Archery action recognition method and electronic device

Abstract

The application provides an archery action recognition method and an electronic device, and the method comprises the following steps: determining key point position information of a plurality of key points in each first image according to a plurality of first images related to archery actions in the process of archery of a user, performing first archery action recognition processing according to the key point position information of each key point, obtaining a first archery action stage recognition result, performing second archery action recognition processing according to the plurality of first images based on an image classification model, obtaining a second archery action stage recognition result, and obtaining a target archery action stage recognition result according to the first archery action stage recognition result and the second archery action stage recognition result. In this way, a more accurate target archery action stage recognition result can be obtained according to the first archery action stage recognition result and the second archery action stage recognition result, so that the user can more accurately adjust the archery action according to the archery action stage recognition result, and the archery is more accurate.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method and electronic device for recognizing archery actions. Background Technology

[0002] Archery is a sport that uses the elasticity of a bow to launch an arrow, competing for accuracy over a certain distance. Currently, archery is increasingly popular in both competitive and recreational settings. During archery, the archery movements differ at different stages. Therefore, it is necessary to identify these movements so that users can analyze them based on the identification results, identify problems, and adjust their shooting techniques for greater accuracy.

[0003] In summary, how to more accurately identify the stages of archery action so that users can analyze the different stages of archery action is a problem that urgently needs to be solved. Summary of the Invention

[0004] This application provides an archery action recognition method and electronic device, which can more accurately identify the archery action stages during the archery process and obtain the archery action stage recognition results, so that users can adjust the archery action at different stages according to the archery action stage recognition results, thereby making the archery results more accurate.

[0005] To address the aforementioned technical problems, in a first aspect, embodiments of this application provide an archery action recognition method. The method includes: determining multiple first images related to the archery action during a user's archery process; determining key point location information of multiple key points in each first image based on each first image, and performing first archery action recognition processing based on the key point location information of each key point in each first image to obtain a first archery action stage recognition result corresponding to the multiple first images; performing second archery action recognition processing based on an image classification model using the multiple first images to obtain a second archery action stage recognition result corresponding to the multiple first images; and obtaining a target archery action stage recognition result corresponding to the multiple first images based on the first and second archery action stage recognition results.

[0006] Using the above technical solution, the key point position information of multiple key points in each frame of the first image is determined based on the multiple frames of the first image related to the archery action during the user's archery process. Then, the first archery action recognition processing is performed based on the key point position information of each key point to obtain the first archery action stage recognition result corresponding to the multiple frames of the first image. Based on the image classification model, the second archery action recognition processing is performed based on the multiple frames of the first image to obtain the second archery action stage recognition result corresponding to the multiple frames of the first image. The target archery action stage recognition result is obtained based on the first archery action stage recognition result and the second archery action stage recognition result. Thus, by performing first archery action recognition processing on the key point position information of the user's archery process in multiple first images, the first archery action stage recognition result is obtained. Based on the image classification model, the user's archery action is then processed into second archery action recognition based on multiple first images, resulting in the second archery action stage recognition result. Furthermore, based on the first and second archery action stage recognition results, the target archery action stage recognition result is obtained. This makes the archery action stage recognition during the user's archery process more accurate, resulting in more precise archery action stage recognition results. Consequently, the user can adjust their archery action based on the archery action stage recognition results, making archery more accurate.

[0007] In one possible implementation of the first aspect described above, after determining the key point location information of multiple key points in each first image frame based on each first image frame, the method further includes: generating a key point location information time sequence of key points based on the key point location information of multiple key points in each first image frame, so as to perform a first archery action recognition process based on the key point location information of each key point corresponding to each first image frame included in the key point location information time sequence, wherein the key point location information time sequence includes the key point location information of each key point corresponding to each first image frame arranged in a forward time sequence of each first image frame.

[0008] In one possible implementation of the first aspect described above, the first archery action recognition process is performed based on the key point position information of each key point corresponding to each frame of the first image to obtain the recognition result of the first archery action stage corresponding to multiple frames of the first image. This includes: determining the distance information between each key point corresponding to each frame of the first image based on the key point position information of each key point corresponding to each frame of the first image, and obtaining the distance change trend of the key points corresponding to multiple frames of the first image based on the distance change trend of the key points corresponding to multiple frames of the first image; obtaining the first archery action stage corresponding to the first target frame image in the multiple frames of the first image and the first confidence indication information corresponding to the first archery action stage based on the distance change trend of the key points corresponding to the multiple frames of the first image; and obtaining the recognition result of the first archery action stage corresponding to multiple frames of the first image based on the first archery action stage corresponding to the first target frame image and the first confidence indication information corresponding to the first archery action stage.

[0009] By employing the above technical solution, the key points of the archery process are analyzed based on the distance variation trends between key points in multiple frames of the first image, yielding archery action stages and corresponding confidence level indications. Thus, analyzing archery actions based on the distance information of key points makes the archery action recognition process more refined, resulting in more accurate identification results for different archery action stages.

[0010] In one possible implementation of the first aspect described above, the first archery action stage corresponding to the target frame image in the multi-frame first images is obtained based on the distance change trend of key points corresponding to key points in the multi-frame first images, including: if the distance change of key points corresponding to the first target frame image in the multi-frame first images satisfies a first change condition based on the distance change trend of key points corresponding to key points in the multi-frame first images, then the first archery action stage corresponding to the first target frame image is determined to be the drawing-the-bow stage; if the distance change of key points corresponding to the first target frame image in the multi-frame first images satisfies a second change condition based on the key point change trend of key points corresponding to key points in the multi-frame first images, then the first archery action stage corresponding to the first target frame image is determined to be the charging-up stage; if the distance change of key points corresponding to the first target frame image in the multi-frame first images satisfies a third change condition based on the key point change trend of key points corresponding to key points in the multi-frame first images, then the first archery action stage corresponding to the first target frame image is determined to be the releasing stage.

[0011] By adopting the above technical solution, the distance change trend between each key point is determined, and the archery action stage corresponding to at least one set of target frame images in multiple first images is determined based on the distance change trend between every two key points, so as to achieve more accurate identification of the archery action stage and obtain more accurate archery action identification results.

[0012] In one possible implementation of the first aspect above, the first change condition is that the difference between the distances to the key points corresponding to the first target frame image in the distance change trend and between any two frames is the same and the difference is greater than zero and less than a preset first difference threshold; the second change condition is that the difference between the distances to the key points corresponding to the first target frame image in the distance change trend and between any two frames tends to zero; the third change condition is that the difference between the distances to the key points corresponding to the first target frame image in the distance change trend and between any two frames is greater than a preset second difference threshold.

[0013] By adopting the above technical solution, based on the distance change characteristics of the user's key points during archery, the archery action stage corresponding to the target frame image is determined. This allows the user to know which archery action stage is being performed at different times based on the archery action recognition results, and to correct the archery action at different stages to make archery more accurate.

[0014] In one possible implementation of the first aspect described above, based on an image classification model, a second archery action recognition process is performed on multiple frames of first images to obtain the recognition result of the second archery action stage corresponding to the multiple frames of first images. This includes: performing the second archery action recognition process on multiple frames of first images based on an image classification model to obtain the archery action category and archery action category confidence indication information for each frame of first images; obtaining the second archery action stage corresponding to the second target frame image in the multiple frames of first images and the second confidence indication information corresponding to the second archery action stage based on the archery action category and the archery action category confidence indication information for each frame of first images; and obtaining the recognition result of the second archery action stage corresponding to the multiple frames of first images based on the second archery action stage corresponding to the second target frame image and the second confidence indication information corresponding to the second archery action stage.

[0015] Using the above technical solution, each frame of the first image is classified based on an image classification model to obtain the archery action category and confidence level information for each frame. This, in turn, yields the second archery action stage corresponding to the second target frame image across multiple first images, and the second confidence level information corresponding to that stage. Thus, a second archery action recognition is performed based on the image classification model for subsequent fusion, resulting in a more accurate archery action recognition outcome.

[0016] In one possible implementation of the first aspect described above, the target archery action phase recognition result includes the target archery action phase corresponding to the target frame image. Based on the first and second archery action phase recognition results, the target archery action phase recognition result corresponding to multiple first frames is obtained, including: when it is determined that the first target frame image and the second target frame image have the same frame, and the first archery action phase corresponding to the first target frame image and the second archery action phase corresponding to the second target frame image are the same archery action phase, a first weight coefficient corresponding to the first confidence indication information and a second weight coefficient corresponding to the second confidence indication information are determined based on the second confidence indication information. If the second confidence level indication information is greater than the preset first confidence level threshold, then the first weight coefficient is less than the second weight coefficient, and the target frame image is the second target frame image. If the second confidence level indication information is less than the preset second confidence level threshold, then the first weight coefficient is greater than the second weight coefficient, and the target frame image is the first target frame image. Based on the first confidence level indication information, the second confidence level indication information, the first weight coefficient, and the second weight coefficient, the target confidence level indication information corresponding to the target frame image is obtained. If it is determined that the target confidence level indication information is greater than the preset third confidence level threshold, the first archery action stage or the second archery action stage is determined to be the target archery action stage corresponding to the target frame image.

[0017] By employing the above technical solution, a more accurate archery action recognition result can be obtained by fusing the first and second archery action recognition results. Furthermore, by dynamically adjusting the weight coefficients based on the second confidence level indication information, the target confidence level indication information can better reflect the accuracy of the target archery action stage corresponding to the target frame image, thereby making the target archery action recognition of the target frame image more accurate.

[0018] In one possible implementation of the first aspect described above, the multiple first images have corresponding timestamp information, and the target archery action phase recognition result also includes the start time, duration, and corresponding stability information of the target archery action phase. Then, based on the first and second archery action phase recognition results, the target archery action phase recognition result corresponding to the multiple first images is obtained, including: obtaining the start time and duration of the target archery action phase based on the timestamp information corresponding to the target frame image; and determining the detection stability information of the target archery action phase based on the first and second confidence indication information corresponding to the target frame image.

[0019] By adopting the above technical solution, by determining the start time and duration of the target archery action phase corresponding to the target frame image, as well as detecting stability information, it is more beneficial for users to analyze the corresponding archery action phase.

[0020] In one possible implementation of the first aspect above, determining multiple first images related to the archery action during the user's archery process includes: acquiring an archery action video related to the archery action during the user's archery process, and obtaining multiple first images based on the archery action video; or acquiring multiple consecutive images related to the archery action during the user's archery process to obtain multiple first images.

[0021] By adopting the above technical solution, archery action recognition based on archery action video or multiple consecutive images can be made more accurate.

[0022] In one possible implementation of the first aspect above, determining the key point location information of multiple key points in each frame of the first image based on each frame of the first image includes: inputting each frame of the first image to a key point detection model so that the key point detection model determines the key point location information of multiple key points in each frame of the first image based on each frame of the first image, the key points including hand key points.

[0023] By adopting the above technical solution, key point location information is identified based on the key point detection model, which accelerates the key point identification efficiency of multiple frames of first image.

[0024] In one possible implementation of the first aspect above, the second archery action recognition processing based on the image classification model according to multiple frames of first images includes: cropping the target region in the multiple frames of first images to obtain multiple frames of second images, and inputting the multiple frames of second images into the image classification model so that the image classification model can perform the second archery action recognition processing based on the multiple frames of second images, wherein the target region includes at least one of the hand region, the face region, and the bow and arrow equipment region; or inputting the multiple frames of first images into the image classification model so that the image classification model can perform the second archery action recognition processing based on the multiple frames of first images.

[0025] Using the above technical solution, at least one region from the hand region, face region, and bow and arrow equipment region in the first image that better represents the archery action is cropped to obtain the second image. This allows the image classification model to perform second archery action recognition processing based on multiple frames of the second image, so that the input is an image that better represents the archery action, and more accurate archery action category and archery action category confidence information can be obtained.

[0026] Secondly, this application also discloses an electronic device, including: a processor and a memory communicatively connected to the processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to enable the electronic device to implement the archery action recognition method provided by any of the implementations of the first aspect.

[0027] Thirdly, this application also discloses a computer-readable storage medium storing a computer program that can be executed by a computer cluster to implement the neural network model-based reasoning method provided by any of the implementations of the first aspect above, and / or the archery action recognition method provided by any of the implementations of the second aspect above.

[0028] Fourthly, this application also discloses a computer program product, including a computer program that, when executed by a computer cluster, implements the reasoning method based on a neural network model provided by any of the implementations of the first aspect above, and / or the archery action recognition method provided by any of the implementations of the second aspect above.

[0029] The relevant beneficial effects of the second to fourth aspects mentioned above can be found in the relevant descriptions in the first or second aspects mentioned above, and will not be repeated here. Attached Figure Description

[0030] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0031] Figure 1A flowchart illustrating the archery action recognition method provided in this application embodiment;

[0032] Figure 2 A flowchart illustrating the determination of the identification result of the first archery action phase provided in an embodiment of this application;

[0033] Figure 3 A flowchart illustrating the determination of the identification result of the second archery action phase provided in this application embodiment;

[0034] Figure 4 A flowchart illustrating the identification results of the target archery action phase provided in this application embodiment;

[0035] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0036] As mentioned earlier, the archery movements differ at different stages of the shooting process. Whether in archery simulation or training, it is necessary to identify these movements to pinpoint the different stages the user is performing. This allows for analysis of the archery actions at each stage, enabling adjustments to the movements to ensure greater standardization and ultimately, more accurate shooting results.

[0037] Existing archery motion recognition methods mostly involve placing simple sensors such as accelerometers, gyroscopes, and magnetometers on the user's wrist or bow and arrow. These methods rely on traditional motion recognition systems to identify changes in the user's hand movements or bow and arrow movements based on the acquired motion information, thereby recognizing the user's archery actions. Alternatively, they can use preset rule matching algorithms to identify the user's archery actions during the shooting process. However, both of these methods are not accurate enough in recognizing complex human movements, lack naturalness and interactivity, and traditional motion recognition systems suffer from problems such as high latency, low accuracy, and poor generalization in motion decomposition and action intent judgment.

[0038] Based on this, this application proposes an archery action recognition method that can be applied to real-time recognition during archery and to post-archery process analysis. By acquiring multiple frames of first images related to the user's archery actions during the shooting process, the method determines the key point location information of multiple key points in each frame. Based on this key point location information, a first archery action recognition process is performed to obtain the first archery action stage recognition result. Further, based on an image classification model and multiple frames of first images, a second archery action recognition process is performed to obtain the second archery action stage recognition result. Finally, based on the first and second archery action stage recognition results, the final archery action stage recognition result is obtained. Thus, during archery, the current archery action stage can be identified in real time, and different archery action stages throughout the entire archery process can also be identified. Furthermore, this dual recognition makes the archery action stage recognition during the user's archery process more accurate, resulting in a more precise archery action stage recognition result. This allows the user to adjust their archery actions based on the archery action stage recognition result, making archery more accurate.

[0039] See Figure 1 The archery action recognition method provided in this application specifically includes the following steps.

[0040] S100, determine the first image of multiple frames related to the archery action during the user's archery process.

[0041] S200: Determine the key point position information of multiple key points in each frame of the first image, and perform the first archery action recognition processing based on the key point position information of each key point in each frame of the first image to obtain the recognition result of the first archery action stage corresponding to multiple frames of the first image.

[0042] S300, based on the image classification model, performs second archery action recognition processing based on multiple frames of the first image, and obtains the recognition result of the second archery action stage corresponding to the multiple frames of the first image.

[0043] S400: Based on the recognition results of the first and second archery action stages, the target archery action stage recognition results corresponding to multiple frames of the first image are obtained.

[0044] The archery action recognition method provided in this application determines the key point position information of multiple key points in each frame of the first image based on multiple frames of first images related to the archery action during the user's archery process. Then, it performs first archery action recognition processing based on the key point position information of each key point to obtain the first archery action stage recognition result corresponding to the multiple frames of first images. Based on an image classification model, it performs second archery action recognition processing based on the multiple frames of first images to obtain the second archery action stage recognition result corresponding to the multiple frames of first images. Finally, it obtains the target archery action stage recognition result based on the first and second archery action stage recognition results. Thus, by performing first archery action recognition processing on the key point position information of the user's archery process in multiple first images, the first archery action stage recognition result is obtained. Based on the image classification model, the user's archery action is then processed into second archery action recognition based on multiple first images, resulting in the second archery action stage recognition result. Furthermore, based on the first and second archery action stage recognition results, the target archery action stage recognition result is obtained. This makes the archery action stage recognition during the user's archery process more accurate, resulting in more precise archery action stage recognition results. Consequently, the user can adjust their archery action based on the archery action stage recognition results, making archery more accurate.

[0045] First, step S100 is executed to determine the first multi-frame images related to the archery action during the user's archery process.

[0046] In one implementation of this application, determining multiple first images related to the archery action during a user's archery process includes: acquiring an archery action video related to the archery action during the user's archery process, and obtaining multiple first images based on the archery action video.

[0047] For example, an image acquisition device is set up near the user's archery area. The front or side camera of the image acquisition device captures video of the archery action, including the complete archery motion. Multi-frame extraction is performed on the archery action video to obtain multiple first images. Each image includes key information such as the user's hand area, face area, and bow and arrow equipment.

[0048] Furthermore, the archery action video can be a video of the entire archery process or a video of a portion of the archery process (e.g., the archery action video corresponding to the currently ongoing archery process). Therefore, the multi-frame first image can be an image of the archery action of the entire archery process or an image of the archery action of a portion of the archery process.

[0049] In another implementation of this application, determining the multiple first images related to the archery action during the user's archery process includes: acquiring multiple consecutive images related to the archery action during the user's archery process to obtain the multiple first images.

[0050] For example, multiple frames of continuous archery action images containing the complete archery action process are captured by the front or side camera of the image acquisition device to obtain multiple first images.

[0051] Among them, multiple consecutive images are archery action images of the entire archery process or archery action images of a part of the archery process (e.g., the archery action image corresponding to the currently ongoing archery process). Therefore, multiple first images are archery action images of the entire archery process or archery action images of a part of the archery process.

[0052] It should be noted that each frame of the first image carries a corresponding timestamp.

[0053] Next, step S200 is executed, which determines the key point position information of multiple key points in each frame of the first image, and performs the first archery action recognition processing based on the key point position information of each key point in each frame of the first image, so as to obtain the recognition result of the first archery action stage corresponding to multiple frames of the first image.

[0054] In one implementation of this application, the key points include at least one of hand key points, bow and arrow key points, and face key points.

[0055] Among them, key hand points include the wrist point of the hand holding the bow and at least one joint point of at least one of the five fingers (thumb, index finger, middle finger, ring finger, and little finger) of the hand holding the bowstring.

[0056] The key points of a bow and arrow include at least one key point on the bow arm and at least one key point on the bowstring.

[0057] Facial key points include at least one upper eyelid key point and at least one lower eyelid key point.

[0058] Preferably, the key points include hand key points, which include the wrist point of the hand holding the bow wall and the five fingers of the hand holding the bowstring, the four joint points of each of the index, middle, ring, and little fingers (tip, dip, interphalangeal joint, and metacarpophalangeal joint), and the three joint points of the thumb, for a total of 20 key points.

[0059] Of course, in another implementation of this application, in addition to the above 20 key points, another joint point can be set between the metacarpophalangeal joint of the thumb of the hand holding the bowstring and the wrist joint, resulting in a total of 21 key points.

[0060] Preferably, the key points include at least one of the facial key points and the bowstring key point.

[0061] Furthermore, image detection processing is performed on multiple frames of the first image to obtain the key point position information of each key point in each frame of the first image. Based on the frame order of the first image (e.g., 1, 2, ..., n), a temporal sequence of position information of each key point (e.g., a, b, ..., k) is generated to obtain the temporal sequence of position information of the key points, which is used for subsequent analysis of the archery action stage.

[0062] The time sequence of the location information of the key points is as follows:

[0063]

[0064] in, This refers to the position information of key point a in the first image of frame n. This refers to the position information of key point b in the first image of frame n. This refers to the position information of key point k in the first image of frame n.

[0065] Of course, the time sequence of key point location information can also be in the following form:

[0066]

[0067] in, This refers to the positional information of key points a, b, ..., k in the first frame of the first image. This refers to the positional information of key points a, b, ..., k in the first image of the second frame. This refers to the positional information of key points a, b, ..., k in the first image of the nth frame.

[0068] In the implementation of this application, the location information of the key point can be the three-dimensional coordinate information of the key point.

[0069] Next, taking key points including hand key points as an example, we will explain in detail the specific process of obtaining the key point location information of each key point in the first frame of each image.

[0070] In the implementation of this application, determining the key point location information of multiple key points in each frame of the first image based on each frame of the first image includes: inputting each frame of the first image into the key point detection model so that the key point detection model can determine the key point location information of multiple key points in each frame of the first image based on each frame of the first image.

[0071] For example, a keypoint detection model can be pre-trained based on MediaPipe to obtain a keypoint detection model, such as a hand keypoint detection model.

[0072] MediaPipe is an open-source multimedia machine learning model application framework, which includes an open-source framework for hand landmarker detection to generate hand landmarker detection models.

[0073] The hand keypoint detection model consists of a palm detection model and a hand keypoint detection model that work together. Taking each frame of the first image as input, the first frame of the first image is detected first to generate the keypoint position information corresponding to the first frame of the first image. Then the second frame of the first image is detected to generate the keypoint position information corresponding to the second frame of the first image, and so on, until the detection of all frames of the first image is completed, and a time sequence of keypoint position information corresponding to the keypoints of the first images of multiple frames is obtained. The time sequence of keypoint position information includes the keypoint position information of each keypoint corresponding to each frame of the first image arranged in the forward time sequence of each frame of the first image.

[0074] It should be noted that because the first image contains the complete archery action, the number and type of key points detected in each frame of the first image are the same, and the position information of the same key points may be the same or different in different frames of the first image.

[0075] Furthermore, based on the key point location information of each key point corresponding to each frame of the first image included in the key point location information time sequence, the first archery action recognition processing is performed to obtain the first archery action stage recognition result corresponding to multiple frames of the first image.

[0076] In the implementation method of this application, such as Figure 2 As shown, based on the key point position information of each key point corresponding to the first image in each frame, the first archery action recognition process is performed to obtain the recognition results of the first archery action stage corresponding to multiple frames of the first image, including the following steps.

[0077] S210, based on the key point position information of each key point corresponding to each frame of the first image, determine the distance information between each key point corresponding to each frame of the first image, and obtain the distance change trend of key points corresponding to multiple frames of the first image based on the distance information.

[0078] For example, the hand keypoint detection model determines the distance information between each pair of keypoints in each frame of the first image based on the keypoint location information of each pair of keypoints in the time sequence of keypoint location information, thereby obtaining multiple distance information between each pair of keypoints in multiple frames of the first image, and obtaining the distance change trend of each keypoint in multiple frames of the first image based on the multiple distance information of each keypoint.

[0079] Specifically, taking the index fingertip (as an example of a hand keypoint) and the wrist point (as another example of a hand keypoint) as examples, the Euclidean distance between the two points is calculated:

[0080]

[0081] in, The distance information between keypoint i and keypoint w corresponding to the first image in frame t is given by the following: The position information of keypoint i in the first image in frame t is given by the following: The location information of key point w in the first image of frame t is ( ).

[0082] In this way, the distance information of key point i (e.g., the tip of the index finger) and key point w (e.g., the wrist point) in multiple frames of the first image is obtained, so as to obtain multiple distance information of key point i and key point w corresponding to multiple frames of the first image, thereby analyzing the change trend of the distance between the two key points in the time dimension in multiple frames of the first image, and obtaining the distance change trend of the two key points.

[0083] In this way, the trend of distance changes between all key points can be obtained.

[0084] The distance change trend of each key point can be specifically represented as a time (i.e., frame)-distance change graph between each key point.

[0085] S220, based on the distance change trend of key points corresponding to multiple first images, obtain the first archery action stage corresponding to the first target frame image in the multiple first images and the first confidence level indication information corresponding to the first archery action stage.

[0086] For example, the hand key point detection model divides the multiple first images into multiple first images based on the distance change trend of key points corresponding to the first target frame image, and obtains the first archery action stage corresponding to the first target frame image and the first confidence indication information corresponding to the first archery action stage.

[0087] The first target frame image comprises at least one group, and each group includes at least one first image. The first confidence level indication information corresponding to each group of first target frame images is the average of the first confidence level indication information of each first image included in the first target frame image. The first confidence level indication information of each first image is the average of the first confidence level indication information of each key point in each first image. That is, the first confidence level indication information of each key point in each first image can specifically be the score of the archery action in each first image included in the first target frame image as the first archery action stage, or it can be the probability information of the archery action in each first image included in the first target frame image as the first archery action stage.

[0088] Furthermore, in the implementation of this application, if the distance change of the key point corresponding to the first target frame image in the multi-frame first images is determined to satisfy the first change condition based on the distance change trend of the key points corresponding to the multi-frame first images, then the first archery action stage corresponding to the first target frame image is determined to be the bow-drawing stage.

[0089] For example, the first change condition is that the difference between the distances of the key points corresponding to the first target frame image in the distance change trend is the same between every two frames, and the difference is greater than zero and less than a preset first difference threshold. That is, the first change condition is that the distance corresponding to the target frame in the distance change trend increases steadily frame by frame. The steady increase in the inter-frame distance indicates that the bowstring is being steadily drawn, indicating that the bowstring is currently being pulled back.

[0090] Therefore, the first image in the multi-frame first image whose distance increases steadily frame by frame (that is, satisfying the first change condition) is determined as the first target frame image, and the first archery action stage corresponding to the first target frame image is the bow-drawing stage.

[0091] If, based on the trend of key point changes in the key points corresponding to the first target frame image in multiple first images, it is determined that the distance changes of the key points in the first target frame image in multiple first images satisfy the second change condition, then the first archery action phase corresponding to the first target frame image is determined to be the charging phase.

[0092] For example, the second change condition is that the difference between the distances of the key points corresponding to the first target frame image in the distance change trend tends to zero between every two frames. That is, the second change condition is that the distances corresponding to the target frames in the distance change trend are relatively stable and the distance differences tend to zero. The relatively stable distances between frames indicate that the bowstring is also in a taut state, indicating that it is currently accumulating power.

[0093] Therefore, the first image in the multi-frame first image whose distance is relatively stable (that is, satisfying the second change condition) is determined as the first target frame image, and the first archery action phase corresponding to the first target frame image is the charging phase.

[0094] If, based on the key point change trend of the key points corresponding to the first target frame image in multiple first images, it is determined that the distance change of the key points in the first target frame image in multiple first images satisfies the third change condition, then the first archery action stage corresponding to the first target frame image is determined to be the release stage.

[0095] For example, the third change condition is that the difference between the distances of key points corresponding to the first target frame image in the distance change trend and every two frames is greater than a preset second difference threshold. That is, the third change condition is that the distance corresponding to the target frame in the distance change trend shows a sharp jump (for example, an increase in the distance slope corresponding to multiple frames). A sharp jump indicates an increase in the inter-frame speed of the fingertip, which means that the finger has released the bowstring, indicating that the bow is currently being released.

[0096] Therefore, the first image in the multi-frame first image that has a drastic change in distance (that is, meets the second change condition) is determined as the first target frame image, and the first archery action phase corresponding to the first target frame image is the release phase.

[0097] Therefore, based on the distance change trend of each key point, the first change condition, the second change condition, and the third change condition, it is possible to identify the first archery action stage corresponding to each first target frame image in multiple first images.

[0098] Furthermore, the hand key point detection model also scores or performs probability calculations on the probability that each frame of the first target frame image represents the first archery action stage, so as to obtain the first confidence level indication information that the archery action in each frame of the first target frame image represents the first archery action stage.

[0099] S230, based on the first archery action stage corresponding to the first target frame image and the first confidence level indication information corresponding to the first archery action stage, the recognition result of the first archery action stage corresponding to multiple frames of the first image is obtained.

[0100] For example, if the multiple first images include only one set of first target frame images, then the first archery action stage corresponding to the set of first target frame images and the first confidence level indication information corresponding to the first archery action stage are used as the recognition result of the first archery action stage corresponding to the multiple first images.

[0101] If the multiple first images include multiple sets of first target frame images, then the first archery action stage corresponding to each set of first target frame images and the first confidence level indication information corresponding to the first archery action stage are used as the recognition result of the first archery action stage corresponding to the multiple first images.

[0102] Next, step S300 is executed, which uses an image classification model to perform second archery action recognition processing based on multiple frames of the first image, and obtains the recognition results of the second archery action stage corresponding to the multiple frames of the first image.

[0103] For example, to enhance the fault tolerance capability for abnormal states, an image classification model (MobileNetV) is introduced. Based on the image classification model, the second archery action recognition process is performed according to multiple frames of the first image to obtain the recognition result of the second archery action stage corresponding to the multiple frames of the first image.

[0104] In this application, the second archery action recognition process is performed based on multiple frames of first images using an image classification model to obtain the recognition result of the second archery action stage corresponding to the multiple frames of first images. This includes: cropping the target region in the multiple frames of first images to obtain multiple frames of second images, and inputting the multiple frames of second images into the image classification model so that the image classification model can perform the second archery action recognition process based on the multiple frames of second images. The target region includes at least one of the hand region, face region, and bow and arrow equipment region.

[0105] For example, at least one of the hand region, face region, and bow and arrow equipment region in the first image is cropped to obtain the target region, so as to obtain multiple frames of second images based on the target region of multiple frames of the first image.

[0106] For example, each frame of the second image includes either a hand and face area or a hand and bow and arrow equipment area.

[0107] Furthermore, in another implementation of this application, multiple frames of the first image are directly input into the image classification model so that the image classification model performs the second archery action recognition processing based on the multiple frames of the first image.

[0108] Among them, the image classification model learns the static image features of typical actions through the training set, and can accurately distinguish still frames.

[0109] Furthermore, such as Figure 3 As shown, in the implementation of this application, based on an image classification model, the second archery action recognition process is performed according to multiple frames of the first image to obtain the recognition result of the second archery action stage corresponding to the multiple frames of the first image, including the following steps.

[0110] S310, based on the image classification model, performs second archery action recognition processing on multiple frames of the first image to obtain the archery action category and archery action category confidence indication information for each frame of the first image.

[0111] For example, multiple frames of the first image are input into the image classification model, which then performs the second archery action recognition process and outputs the archery action category and the corresponding archery action category confidence indication information for each frame.

[0112] The archery action categories include drawing the bow, charging, releasing, idle, and abnormal. One frame of image corresponds to one or more categories, and each category has a confidence index for the archery action category.

[0113] The confidence index information of the archery action category corresponding to each frame image is obtained based on the softmax function and is used for subsequent fusion judgment.

[0114] The confidence index information for the archery action category corresponding to each frame image can specifically be the probability information corresponding to each archery action category. (This can also be the score information for each archery action category.)

[0115] S320, based on the archery action category and archery action category confidence indication information of each frame of the first image, obtain the second archery action stage corresponding to the second target frame image in the multi-frame first image and the second confidence indication information corresponding to the second archery action stage.

[0116] For example, multiple consecutive frames of first images of the same archery action category are taken as a group of second target frame images to obtain the second archery action stage corresponding to each group of second target frame images, and the average value of the archery action category confidence indication information corresponding to the multiple consecutive frames of first images of the same archery action category is taken as the second confidence indication information of the second archery action corresponding to the group of second target frame images.

[0117] S330, based on the second archery action stage corresponding to the second target frame image and the second confidence level indication information corresponding to the second archery action stage, the recognition result of the second archery action stage corresponding to multiple frames of the first image is obtained.

[0118] For example, if the multiple first images include only one set of second target frame images, then the second archery action stage corresponding to the set of second target frame images and the second confidence level indication information corresponding to the second archery action stage are used as the recognition result of the second archery action stage corresponding to the multiple first images.

[0119] If the multiple first images include multiple sets of second target image frames, then the second archery action phase corresponding to each set of second target image frames and the second confidence level indication information corresponding to the second archery action phase are used as the recognition result of the second archery action phase corresponding to the multiple first images.

[0120] It should be noted that the method for recognizing the second archery action based on multiple frames of the second image is the same as the method based on multiple frames of the first image. Since the second archery action recognition is performed based on multiple frames of the second image, only the key regions are retained in each frame, resulting in more accurate recognition.

[0121] Next, step S400 is executed to obtain the target archery action stage recognition results corresponding to multiple frames of the first image based on the recognition results of the first archery action stage and the recognition results of the second archery action stage.

[0122] In the implementation of this application, the target archery action phase identification result includes the target archery action phase, the start time of the target archery action phase, the duration of the target archery action phase, and the corresponding stability information of the target archery action phase.

[0123] In the implementation method of this application, such as Figure 4 As shown, based on the recognition results of the first and second archery action stages, the target archery action stage recognition results corresponding to multiple frames of the first image are obtained, including the following steps.

[0124] S410, if it is determined that the first target frame image and the second target frame image have the same frame, and the first archery action phase corresponding to the first target frame image and the second archery action phase corresponding to the second target frame image are the same archery action phase, the first weight coefficient corresponding to the first confidence indication information and the second weight coefficient corresponding to the second confidence indication information are determined according to the second confidence indication information.

[0125] For example, the first target frame images and the second target frame images in the recognition results of the first archery action stage are compared. If it is determined that the first target frame image and the second target frame image have the same frame, it indicates that it is the recognition of the same first image in multiple first images. Furthermore, if it is determined that the first archery action stage corresponding to the first target frame image and the second archery action stage corresponding to the second target frame image are the same archery action stage, it indicates that the first archery action stage and the second archery action stage corresponding to the same first image are the same.

[0126] Wherein, if the second confidence level indication information is greater than the preset first confidence level threshold, then the target frame image is the second target frame image, and the first weight coefficient... Less than the second weighting coefficient If the second confidence level indication is less than the preset second confidence level threshold, then the target frame image is the first target frame image, and the first weighting coefficient is used. Greater than the second weighting coefficient .

[0127] In this implementation, if the second confidence level indicator information corresponding to the second archery action phase identified by the image classification model is greater than a preset first confidence level threshold (e.g., 0.85), it indicates that the image classification confidence is high, and image classification is prioritized. .

[0128] If the second confidence level indicator is less than the preset second confidence level threshold (e.g., 0.4) or the image classification model output is null, then the key point recognition result (the first archery action recognition result) will be used as the primary indicator. .

[0129] The preset second confidence threshold is less than the preset first confidence threshold.

[0130] Furthermore, if the second confidence level indication information is greater than or equal to a preset second confidence level threshold and less than or equal to a preset first confidence level threshold, then If the first confidence level indication information is greater than or equal to the second confidence level indication information, then the target frame image is the first target frame image; if the first confidence level indication information is less than the second confidence level indication information, then the target frame image is the second target frame image.

[0131] It should be noted that, in the implementation method of this application, it can be... ,but .

[0132] Furthermore, it can also be used for .

[0133] Furthermore, in this implementation, a first weighting coefficient and a second weighting coefficient can be determined based on the first confidence level indication information. If the first confidence level indication information is greater than a preset first confidence level threshold, then the target frame image is the first target frame image, and the first weighting coefficient... Regarding the second weighting coefficient, if the first confidence level indication information is less than the preset second confidence level threshold, then the target frame image is the second target frame image, and the first weighting coefficient... The second weighting coefficient.

[0134] Furthermore, a first weighting coefficient and a second weighting coefficient can be determined based on the first confidence level indication information and the second confidence level indication information. If the first confidence level indication information is greater than the second confidence level indication information, then the target frame image is the first target frame image, and the first weighting coefficient is greater than the second weighting coefficient. If the first confidence level indication information is less than the second confidence level indication information, then the target frame image is the second target frame image, and the first weighting coefficient is less than the second weighting coefficient.

[0135] S420: Based on the first confidence level indication information, the second confidence level indication information, the first weighting coefficient, and the second weighting coefficient, the target confidence level indication information corresponding to the target frame image is obtained.

[0136] For example, for each set of target frame images, the target confidence indication information corresponding to the target frame image is obtained based on the first confidence indication information, the second confidence indication information, the first weight coefficient, and the second weight coefficient.

[0137] Taking the first confidence level indicator and the second confidence level indicator as scores as an example, the target confidence level indicator of the target frame image is obtained in the following way:

[0138]

[0139] in, Target confidence indication information for the target frame image. This is the second weighting coefficient. The first judgment score (as an example of the second confidence level information). As the first weighting coefficient, This is the second judgment score (as an example of the first confidence level indication information). , The fusion coefficient is adjustable in the range of [0, 1].

[0140] when At that time, the target confidence indication information of the target frame image is obtained in the following way:

[0141]

[0142] Thus, the target confidence indication information of the target frame image is obtained.

[0143] S430, if the target confidence indication information is greater than the preset third confidence threshold, determine the first archery action phase or the second archery action phase as the target archery action phase.

[0144] For example, if the target confidence indication information is greater than a preset third confidence threshold (e.g., 0.8), the target archery action phase is output as either the first archery action phase or the second archery action phase.

[0145] S440, based on the timestamp information corresponding to the second target frame image, obtain the start time and duration of the target archery action phase.

[0146] For example, the timestamp information of the target frame image is recorded simultaneously. The timestamp information with the smallest time among the multiple frames included in the target frame image is taken as the start time of the target archery action phase. The duration is obtained based on the timestamp information with the largest time and the timestamp information with the smallest time among the multiple frames included in the target frame image. In this way, the target archery action phase-time mapping table is obtained.

[0147] S450, determine the detection stability information of the target archery action phase based on the first confidence indication information and the second confidence indication information corresponding to the target frame image.

[0148] For example, the detection stability information corresponding to the target frame image is obtained by weighting and averaging the first confidence indicator information corresponding to the first target frame image and the second confidence indicator information corresponding to the second target frame image. This is the detection stability information during the target archery action phase. For example, it is the detection stability score.

[0149] Furthermore, if it is determined that the first target frame image and the second target frame image have the same frame, and the first archery action stage corresponding to the first target frame image and the second archery action stage corresponding to the second target frame image are not the same archery action stage, then the target archery action stage corresponding to the target frame image is determined to be another stage.

[0150] If the target confidence level indication information is determined to be less than the preset third confidence level threshold, the second anomaly information is obtained.

[0151] If it is the first abnormal information or the second abnormal information, then the target archery action stage corresponding to the target frame image is determined to be another stage.

[0152] Output the identified archery action phases (e.g., drawing the bow / charging / releasing / others), the switching time points of the archery action phases, the duration of the archery action phases, and the detection stability score.

[0153] In this way, we can obtain the target archery action phase and the switching time (i.e., start time), duration, and detection stability information of one or more target archery action phases in the first image of multiple frames.

[0154] The archery action recognition method provided in this application is a motion-sensing archery action recognition method based on the temporal changes of key points detected by MediaPipe and a self-trained AI classification model (i.e., an intelligent image classification model). For example, it is an archery action stage recognition method based on a combination of temporal changes of hand key points and image classification. This method can accurately identify various key actions of the user during archery and provide real-time feedback on archery action information at each stage, improving the archery interaction experience. It also allows users to adjust their archery actions at different stages based on the archery action stage recognition results, thereby making the archery results more accurate.

[0155] The archery motion recognition method provided in this application can be applied to electronic devices.

[0156] Figure 5 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application. Figure 5 As shown, the electronic device may include: transceiver 121, processor 122, and memory 123.

[0157] The processor 122 executes computer execution instructions stored in the memory, causing the processor 122 to perform the technical solution of the archery action recognition method in the above embodiments. The processor 122 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital data processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0158] The memory 123 is connected to the processor 122 via the system bus and completes communication between them. The memory 123 is used to store computer program instructions.

[0159] For example, and not as a limitation, memory 123 may include a hard disk drive (HDD), a floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 123 may include removable or non-removable (or fixed) media. Where appropriate, memory 123 may be internal or external to the integrated gateway device. In a particular embodiment, memory 123 is non-volatile solid-state memory. In a particular embodiment, memory 123 includes read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only ROM (PROM), an erasable programmable read-only ROM (EPROM), an electrically erasable programmable read-only ROM (EEPROM), an electrically alterable read-only ROM (EAROM), or flash memory, or a combination of two or more of these. Transceiver 121 can be used to obtain the task to be run and its configuration information.

[0160] The system bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.

[0161] Furthermore, the electronic device can be, for example, a computer, a mobile phone, a server, or other electronic devices.

[0162] This application also provides a chip for executing instructions, which is used to execute the archery action recognition method described in the above embodiments.

[0163] This application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on the processor of a computer device, the processor of the computer device executes the technical solution of the archery action recognition method described above.

[0164] In some possible implementations, various aspects of the methods provided in this application can also be implemented as a program product, which includes program code. When the program product is run on the processor of a computer device, the program code is used to cause the processor of the computer device to perform the steps in the methods of the various exemplary implementations of this application described above. For example, the computer device can execute the archery action recognition method described in the embodiments of this application.

[0165] The program product may take the form of any combination of one or more readable media. A readable medium may be a readable data medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CDROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0166] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium, and when the at least one processor executes the computer program, it can implement the technical solution of the archery action recognition method in the above embodiments.

[0167] It should be noted that, in addition to the specific embodiments described above, those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Although the description of this application is presented in conjunction with preferred embodiments, this does not mean that the features of this invention are limited to this implementation. On the contrary, the purpose of describing the invention in conjunction with the implementation is to cover other options or modifications that may be derived from this application. To provide a thorough understanding of this application, many specific details are included in the above description, and this application may also be implemented without using these details. Furthermore, to avoid confusion or obscuring the focus of this application, some specific details will be omitted in the description. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0168] It should be noted that in this specification, similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0169] It should be noted that the terms "first" and "second" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0170] It should be noted that some structural or methodological features may be shown in the accompanying drawings in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be necessary. Rather, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. Furthermore, including structural or methodological features in a particular figure does not imply that such features are required in all embodiments, and in some embodiments, these features may be omitted or may be combined with other features.

[0171] Although this application has been illustrated and described with reference to certain preferred embodiments, those skilled in the art should understand that the above description is a further detailed explanation of the application in conjunction with specific implementations, and should not be construed as limiting the specific implementation of the application to these descriptions. Those skilled in the art can make various changes in form and detail, including some simple deductions or substitutions, without departing from the spirit and scope of this application.

Claims

1. A method for recognizing archery actions, characterized in that, The method includes: Determine the first images of multiple frames related to the archery action during the user's archery process; Based on each frame of the first image, determine the key point location information of multiple key points in each frame of the first image, and based on the key point location information of each key point corresponding to each frame of the first image, determine the distance information between each key point corresponding to each frame of the first image, and obtain the distance change trend of the key points corresponding to the multiple frames of the first image based on the distance information. Based on the distance change trend of the key points corresponding to the multi-frame first images, a first archery action stage corresponding to the first target frame image in the multi-frame first images and a first confidence level indication information corresponding to the first archery action stage are obtained. Specifically, if the distance change of the key points corresponding to the first target frame image in the multi-frame first images satisfies a first change condition based on the distance change trend of the key points corresponding to the key points in the multi-frame first images, then the first archery action stage corresponding to the first target frame image is determined to be a drawing-the-bow stage; if the distance change of the key points corresponding to the first target frame image in the multi-frame first images satisfies a second change condition based on the key point change trend of the key points in the multi-frame first images, then the first archery action stage corresponding to the first target frame image is determined to be a charging-up stage; if the distance change of the key points corresponding to the first target frame image in the multi-frame first images satisfies a third change condition based on the key point change trend of the key points in the multi-frame first images, then the first archery action stage corresponding to the first target frame image is determined to be a releasing stage. Based on the first archery action stage corresponding to the first target frame image and the first confidence indication information corresponding to the first archery action stage, the recognition result of the first archery action stage corresponding to the multiple first frames of images is obtained; and Based on the image classification model, the second archery action recognition process is performed on the multiple frames of the first image to obtain the recognition result of the second archery action stage corresponding to the multiple frames of the first image; Based on the first archery action phase recognition result and the second archery action phase recognition result, the target archery action phase recognition result corresponding to the multi-frame first image is obtained.

2. The archery action recognition method according to claim 1, characterized in that, Based on the image classification model, the second archery action recognition process is performed on the multiple frames of the first image to obtain the recognition result of the second archery action stage corresponding to the multiple frames of the first image, including: Based on the image classification model, the second archery action recognition process is performed on the multiple frames of the first image to obtain the archery action category and archery action category confidence indication information for each frame of the first image; Based on the archery action category and archery action category confidence indication information of the first image in each frame, the second archery action stage corresponding to the second target frame image in the multi-frame first image and the second confidence indication information corresponding to the second archery action stage are obtained. Based on the second archery action stage corresponding to the second target frame image and the second confidence level indication information corresponding to the second archery action stage, the recognition result of the second archery action stage corresponding to the multiple first images is obtained.

3. The archery action recognition method according to claim 2, characterized in that, The target archery action phase recognition result includes the target archery action phase corresponding to the target frame image. Based on the first archery action phase recognition result and the second archery action phase recognition result, the target archery action phase recognition result corresponding to the multiple first frames of images is obtained, including: If it is determined that the first target frame image and the second target frame image have the same frame, and the first archery action phase corresponding to the first target frame image and the second archery action phase corresponding to the second target frame image are the same archery action phase, then according to the second confidence level indication information, the first weight coefficient corresponding to the first confidence level indication information and the second weight coefficient corresponding to the second confidence level indication information are determined. If the second confidence level indication information is greater than the preset first confidence level threshold, then the first weight coefficient is less than the second weight coefficient, and the target frame image is the second target frame image. If the second confidence level indication information is less than the preset second confidence level threshold, then the first weight coefficient is greater than the second weight coefficient, and the target frame image is the first target frame image. Based on the first confidence level indication information, the second confidence level indication information, the first weight coefficient, and the second weight coefficient, the target confidence level indication information corresponding to the target frame image is obtained; If the target confidence indication information is determined to be greater than a preset third confidence threshold, the first archery action phase or the second archery action phase is determined to be the target archery action phase corresponding to the target frame image.

4. The archery action recognition method according to claim 3, characterized in that, The multiple first images have corresponding timestamp information. The target archery action phase recognition result also includes the start time, duration, and stability information corresponding to the target archery action phase. Based on the first and second archery action phase recognition results, the target archery action phase recognition result corresponding to the multiple first images is obtained, including: The start time and duration of the target archery action phase are obtained based on the timestamp information corresponding to the target frame image; The detection stability information of the target archery action phase is determined based on the first confidence indication information and the second confidence indication information corresponding to the target frame image.

5. The archery action recognition method according to claim 4, characterized in that, Determine multiple first images related to the archery action during the user's archery process, including: Acquire video of the archery actions related to the user's archery movements during the archery process, and obtain the multi-frame first image based on the archery action video; or Multiple consecutive images related to the archery action during the user's archery process are acquired to obtain the first multi-frame image.

6. The archery action recognition method according to claim 5, characterized in that, Based on each frame of the first image, determine the key point location information of multiple key points in each frame of the first image, including: Each frame of the first image is input into the keypoint detection model so that the keypoint detection model determines the keypoint location information of multiple keypoints in each frame of the first image, including hand keypoints.

7. The archery action recognition method according to any one of claims 1-6, characterized in that, Based on an image classification model, the second archery action is identified using the multiple frames of the first image, including: The target region in the multiple first images is cropped to obtain multiple second images. These multiple second images are then input into the image classification model, enabling the model to perform a second archery action recognition based on the multiple second images. The target region includes at least one of a hand region, a face region, and a bow and arrow equipment region; or The first multi-frame image is input into the image classification model so that the image classification model can perform a second archery action recognition process based on the first multi-frame image.

8. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes the computer execution instructions stored in the memory to perform the archery action recognition method as described in any one of claims 1-7.

Images