initiator of the image processing, and as the initiator of the image processing request, usually the user initiates the request through the user equipment.

The above-mentioned data processing device may be a device or server having a data processing function such as a cloud server, a network server, an application server, and a management server. The data processing device receives the image processing request from the intelligent terminal through the interactive interface, and then performs image processing such as machine learning, deep learning, search, reasoning, and decision-making through the memory for storing data and the processor link of data processing. The storage in the data processing device can be a general term, including local storage and databases for storing historical data. The database can be on the data processing device or on other network servers.

In the image processing system shown in Figure 4a, the user equipment can receive user instructions, for example, the user equipment can obtain an image input/selected by the user, and then initiate a request to the data processing equipment, so that the data processing equipment can obtain the user equipment. An image processing application is performed on the image (for example, recognizing a pose of the target object, etc.), so as to obtain a corresponding processing result for the image. Exemplarily, the user device may acquire an image input by the user, and then initiate a gesture recognition request of the target object to the data processing device, so that the data processing device classifies the image, thereby obtaining the pose parameters of the target object in the image, thereby determining The pose of the target object in the image.

In FIG. 4a, the data processing device can execute the image processing method of the embodiment of the present application.

Fig. 4b is another schematic structural diagram of the image processing system provided by the embodiment of the present application. In Fig. 4b, the user equipment is directly used as a data processing equipment, and the user equipment can directly obtain the input from the user and directly perform the processing by the hardware of the user equipment itself. For processing, the specific process is similar to that in FIG. 4a , and reference may be made to the above description, and details are not repeated here.

In the image processing system shown in FIG. 4b, the user equipment may receive an instruction from the user, for example, the user equipment may acquire an image selected by the user in the user equipment, and then the user equipment itself executes an image processing application ( For example, the recognition gesture of the target object, etc.), so as to obtain the corresponding processing results for the image.

In FIG. 4b, the user equipment itself can execute the gesture recognition method of the embodiment of the present application.

Fig. 4c is a schematic diagram of related equipment for image processing provided by the embodiment of the present application.

The above-mentioned user equipment in FIG. 4a and FIG. 4b may specifically be the local device 301 or local device 302 in FIG. 4c, and the data processing device in FIG. 4a may specifically be the execution device 210 in FIG. To store the data to be processed by the execution device 210, the data storage system 250 may be integrated on the execution device 210, or set on the cloud or other network servers.

The processors in Figure 4a and Figure 4b can perform data training/machine learning/deep learning through a neural network model or other models (for example, a model based on a support vector machine), and use the data to finally train or learn the model for image execution Image processing application, so as to obtain the corresponding processing results.

FIG. 5 is a schematic diagram of the architecture of the system 100 provided by the embodiment of the present application. In FIG. 5, the execution device 110 is configured with an input/output (I/O) interface 112 for data interaction with external devices, and the user Data can be input to the I/O interface 112 through the client device 140, and the input data in this embodiment of the application may include: various tasks to be scheduled, callable resources, and other parameters.

When the execution device 110 preprocesses the input data, or when the calculation module 111 of the execution device 110 executes calculations and other related processing (such as implementing the function of the neural network in this application), the execution device 110 can call the data storage system 150 The data, codes, etc. in the system can be used for corresponding processing, and the data, instructions, etc. obtained by corresponding processing can also be stored in the data storage system 150 .

Finally, the I/O interface 112 returns the processing result to the client device 140, thereby providing it to the user.

It is worth noting that the training device 120 can generate corresponding target models/rules based on different training data for different goals or different tasks, and the corresponding target models/rules can be used to achieve the above-mentioned goals or complete the above-mentioned tasks , giving the user the desired result. Wherein, the training data can be stored in the database 130, and come from the training samples collected by the data collection device 160.

In the situation shown in FIG. 5 , the user can manually specify the input data, and the manual specification can be operated through the interface provided by the I/O interface 112 . In another case, the client device 140 can automatically send the input data to the I/O interface 112 . If the client device 140 is required to automatically send the input data to obtain the user's authorization, the user can set the corresponding authority in the client device 140 . The user can view the results output by the execution device 110 on the client device 140, and the specific presentation form may be specific ways such as display, sound, and action. The client device 140 can also be used as a data collection terminal, collecting the input data input to the I/O interface 112 as shown in the figure and the output results of the output I/O interface 112 as new sample data, and storing them in the database 130 . Of course, it is also possible not to collect through the client device 140, but the I/O interface 112 directly uses the input data input to the I/O interface 112 as shown in the figure and the output result of the output I/O interface 112 as a new sample The data is stored in database 130 .

It should be noted that FIG. 5 is only a schematic diagram of a system architecture provided by the embodiment of the present application, and the positional relationship between devices, devices, modules, etc. shown in the figure does not constitute any limitation. For example, in FIG. 5, the data The storage system 150 is an external memory relative to the execution device 110 , and in other cases, the data storage system 150 may also be placed in the execution device 110 . As shown in FIG. 5 , the neural network can be obtained by training according to the training device 120 .

An embodiment of the present application also provides a chip, the chip includes a neural network processor (NPU). The chip can be set in the execution device 110 shown in FIG. 5 to complete the computing work of the computing module 111 . The chip can also be set in the training device 120 shown in FIG. 5 to complete the training work of the training device 120 and output the target model/rule.

The neural network processor NPU is mounted on the main central processing unit (central processing unit, CPU) (host CPU) as a coprocessor, and the main CPU assigns tasks. The core part of the NPU is the operation circuit, and the controller controls the operation circuit to extract the data in the memory (weight memory or input memory) and perform operations.

In some implementations, the operation circuit includes multiple processing units (process engine, PE). In some implementations, the arithmetic circuit is a two-dimensional systolic array. The arithmetic circuit may also be a one-dimensional systolic array or other electronic circuitry capable of performing mathematical operations such as multiplication and addition. In some implementations, the arithmetic circuit is a general purpose matrix processor.

For example, suppose there is an input matrix A, a weight matrix B, and an output matrix C. The operation circuit fetches the data corresponding to the matrix B from the weight memory, and caches it on each PE in the operation circuit. The operation circuit takes the data of matrix A from the input memory and performs matrix operation with matrix B, and the obtained partial or final results of the matrix are stored in the accumulator.

The vector calculation unit can further process the output of the operation circuit, such as vector multiplication, vector addition, exponential operation, logarithmic operation, size comparison and so on. For example, the vector computing unit can be used for network calculations of non-convolution/non-FC layers in neural networks, such as pooling, batch normalization, local response normalization, etc.

In some implementations, the vector computation unit can store the processed output vectors to a unified register. For example, a vector computation unit may apply a non-linear function to the output of the arithmetic circuit, such as a vector of accumulated values, to generate activation values. In some implementations, the vector computation unit generates normalized values, merged values, or both. In some implementations, the vector of processed outputs can be used as an activation input to an operational circuit, for example for use in a subsequent layer in a neural network.

Unified memory is used to store input data and output data.

The weight data directly transfers the input data in the external memory to the input memory and/or the unified memory through the storage unit access controller (direct memory access controller, DMAC), stores the weight data in the external memory into the weight memory, and stores the weight data in the unified memory Store the data in the external memory.

The bus interface unit (bus interface unit, BIU) is used to realize the interaction between the main CPU, DMAC and instruction fetch memory through the bus.

The instruction fetch buffer connected to the controller is used to store the instructions used by the controller;

The controller is used for invoking instructions cached in the memory to control the working process of the computing accelerator.

Generally, the unified memory, the input memory, the weight memory and the instruction fetch memory are all on-chip (On-Chip) memory, and the external memory is the memory outside the NPU, and the external memory can be a double data rate synchronous dynamic random access memory (double data rate synchronous dynamic random access memory, DDR SDRAM), high bandwidth memory (high bandwidth memory, HBM) or other readable and writable memory.

Since the embodiment of the present application involves the application of a large number of neural networks, in order to facilitate understanding, the following first introduces related terms and neural network related concepts involved in the embodiment of the present application.

(1) neural network

The neural network can be composed of neural units, and the neural unit can refer to an operation unit that takes xs and intercept 1 as input, and the output of the operation unit can be:

Wherein, s=1, 2, ... n, n is a natural number greater than 1, Ws is the weight of xs, and b is the bias of the neuron unit. f is the activation function of the neural unit, which is used to introduce nonlinear characteristics into the neural network to convert the input signal in the neural unit into an output signal. The output signal of this activation function can be used as the input of the next convolutional layer. The activation function may be a sigmoid function. A neural network is a network formed by connecting many of the above-mentioned single neural units, that is, the output of one neural unit can be the input of another neural unit. The input of each neural unit can be connected with the local receptive field of the previous layer to extract the features of the local receptive field. The local receptive field can be an area composed of several neural units.

The work of each layer in the neural network can be described by the mathematical expression y=a(Wx+b): From the physical level, the work of each layer in the neural network can be understood as through five kinds of input space (input vector set) to complete the transformation from the input space to the output space (that is, the row space of the matrix to the column space), these five operations include: 1. Dimension enhancement/dimension reduction; Translation; 5. "Bending". Among them, the operations of 1, 2, and 3 are completed by Wx, the operation of 4 is completed by +b, and the operation of 5 is realized by a(). The reason why the word "space" is used here is because the classified object is not a single thing, but a kind of thing, and space refers to the collection of all individuals of this kind of thing. Wherein, W is a weight vector, and each value in the vector represents the weight value of a neuron in this layer of neural network. The vector W determines the space transformation from the input space to the output space described above, that is, the weight W of each layer controls how to transform the space. The purpose of training the neural network is to finally obtain the weight matrix of all layers of the trained neural network (the weight matrix formed by the vector W of many layers). Therefore, the training process of the neural network is essentially to learn the way to control the space transformation, and more specifically, to learn the weight matrix.

Because we want the output of the neural network to be as close as possible to the value we really want to predict, we can compare the predicted value of the current network with the target value we really want, and then update each layer of neural network according to the difference between the two (Of course, there is usually an initialization process before the first update, that is, to pre-configure parameters for each layer in the neural network). For example, if the network's predicted value is high, adjust the weight vector to make it predict low Some, keep adjusting until the neural network can predict the desired target value. Therefore, it is necessary to pre-define "how to compare the difference between the predicted value and the target value", which is the loss function (loss function) or objective function (objective function), which is used to measure the difference between the predicted value and the target value important equation. Among them, taking the loss function as an example, the higher the output value (loss) of the loss function, the greater the difference, so the training of the neural network becomes a process of reducing the loss as much as possible.

(2) Back propagation algorithm

The neural network can use the error back propagation (back propagation, BP) algorithm to correct the size of the parameters in the initial neural network model during the training process, so that the reconstruction error loss of the neural network model becomes smaller and smaller. Specifically, passing the input signal forward until the output will generate an error loss, and updating the parameters in the initial neural network model by backpropagating the error loss information, so that the error loss converges. The backpropagation algorithm is a backpropagation movement dominated by error loss, aiming to obtain the optimal parameters of the neural network model, such as the weight matrix.

The method provided by this application is described below from the training side of the neural network and the application side of the neural network.

The model training method provided in the embodiment of the present application involves image processing, and can be specifically applied to data processing methods such as data training, machine learning, and deep learning to symbolize and form the training data (such as the image to be processed in this application) intelligent information modeling, extraction, preprocessing, training, etc., and finally a trained neural network (such as the gesture recognition model in this application); and, the gesture recognition method provided in the embodiment of this application can use the above-mentioned trained A neural network, inputting input data (such as the target image in the present application, the target area of the target image) into the trained neural network to obtain output data (such as the pose parameters in the present application, etc.). It should be noted that the model training method and gesture recognition method provided in the embodiment of the present application are inventions based on the same idea, and can also be understood as two parts in a system, or two stages of an overall process: such as model training phase and model application phase.

Fig. 6 is a schematic flow chart of the gesture recognition method provided by the embodiment of the present application. As shown in Fig. 6, the method includes:

601. Acquire a target image.

In this embodiment, when it is necessary to perform gesture recognition on the target image, the target image may be acquired first, and the target image may be a certain frame image in the video stream, or may be a single image. The target image usually presents the target object and the environment in which the target object is located, so the goal of gesture recognition is to obtain the pose of the target object in the target image. For example, as shown in Figure 7 (Figure 7 is a schematic diagram of an application example of gesture recognition provided by the embodiment of the present application), the content presented by the target image is the human body and the environment in which the human body is located, so the gesture recognition target of the target image is Recognize the pose of the human body in the target image.

In order to reduce the computational load of the pose recognition model, the target image can be preprocessed. Specifically, the target image can be detected to determine the area where the target object is located in the target image, and this area can be called a target area (also called a detection frame). Still as in the above example, after the area where the human body is located in the target image is detected, the area where the human body is located can be determined as the target area (that is, the area enclosed by the red frame in FIG. 7 ).

Further, the position information of the target area in the target image can also be obtained, and the position information can have many situations: (1) The position information of the target area in the target image includes the coordinates of the center point of the target area in the image coordinate system As well as the size of the target area, the image coordinate system is constructed based on the target image. Specifically, a certain vertex of the target image (for example, the vertex in the upper left corner of the target image) can be used as the origin of the image coordinate system, then the coordinates of the center point of the target area in the image coordinate system can be determined, as well as the length of the target area and width (which is equivalent to obtaining the size of the target area), these information can be used to indicate the position of the target area in the target image. (2) The position information of the target area in the target image includes the coordinates of the vertices of the target area in the image coordinate system, and the image coordinate system is constructed based on the target image. Specifically, a certain vertex of the target image (for example, the vertex in the upper left corner of the target image) can be used as the origin of the image coordinate system, then the coordinates of all vertices of the target area in the image coordinate system can be determined, and this information can be used to indicate The position of the target area in the target image and so on.

Furthermore, after obtaining the position information of the target area in the target image, the position information may also be normalized to obtain the normalized position information of the target area in the target image. Specifically, assuming that the position information of the target area in the target image is I, and the position information of the target area in the target image after normalization processing is I', then, I'=(I-mean(I))/F , F=sqrt(w*w+h*h) Wherein, mean (I) is the mean value of the target area, w is the width of the target area, and h is the length of the target area. In this way, after obtaining the normalized information, it is beneficial to realize the posture recognition operation of the subsequent model.

602. Process the target area of the target image and the position information of the target area in the target image through a gesture recognition model to obtain a pose parameter, where the target area is the area where the target object is located, and the pose parameter is used to obtain the pose of the target object.

After obtaining the target area and the normalized position information of the target area in the target image, the target area and the normalized position information of the target area in the target image can be input to the pose recognition model (for the trained neural network Network), to process the target area of the target image and the position information of the normalized target area in the target image through the gesture recognition model (for example, a series of operations such as feature extraction) to obtain the pose parameters. Then, the pose of the target object in the target image can be obtained based on the pose parameters.

It should be noted that the pose recognition model can output not only pose parameters, but also shape parameters and displacement parameters. Among them, the posture parameters are usually composed of two parts of parameters, one part of parameters is used to indicate the rotation angle of the target object relative to the camera, and the other part of parameters is used to indicate the angles between the joints of the target object itself. The shape parameter is used to indicate the three-dimensional shape of the target object. The displacement parameter is used to indicate the range occupied by the target object in the target area, and the offset of the target object in the target area (for example, taking the center point of the target area as a reference point, the degree to which the target object shifts to the left or to degree of right offset). Based on the calculation of attitude parameters, shape parameters and displacement parameters, the attitude of the target object can be obtained. The attitude of the target object can be represented by multiple three-dimensional key points (3D key points) of the target object. These three-dimensional key points can describe the position of the target object. The orientation in the camera coordinate system (that is, the orientation of the target object relative to the camera) and the body behavior of the target object in the camera coordinate system (that is, the movement of the target object in three-dimensional space), where the origin of the camera coordinate system is the shooting target image camera. Still as in the above example, as shown in Figure 7, after the pose recognition model outputs pose parameters θ, shape parameters β, and displacement parameters S, T, these parameters can be calculated to obtain multiple 3D key points of the target object. These 3D key points The points are combined to show the orientation of the target object in the camera coordinate system and the body behavior of the target object in the camera coordinate system.

So far, the gesture recognition of the target image is completed. If the target image is a certain frame image in the video stream, for the rest of the frame images in the video stream, the same operation as that performed on the target image can be performed to obtain the continuous multiple postures of the target object and form the motion information of the target object , so as to meet the various application needs of users.

Fig. 8 is another schematic flowchart of the posture recognition method provided by the embodiment of the present application. As shown in Fig. 8, the method includes:

801. Acquire a target image.

In this embodiment, when it is necessary to perform gesture recognition on the target image, the target image may be acquired first, and the target image may be a certain frame image in the video stream, or may be a single image. The target image usually presents the target object and the environment in which the target object is located, so the goal of gesture recognition is to obtain the pose of the target object in the target image. For example, as shown in Figure 9a (Figure 9a is a schematic diagram of another application example of gesture recognition provided by the embodiment of the present application), the content presented by the target image is the human body and the environment in which the human body is located, so the gesture recognition target of the target image In order to recognize the posture presented by the human body in the target image.

Furthermore, the position information of the pixel in the target image can also be obtained, the position information includes the coordinates of the target image and the pixel in the image coordinate system, and the image coordinate system is constructed based on the target image. Specifically, a certain vertex of the target image (for example, the vertex in the upper left corner of the target image) can be used as the origin of the image coordinate system, then the coordinates of all pixels in the target image in the image coordinate system can be determined, and these information can be used as pose The input of the recognition model, so that the input of the pose recognition model not only includes each pixel of the target image, but also includes the coordinates of each pixel in the target image in the image coordinate system.

Furthermore, after obtaining the position information of the pixels in the target image, the position information may also be normalized to obtain the position information of the pixels in the target image after normalization processing. Specifically, assuming that the positional information of the pixel in the target image is I, the positional information of the pixel in the target image after the normalization process is I', then, I'=(I-mean(I))/F, F =sqrt(w×w+h×h) wherein, mean(I) is the mean value of the target image, w is the width of the target image, and h is the length of the target image. In this way, after obtaining the normalized information, it is beneficial to realize the posture recognition operation of the subsequent model.

802. Process the target image and the position information of the pixels in the target image through the gesture recognition model to obtain a gesture parameter, and the gesture parameter is used to acquire the gesture of the target object included in the target image.

After obtaining the target image and the position information of the pixels in the normalized target image, the target image and the position information of the pixels in the normalized target image can be input to the gesture recognition model (a trained neural network) , to process the target area of the target image and the position information of the pixels in the normalized target image through the gesture recognition model (for example, a series of operations such as feature extraction) to obtain the gesture parameters. Then, the pose of the target object in the target image can be obtained based on the pose parameters.

As shown in Figure 9b (Figure 9b is a schematic structural diagram of the gesture recognition model provided by the embodiment of the present application), the gesture recognition model can include two parts, one part is an encoder (encoder), and the other part is a plurality of convolutional layers (convolution ). Among them, after the encoder receives the target image, it can perform feature extraction processing on the target image to obtain a feature image (featrue map), and send the feature image to the convolutional layer. Multiple convolutional layers can perform at least one convolution process on the feature image and the position information (location map) of the target image after normalization to obtain the attitude parameters, so the position of the target object in the target image can be determined based on the attitude parameters. attitude.

For descriptions of step 802, reference may be made to relevant descriptions of step 602 in the embodiment shown in FIG. 6 , and details are not repeated here.

It should be understood that this embodiment is only schematically introduced with the example shown in FIG. 9b, and does not limit the structure of the gesture recognition model in this application. The gesture recognition model can also have other various structures, as long as the gesture recognition function can be realized. Can.

In the embodiment of the present application, after the target image is acquired, the target image and the position information of the pixel points in the target image can be input to the gesture recognition model, so as to process the target image and the position information of the pixel points in the target image through the gesture recognition model , to obtain the pose parameter, so the pose of the target object presented by the target image can be obtained based on the pose parameter. In the aforementioned process, the input of the gesture recognition model not only includes the target image, but also includes the position information of the pixels in the target image. Therefore, when the gesture recognition model performs image processing, it not only considers the image information of the target image itself to the pose of the target object. The impact caused by this method also takes into account the influence of the position information of the pixel points in the target image on the attitude of the target object. The factors considered are relatively comprehensive, so the attitude of the target object obtained based on this method has high accuracy. degree, which in turn helps to improve the accuracy of the acquired motion information of the target image.

In addition, the gesture recognition model provided by the embodiment of the present application can also be compared with the gesture recognition model of the related art, and the comparison results are shown in Table 1 and Table 2:

Table 1

Table 2

Based on Table 1, it can be seen that in the data set 1, compared with other related technologies, the posture recognition model provided by the embodiment of the present application has significantly reduced various error indicators, that is, the posture recognition model provided by the embodiment of the present application has a significant reduction in each error index. All the error indicators have excellent performance.

Based on Table 2, it can be seen that, compared with other related technologies, the posture recognition model provided by the embodiment of the present application has significantly reduced various error indicators in the data set 1 and data set 2, that is, the error index provided by the embodiment of the present application The gesture recognition model also has superior performance in various error indicators.

The above is a detailed description of the gesture recognition method provided by the embodiment of the present application, and the model training method provided by the embodiment of the present application will be introduced below. Fig. 10 is a schematic diagram of the model training method provided by the embodiment of the present application. As shown in Fig. 10, the method includes:

1001. Acquire an image to be processed.

When the model to be trained needs to be trained, a batch of training samples can be obtained, that is, images to be processed for training. It is worth noting that, for the image to be processed, the real pose of the target object in the image to be processed is known.

After the image to be processed is obtained, the image to be processed can be detected and processed to determine the target area in the image to be processed. The location information of the target area in the image to be processed can also be obtained and normalized.

In a possible implementation manner, the target object is a human body.

In a possible implementation manner, the position information of the target area in the image to be processed includes the coordinates of the center point of the target area in the image coordinate system and the size of the target area, and the image coordinate system is constructed based on the image to be processed.

In a possible implementation manner, the position information of the target area in the image to be processed includes coordinates of vertices of the target area in an image coordinate system, and the image coordinate system is constructed based on the image to be processed.

For descriptions of step 1001, reference may be made to relevant descriptions of step 601 in the embodiment shown in FIG. 6 , and details are not repeated here.

1002. Process the target area of the image to be processed by the model to be trained and the position information of the target area in the image to be processed to obtain attitude parameters, where the target area is the area where the target object is located.

After obtaining the position information of the image to be processed and the normalized target area in the image to be processed, the image to be processed and the position information of the normalized target area in the image to be processed can be input to the model to be trained to The pose parameters are obtained by processing the target area of the image to be processed by the model to be trained and the normalized position information of the target area in the image to be processed.

In a possible implementation, the method further includes: processing the target area of the image to be processed and the normalized position information of the target area in the image to be processed by the gesture recognition model to obtain shape parameters and displacement parameters, Pose parameters, shape parameters and displacement parameters are jointly used to obtain the pose of the target object.

For the description of step 1002, reference may be made to the related description of step 602 in the embodiment shown in FIG. 6 , which will not be repeated here.

1003. Acquire a predicted pose of the target object based on the pose parameter.

After the attitude parameters, shape parameters and displacement parameters are obtained, the calculation can be performed based on the attitude parameters, shape parameters and displacement parameters to obtain the predicted attitude of the target object. express.

For the description of step 1003, reference may be made to the related description of step 602 in the embodiment shown in FIG. 6 , and details are not repeated here.

1004. Based on the predicted pose of the target object and the real pose of the target object, train the model to be trained to obtain a pose recognition model.

After obtaining the predicted pose of the target object, since the real pose of the target object is known, the model to be trained can be trained based on the predicted pose of the target object and the real pose of the target object to obtain a pose recognition model.

In a possible implementation, the model to be trained is trained based on the predicted pose of the target object and the real pose of the target object, and the gesture recognition model obtained includes: a predicted projection result based on the predicted pose of the target object on the image to be processed and The real projection result of the real pose of the target object on the image to be processed is used to train the model to be trained to obtain a pose recognition model. Specifically, as shown in Figure 11 (Figure 11 is a schematic diagram of the projection result provided by the embodiment of the present application), since the predicted pose of the target object can be represented by multiple predicted 3D key points of the target object, and the predicted pose of the target object The posture can be represented by multiple real 3D key points of the target object (these real 3D key points can be marked in advance and are known), so multiple predicted 3D key points of the target object can be projected onto the image to be processed, Obtain multiple predicted two-dimensional key points of the target object (that is, the predicted projection result of the predicted pose of the target object on the image to be processed), and similarly, multiple real three-dimensional key points of the target object can also be projected onto the image to be processed , to obtain multiple real two-dimensional key points of the target object (that is, the real projection result of the real pose of the target object on the image to be processed). Then, the target loss can be calculated based on multiple predicted 2D key points and multiple real 2D key points of the target object, and the target loss is used to indicate multiple predicted 2D key points and multiple real 2D key points of the target object difference between. Then, the model parameters of the model to be trained can be updated based on the target loss, and the next batch of training samples can be used to train the model to be trained after the updated parameters (that is, re-execute steps 1002 to 1004), until the model training conditions are met (such as , the target loss reaches convergence, etc.), the gesture recognition model in the embodiment shown in FIG. 6 can be obtained.

The gesture recognition model trained in the embodiment of the present application can make the gesture recognition model perceive the position of the target object in the whole picture, and learn the relationship between the position and the pose of the target object, so as to more accurately capture the pose of the target object. When the gesture recognition model is in image processing, the input of the gesture recognition model not only includes the cropped target area, but also includes the position information of the target area in the target image, so the gesture recognition model not only considers the image information of the target area itself. The influence caused by the posture of the target object also takes into account the influence of the position information of the target area in the target image on the posture of the target object. The factors considered are relatively comprehensive, so the posture of the target object obtained based on this method, It has higher accuracy, which is beneficial to improve the accuracy of the acquired motion information of the target image.

Fig. 12 is another schematic diagram of the model training method provided by the embodiment of the present application. As shown in Fig. 12, the method includes:

1201. Acquire images to be processed.

After obtaining the image to be processed, the position information of the pixels in the target area can be obtained and normalized.

In a possible implementation manner, the target object is a human body.

In a possible implementation manner, the position information of the pixels in the image to be processed includes coordinates of the pixels in an image coordinate system, and the image coordinate system is constructed based on the image to be processed.

For the description of step 1201, reference may be made to the related description of step 801 in the embodiment shown in FIG. 8 , and details are not repeated here.

1202. Process the image to be processed and the position information of the pixels in the image to be processed by the model to be trained to obtain the pose parameters.

After obtaining the position information of the pixels in the image to be processed and the normalized image to be processed, the position information of the pixels in the image to be processed and the normalized image to be processed can be input to the model to be trained, so as to pass the The training model processes the target area of the image to be processed and the normalized position information of the pixels in the image to be processed to obtain the pose parameters.

In a possible implementation, the method further includes: processing the target area of the image to be processed and the normalized position information of pixels in the image to be processed by the gesture recognition model to obtain shape parameters and displacement parameters, and pose parameters, shape parameters and displacement parameters are jointly used to obtain the pose of the target object.

For the description of step 1202, reference may be made to the relevant description of step 802 in the embodiment shown in FIG. 8 , and details are not repeated here.

1203. Acquire a predicted pose of the target object based on the pose parameter.

For descriptions of step 1203, reference may be made to relevant descriptions of step 1003 in the embodiment shown in FIG. 10 , and details are not repeated here.

1204. Based on the predicted pose of the target object and the real pose of the target object, train the model to be trained to obtain a pose recognition model.

After obtaining the predicted pose of the target object, since the real pose of the target object is known, the model to be trained can be trained based on the predicted pose of the target object and the real pose of the target object, and the pose recognition in the embodiment shown in Figure 8 is obtained Model.

For the description of step 1204, reference may be made to the related description of step 1004 in the embodiment shown in FIG. 10 , which will not be repeated here.

The gesture recognition model trained in the embodiment of the present application can make the gesture recognition model perceive the position of the target object in the whole picture, and learn the relationship between the position and the pose of the target object, so as to more accurately capture the pose of the target object. When the pose recognition model is in image processing, its input not only includes the target image, but also includes the position information of the pixels in the target image, so the pose recognition model not only considers the influence of the image information of the target image itself on the pose of the target object , it also takes into account the impact of the position information of the pixel points in the target image on the attitude of the target object, and the factors considered are relatively comprehensive, so the attitude of the target object obtained based on this method has high accuracy, and further has It is beneficial to improve the accuracy of the acquired motion information of the target image.

The above is the detailed description of the model training method provided by the embodiment of the present application, and the gesture recognition device and the model training device provided by the embodiment of the present application will be introduced below. Fig. 13 is a schematic structural diagram of the gesture recognition device provided by the embodiment of the present application. As shown in Fig. 13, the device includes:

An acquisition module 1301, configured to acquire a target image;

The processing module 1302 is used to process the target area of the target image and the position information of the target area in the target image through the gesture recognition model to obtain gesture parameters, the target zone is the area where the target object is located, and the gesture parameters are used to obtain the target object attitude.

In a possible implementation, the processing module 1302 is also configured to process the target area of the target image and the position information of the target area in the target image through the gesture recognition model to obtain shape parameters and displacement parameters, gesture parameters, shape The parameters and the displacement parameters are used together to obtain the pose of the target object.

In a possible implementation manner, the device further includes: a normalization module, configured to perform normalization processing on the position information of the target area in the target image to obtain normalized position information; the processing module 1302, It is used to process the target area of the target image and the normalized position information through the gesture recognition model to obtain the gesture parameters.

In a possible implementation manner, the target object is a human body.

Fig. 14 is another schematic structural diagram of the gesture recognition device provided by the embodiment of the present application. As shown in Fig. 14, the device includes:

An acquisition module 1401, configured to acquire a target image;

The processing module 1402 is configured to process the target image and the position information of the pixels in the target image through the gesture recognition model to obtain a gesture parameter, and the gesture parameter is used to acquire the gesture of the target object included in the target image.

In a possible implementation, the processing module 1402 is also configured to process the target image and the position information of the pixels in the target image through the pose recognition model to obtain shape parameters and displacement parameters, and pose parameters, shape parameters and displacement parameters Commonly used to obtain the pose of the target object.

In a possible implementation manner, the device further includes: a normalization module, configured to perform normalization processing on the position information of pixels in the target image to obtain normalized position information; the processing module 1402 uses The target image and the normalized position information are processed by the gesture recognition model to obtain the gesture parameters.

In a possible implementation manner, the target object is a human body.

Fig. 15 is another structural schematic diagram of the model training device provided by the embodiment of the present application. As shown in Fig. 15, the device includes:

A first acquisition module 1501, configured to acquire an image to be processed;

The processing module 1502 is used to process the target area of the image to be processed by the model to be trained and the position information of the target area in the image to be processed to obtain posture parameters, and the target area is the area where the target object is located;

The second acquiring module 1503 is configured to acquire the predicted attitude of the target object based on the attitude parameter;

The training module 1504 is configured to train the model to be trained based on the predicted pose of the target object and the real pose of the target object to obtain a pose recognition model.

The gesture recognition model obtained in the embodiment of the present application can make the gesture recognition model perceive the position of the target object in the whole picture, and learn the relationship between the position and the pose of the target object, so as to capture the pose of the target object more accurately. When the gesture recognition model is in image processing, the input of the gesture recognition model not only includes the cropped target area, but also includes the position information of the target area in the target image, so the gesture recognition model not only considers the image information of the target area itself. The influence caused by the posture of the target object also takes into account the influence of the position information of the target area in the target image on the posture of the target object. The factors considered are relatively comprehensive, so the posture of the target object obtained based on this method, It has higher accuracy, which is beneficial to improve the accuracy of the acquired motion information of the target image.

In a possible implementation, the training module 1504 is configured to perform the training on the model to be trained based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real projection result of the real pose of the target object on the image to be processed. Training to get the gesture recognition model.

In a possible implementation, the processing module 1502 is also configured to process the target area of the image to be processed and the position information of the target area in the image to be processed through the gesture recognition model to obtain shape parameters and displacement parameters, pose parameters, The shape parameter and the displacement parameter are jointly used to obtain the pose of the target object.

In a possible implementation manner, the device further includes: a normalization module, configured to perform normalization processing on the position information of the target area in the image to be processed to obtain the normalized position information; the processing module 1502 , which is used to process the target area of the image to be processed and the normalized position information through the model to be trained to obtain the pose parameters.

In a possible implementation manner, the target object is a human body.

Fig. 16 is another structural schematic diagram of the model training device provided by the embodiment of the present application. As shown in Fig. 16, the device includes:

A first acquisition module 1601, configured to acquire an image to be processed;

The processing module 1602 is used to process the image to be processed and the position information of the pixels in the image to be processed through the model to be trained to obtain the attitude parameter;

The second acquiring module 1603 is configured to acquire the predicted attitude of the target object based on the attitude parameter;

The training module 1604 is configured to train the model to be trained based on the predicted pose of the target object and the real pose of the target object to obtain a pose recognition model.

The gesture recognition model obtained in the embodiment of the present application can make the gesture recognition model perceive the position of the target object in the whole picture, and learn the relationship between the position and the pose of the target object, so as to capture the pose of the target object more accurately. When the pose recognition model is in image processing, its input not only includes the target image, but also includes the position information of the pixels in the target image, so the pose recognition model not only considers the influence of the image information of the target image itself on the pose of the target object , it also takes into account the impact of the position information of the pixel points in the target image on the attitude of the target object, and the factors considered are relatively comprehensive, so the attitude of the target object obtained based on this method has high accuracy, and further has It is beneficial to improve the accuracy of the acquired motion information of the target image.

In a possible implementation, the training module 1604 is configured to perform the training on the model to be trained based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real projection result of the real pose of the target object on the image to be processed. Training to get the gesture recognition model.

In a possible implementation, the processing module 1602 is also used to process the target area of the image to be processed and the position information of the pixels in the image to be processed through the pose recognition model to obtain shape parameters and displacement parameters, pose parameters, shape The parameters and the displacement parameters are used together to obtain the pose of the target object.

In a possible implementation manner, the device further includes: a normalization module, configured to perform normalization processing on the position information of pixels in the image to be processed to obtain normalized position information; the processing module 1602 uses The attitude parameter is obtained by processing the image to be processed and the normalized position information through the model to be trained.

In a possible implementation manner, the target object is a human body.

It should be noted that the information interaction and execution process between the modules/units of the above-mentioned device are based on the same concept as the method embodiment of the present application, and the technical effect it brings is the same as that of the method embodiment of the present application. The specific content can be Reference is made to the descriptions in the foregoing method embodiments shown in the embodiments of the present application, and details are not repeated here.

The embodiment of the present application also relates to an execution device, and FIG. 17 is a schematic structural diagram of the execution device provided in the embodiment of the present application. As shown in FIG. 17 , the execution device 1700 may specifically be a mobile phone, a tablet, a notebook computer, a smart wearable device, a server, etc., which is not limited here. Wherein, the image classification apparatus described in the embodiment corresponding to FIG. 13 or FIG. 14 may be deployed on the execution device 1700 to realize the gesture recognition function in the embodiment corresponding to FIG. 6 or FIG. 8 . Specifically, the execution device 1700 includes: a receiver 1701, a transmitter 1702, a processor 1703, and a memory 1704 (the number of processors 1703 in the execution device 1700 may be one or more, and one processor is taken as an example in FIG. 17 ) , where the processor 1703 may include an application processor 17031 and a communication processor 17032 . In some embodiments of the present application, the receiver 1701 , the transmitter 1702 , the processor 1703 and the memory 1704 may be connected through a bus or in other ways.

The memory 1704 may include read-only memory and random-access memory, and provides instructions and data to the processor 1703 . A part of the memory 1704 may also include a non-volatile random access memory (non-volatile random access memory, NVRAM). The memory 1704 stores processors and operating instructions, executable modules or data structures, or their subsets, or their extended sets, wherein the operating instructions may include various operating instructions for implementing various operations.

The processor 1703 controls the operations of the execution device. In a specific application, various components of the execution device are coupled together through a bus system, where the bus system may include not only a data bus, but also a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the various buses are referred to as bus systems in the figures.

The methods disclosed in the foregoing embodiments of the present application may be applied to the processor 1703 or implemented by the processor 1703 . The processor 1703 may be an integrated circuit chip, which has a signal processing capability. In the implementation process, each step of the above method may be completed by an integrated logic circuit of hardware in the processor 1703 or instructions in the form of software. The above-mentioned processor 1703 may be a general-purpose processor, a digital signal processor (digital signal processing, DSP), a microprocessor or a microcontroller, and may further include an application specific integrated circuit (ASIC), field programmable Field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The processor 1703 may implement or execute various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like. The steps of the method disclosed in connection with the embodiments of the present application may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, register. The storage medium is located in the memory 1704, and the processor 1703 reads the information in the memory 1704, and completes the steps of the above method in combination with its hardware.

The receiver 1701 can be used to receive input digital or character information, and generate signal input related to performing device related settings and function control. The transmitter 1702 can be used to output digital or character information through the first interface; the transmitter 1702 can also be used to send instructions to the disk group through the first interface to modify the data in the disk group; the transmitter 1702 can also include a display device such as a display screen .

In the embodiment of the present application, in one case, the processor 1703 is configured to perform gesture recognition on the image by using the gesture recognition model in the embodiment corresponding to FIG. 6 or FIG. 8 .

The embodiment of the present application also relates to a training device, and FIG. 18 is a schematic structural diagram of the training device provided in the embodiment of the present application. As shown in Figure 18, the training device 1800 is implemented by one or more servers, and the training device 1800 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (central processing units, CPU) 1814 (eg, one or more processors) and memory 1832, one or more storage media 1830 (eg, one or more mass storage devices) for storing application programs 1842 or data 1844. Wherein, the memory 1832 and the storage medium 1830 may be temporary storage or persistent storage. The program stored in the storage medium 1830 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the training device. Furthermore, the central processing unit 1814 may be configured to communicate with the storage medium 1830 , and execute a series of instruction operations in the storage medium 1830 on the training device 1800 .

The training device 1800 can also include one or more power supplies 1826, one or more wired or wireless network interfaces 1850, one or more input and output interfaces 1858; or, one or more operating systems 1841, such as Windows Server™, Mac OS X™ , UnixTM, LinuxTM, FreeBSDTM and so on.

Specifically, the training device may execute the model training method in the embodiment corresponding to FIG. 10 or FIG. 12 .

The embodiment of the present application also relates to a computer storage medium, where a program for signal processing is stored in the computer-readable storage medium, and when the program is run on the computer, the computer executes the steps performed by the aforementioned execution device, or, The computer is caused to perform the steps as performed by the aforementioned training device.

The embodiment of the present application also relates to a computer program product, where instructions are stored in the computer program product, and when executed by a computer, the instructions cause the computer to perform the steps performed by the aforementioned executing device, or cause the computer to perform the steps performed by the aforementioned training device. A step of.

The execution device, training device or terminal device provided in the embodiment of the present application may specifically be a chip. The chip includes: a processing unit and a communication unit. The processing unit may be, for example, a processor, and the communication unit may be, for example, an input/output interface, pins or circuits etc. The processing unit can execute the computer-executed instructions stored in the storage unit, so that the chips in the execution device execute the data processing methods described in the above embodiments, or make the chips in the training device execute the data processing methods described in the above embodiments. Optionally, the storage unit is a storage unit in the chip, such as a register, a cache, etc., and the storage unit may also be a storage unit located outside the chip in the wireless access device, such as only Read-only memory (ROM) or other types of static storage devices that can store static information and instructions, random access memory (random access memory, RAM), etc.

Specifically, please refer to FIG. 19. FIG. 19 is a schematic structural diagram of a chip provided by the embodiment of the present application. The chip can be represented as a neural network processor NPU 1900, and the NPU 1900 is mounted to the main CPU (Host CPU) as a coprocessor ), the tasks are assigned by the Host CPU. The core part of the NPU is the operation circuit 1903, and the operation circuit 1903 is controlled by the controller 1904 to extract matrix data in the memory and perform multiplication operations.

In some implementations, the operation circuit 1903 includes multiple processing units (Process Engine, PE). In some implementations, arithmetic circuit 1903 is a two-dimensional systolic array. The arithmetic circuit 1903 may also be a one-dimensional systolic array or other electronic circuits capable of performing mathematical operations such as multiplication and addition. In some implementations, arithmetic circuit 1903 is a general-purpose matrix processor.

For example, suppose there is an input matrix A, a weight matrix B, and an output matrix C. The operation circuit fetches the data corresponding to the matrix B from the weight memory 1902, and caches it in each PE in the operation circuit. The operation circuit takes the data of matrix A from the input memory 1901 and performs matrix operation with matrix B, and the obtained partial or final results of the matrix are stored in the accumulator (accumulator) 1908 .

The unified memory 1906 is used to store input data and output data. The weight data directly accesses the controller (Direct Memory Access Controller, DMAC) 1905 through the storage unit, and the DMAC is transferred to the weight storage 1902. Input data is also transferred to unified memory 1906 by DMAC.

The BIU is the Bus Interface Unit, that is, the bus interface unit 1913, which is used for the interaction between the AXI bus and the DMAC and the instruction fetch buffer (Instruction Fetch Buffer, IFB) 1909.

The bus interface unit 1913 (Bus Interface Unit, BIU for short), is used for the instruction fetch memory 1909 to obtain instructions from the external memory, and is also used for the storage unit access controller 1905 to obtain the original data of the input matrix A or the weight matrix B from the external memory.

The DMAC is mainly used to move the input data in the external memory DDR to the unified memory 1906 , move the weight data to the weight memory 1902 , or move the input data to the input memory 1901 .

The vector computing unit 1907 includes a plurality of computing processing units, and if necessary, further processes the output of the computing circuit 1903, such as vector multiplication, vector addition, exponent operation, logarithmic operation, size comparison and so on. It is mainly used for non-convolutional/fully connected layer network calculations in neural networks, such as Batch Normalization (batch normalization), pixel-level summation, upsampling of predicted label planes, etc.

In some implementations, the vector computation unit 1907 can store the vector of the processed output to unified memory 1906 . For example, the vector calculation unit 1907 can apply a linear function; or, a non-linear function to the output of the operation circuit 1903, such as performing linear interpolation on the predicted label plane extracted by the convolutional layer, and then for example, a vector of accumulated values to generate an activation value . In some implementations, the vector calculation unit 1907 generates normalized values, pixel-level summed values, or both. In some implementations, the vector of processed outputs can be used as an activation input to operational circuitry 1903, for example for use in subsequent layers in a neural network.

An instruction fetch buffer (instruction fetch buffer) 1909 connected to the controller 1904 is used to store instructions used by the controller 1904;

The unified memory 1906, the input memory 1901, the weight memory 1902 and the fetch memory 1909 are all On-Chip memories. External memory is private to the NPU hardware architecture.

Wherein, the processor mentioned above can be a general-purpose central processing unit, microprocessor, ASIC, or one or more integrated circuits for controlling the execution of the above-mentioned programs.

In addition, it should be noted that the device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be A physical unit can be located in one place, or it can be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the device embodiments provided in the present application, the connection relationship between the modules indicates that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines.

Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary general-purpose hardware, and of course it can also be realized by special hardware including application-specific integrated circuits, dedicated CPUs, dedicated memories, Special components, etc. to achieve. In general, all functions completed by computer programs can be easily realized by corresponding hardware, and the specific hardware structure used to realize the same function can also be varied, such as analog circuits, digital circuits or special-purpose circuit etc. However, for this application, software program implementation is a better implementation mode in most cases. Based on this understanding, the essence of the technical solution of this application or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a readable storage medium, such as a floppy disk of a computer , U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., including several instructions to make a computer device (which can be a personal computer, training device, or network device, etc.) execute the instructions described in various embodiments of the present application method.

In the above embodiments, all or part of them may be implemented by software, hardware, firmware or any combination thereof. When implemented using software, it may be implemented in whole or in part in the form of a computer program product.

The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the processes or functions according to the embodiments of the present application will be generated in whole or in part. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transferred from a website, computer, training device, or data The center transmits to another website site, computer, training device or data center via wired (eg, coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (eg, infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that can be stored by a computer, or a data storage device such as a training device or a data center integrated with one or more available media. The available medium may be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (Solid State Disk, SSD)), etc.

Claims

A gesture recognition method, characterized in that the method comprises:

Get the target image;

Process the target area of the target image and the position information of the target area in the target image through a gesture recognition model to obtain a gesture parameter, the target zone is an area where the target object is located, and the gesture parameter is used for Get the pose of the target object.
The method according to claim 1, wherein the location information includes the coordinates of the center point of the target area in an image coordinate system and the size of the target area, and the image coordinate system is based on the target image build; or,

The location information includes coordinates of vertices of the target area in an image coordinate system, and the image coordinate system is constructed based on the target image.
The method according to claim 1 or 2, wherein the pose recognition model is based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real pose of the target object on the image to be processed. It is obtained by processing the real projection results on the image for training.
The method according to any one of claims 1 to 3, wherein the posture of the target object includes the orientation of the target object in the camera coordinate system and the body behavior of the target object in the camera coordinate system, The camera coordinate system is constructed based on the camera that captures the target image.
The method according to any one of claims 1 to 4, wherein the method further comprises:

Processing the target area of the target image and the position information of the target area in the target image through a gesture recognition model to obtain a shape parameter and a displacement parameter, the posture parameter, the shape parameter and the displacement parameter Commonly used to obtain the pose of the target object.
The method according to any one of claims 1 to 5, wherein the method further comprises:

performing normalization processing on the position information of the target area in the target image to obtain normalized position information;

The gesture recognition model is used to process the target area of the target image and the position information of the target area in the target image, and the gesture parameters obtained include:

The target area of the target image and the normalized position information are processed by a gesture recognition model to obtain gesture parameters.
A gesture recognition method, characterized in that the method comprises:

Get the target image;

The target image and the position information of the pixel points in the target image are processed by a gesture recognition model to obtain a gesture parameter, and the gesture parameter is used to acquire the gesture of the target object included in the target image.
The method according to claim 7, wherein the position information includes coordinates of pixels in an image coordinate system, and the image coordinate system is constructed based on the target image.
The method according to claim 7 or 8, wherein the pose recognition model is based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real pose of the target object on the image to be processed. The real projection result on the image is obtained by training.
The method according to any one of claims 7 to 9, wherein the posture of the target object includes the orientation of the target object in the camera coordinate system and the body behavior of the target object in the camera coordinate system, The camera coordinate system is constructed based on the camera that captures the target image.
The method according to any one of claims 7 to 10, further comprising:

Process the target image and the position information of the pixels in the target image through a gesture recognition model to obtain a shape parameter and a displacement parameter, and the gesture parameter, the shape parameter and the displacement parameter are jointly used to obtain the The pose of the target object.
The method according to any one of claims 7 to 11, further comprising:

performing normalization processing on the position information of pixels in the target image to obtain normalized position information;

The position information of the target image and the pixel points in the target image is processed by the gesture recognition model, and the gesture parameters obtained include:

The target image and the normalized position information are processed by a gesture recognition model to obtain gesture parameters.
A method for model training, characterized in that the method comprises:

Get the image to be processed;

Processing the target area of the image to be processed and the position information of the target area in the image to be processed by the model to be trained to obtain attitude parameters, the target area being the area where the target object is located;

Obtaining a predicted pose of the target object based on the pose parameters;

Based on the predicted pose of the target object and the real pose of the target object, the model to be trained is trained to obtain a pose recognition model.
The method according to claim 13, wherein the training of the model to be trained based on the predicted pose of the target object and the real pose of the target object to obtain a pose recognition model includes:

Based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real projection result of the real pose of the target object on the image to be processed, train the model to be trained to obtain a pose Identify the model.
The method according to claim 13 or 14, wherein the position information includes the coordinates of the center point of the target area in an image coordinate system and the size of the target area, and the image coordinate system is based on the pending image construction; or,

The location information includes coordinates of vertices of the target area in an image coordinate system, and the image coordinate system is constructed based on the image to be processed.
A method for model training, characterized in that the method comprises:

Get the image to be processed;

Processing the image to be processed and the position information of pixels in the image to be processed by the model to be trained to obtain attitude parameters;

Obtaining a predicted pose of the target object based on the pose parameters;

Based on the predicted pose of the target object and the real pose of the target object, the model to be trained is trained to obtain a pose recognition model.
The method according to claim 16, wherein the training of the model to be trained based on the predicted pose of the target object and the real pose of the target object to obtain a pose recognition model includes:

Based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real projection result of the real pose of the target object on the image to be processed, train the model to be trained to obtain a pose Identify the model.
The method according to claim 16 or 17, wherein the position information includes coordinates of pixels in an image coordinate system, and the image coordinate system is constructed based on the image to be processed.
A gesture recognition device, characterized in that the device comprises:

An acquisition module, configured to acquire a target image;

A processing module, configured to process the target area of the target image and the position information of the target area in the target image through a gesture recognition model to obtain gesture parameters, the target area is the area where the target object is located, and the The pose parameters are used to obtain the pose of the target object.
The device according to claim 19, wherein the position information includes the coordinates of the center point of the target area in an image coordinate system and the size of the target area, and the image coordinate system is based on the target image Build; or,

The location information includes coordinates of vertices of the target area in an image coordinate system, and the image coordinate system is constructed based on the target image.
The method according to claim 19 or 20, wherein the pose recognition model is based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real pose of the target object on the image to be processed. It is obtained by processing the real projection results on the image for training.
A gesture recognition device, characterized in that the device comprises:

An acquisition module, configured to acquire a target image;

The processing module is configured to process the target image and the position information of the pixel points in the target image through a gesture recognition model to obtain a gesture parameter, and the gesture parameter is used to acquire the gesture of the target object included in the target image.
The device according to claim 22, wherein the position information includes coordinates of pixels in an image coordinate system, and the image coordinate system is constructed based on the target image.
The device according to claim 22 or 23, wherein the pose recognition model is based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real pose of the target object on the image to be processed. It is obtained by processing the real projection results on the image for training.
A model training device, characterized in that the device comprises:

The first acquisition module is used to acquire the image to be processed;

A processing module, configured to process the target area of the image to be processed and the position information of the target area in the image to be processed by the model to be trained to obtain posture parameters, the target area being the area where the target object is located ;

A second acquisition module, configured to acquire the predicted attitude of the target object based on the attitude parameters;

The training module is used to train the model to be trained based on the predicted pose of the target object and the real pose of the target object to obtain a pose recognition model.
The device according to claim 25, wherein the training module is configured to predict projection results based on the predicted pose of the target object on the image to be processed and the real pose of the target object in the The real projection result on the image to be processed is used to train the model to be trained to obtain a gesture recognition model.
The device according to claim 25 or 26, wherein the position information includes the coordinates of the center point of the target area in an image coordinate system and the size of the target area, and the image coordinate system is based on the pending image construction; or,

The location information includes coordinates of vertices of the target area in an image coordinate system, and the image coordinate system is constructed based on the image to be processed.
A model training device, characterized in that the device comprises:

The first acquisition module is used to acquire the image to be processed;

A processing module, configured to process the image to be processed and the position information of pixels in the image to be processed through the model to be trained to obtain attitude parameters;

A second acquisition module, configured to acquire the predicted attitude of the target object based on the attitude parameters;

A training module, configured to train the model to be trained based on the predicted pose of the target object and the real pose of the target object to obtain a pose recognition model.
The device according to claim 28, wherein the training module is configured to be based on the predicted projection result of the predicted pose of the target object on the image to be processed and the real pose of the target object in the The real projection result on the image to be processed is used to train the model to be trained to obtain a gesture recognition model.
The device according to claim 28 or 29, wherein the position information includes coordinates of pixels in an image coordinate system, and the image coordinate system is constructed based on the image to be processed.
A gesture recognition device, characterized in that the device includes a memory and a processor; the memory stores codes, the processor is configured to execute the codes, and when the codes are executed, the gesture recognition The device executes the method as claimed in any one of claims 1-18.
A computer storage medium, wherein the computer storage medium stores one or more instructions which, when executed by one or more computers, cause the one or more computers to implement any of claims 1 to 18 a method as described.
A computer program product, characterized in that the computer program product stores instructions, and when the instructions are executed by a computer, the computer implements the method according to any one of claims 1 to 18.