A method, system, device, and medium for jump rope counting

Abstract

This invention discloses a method, system, device, and medium for counting jump ropes, aiming to solve the problems of low efficiency and poor accuracy in existing jump rope counting methods. The method includes: analyzing collected student jump rope video frames using a pre-trained target detection model and a jump rope counting detection model; the jump rope counting detection model, based on a network structure including a motion analysis module, extracts multi-scale features of key human body points and outputs a heatmap of human skeletal points; the target detection model outputs the position information of the jump rope; by judging whether the spatial positional relationship of specific human skeletal points (such as wrists, hips, and ankles) in a continuous frame sequence conforms to a preset jumpable state, and combining the transition between the jump rope's airborne and falling states relative to all human skeletal points, a valid jump rope determination and counting is completed. This invention achieves automatic and accurate counting of jump rope activities in physical education classes or tests, improving efficiency and objectivity.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and intelligent sports monitoring technology, and in particular to a method, system, device and medium for counting jump ropes. Background Technology

[0002] Rope skipping is a fundamental activity in physical education teaching and testing in primary and secondary schools, and the accuracy of its counting directly affects the fairness and efficiency of the assessment. Currently, the mainstream counting method still relies on manual visual counting, which suffers from low efficiency, fatigue, and large subjective errors, especially in group tests where labor costs are high. Some electronic counting ropes use built-in sensors for counting, but they are prone to miscounting due to non-standard wrist movements or equipment vibration, resulting in insufficient reliability.

[0003] With the expanding application of artificial intelligence technology in sports assessment, motion analysis based on computer vision provides a new solution for contactless automatic counting. However, rope skipping is characterized by its fast pace, complex human-rope interaction, and numerous background interferences, making it difficult for general motion recognition models to directly achieve highly robust counting. Currently, there is a lack of a complete technical solution that can accurately and stably integrate human posture estimation and rope skipping target detection, and define a clear counting logic accordingly.

[0004] Therefore, a method, system, device, and medium for counting jump ropes are proposed to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method for counting jump ropes to solve the problems mentioned in the background art.

[0006] The first aspect of this invention is achieved through the following technical solution:

[0007] A method for counting jump ropes, the method comprising the following steps:

[0008] S1. Acquire video data containing rope skippers and extract a continuous sequence of image frames from it;

[0009] S2. Input the image frame sequence into the pre-trained rope skipping detection model to obtain the predicted point information of human skeletal key points in each image frame; the rope skipping detection model includes a motion analysis module for extracting multi-scale features of human key points;

[0010] S3. Synchronously input the image frame sequence into the pre-trained target detection model to obtain the position information of the jump rope in each image frame;

[0011] S4. Based on the predicted position information of the key points of the human skeleton, determine whether the jumper is in a preparatory jumping state.

[0012] S5. In a continuous image frame sequence, detect the process of the jump rope position changing from a first state to a second state relative to the whole human body region; wherein, the first state is that the jump rope is located above the whole human body region, and the second state is that the jump rope passes under the whole human body region.

[0013] S6. When a complete transition from the first state to the second state is detected, and the jumper was in the preparatory jump state before the transition, it is determined that a valid jump rope action has been completed, and the count is performed.

[0014] Furthermore, in step S2, the jump rope counting detection model obtains the predicted point information in the following way: the model outputs a heat map of a preset number of key points, analyzes the peak coordinates of each heat map, and takes the local maximum value point as the predicted point of the corresponding human skeletal key point.

[0015] Furthermore, in step S4, determining whether the jumper is in a preparatory jumping state specifically involves determining whether the following conditions are met simultaneously: the nose is higher than the left and right wrists; the left and right wrists are higher than the left and right hips; the left and right hips are higher than the left and right ankles; and the left and right wrists are closest to the hips on the same side.

[0016] Furthermore, the jump rope counting detection model is trained through the following steps:

[0017] S11. Collect images containing various human postures and rope skipping images to construct a training dataset;

[0018] S12. Mark the key points of human skeleton on the human images in the training dataset, and mark the jump rope position information on the jump rope images;

[0019] S13. Using the labeled dataset, train the jump rope counting detection model; wherein the jump rope counting detection model is trained based on a neural network structure that includes the motion parsing module.

[0020] Furthermore, the motion analysis module includes:

[0021] The first processing branch performs layer normalization on the input feature map and then multiplies it with the first learnable weight factor;

[0022] The second processing branch multiplies the input feature map by a second learnable weight factor;

[0023] A feature fusion layer is used to add the outputs of the first processing branch and the second processing branch, and then perform downsampling processing;

[0024] A multi-scale feature extraction layer is used to perform convolution operations with multiple different kernel sizes on the output of the feature fusion layer in parallel to obtain feature maps at multiple scales.

[0025] An adaptive weight fusion layer is used to assign learnable weights to the feature maps at multiple scales and perform weighted summation to obtain fused multi-scale features.

[0026] The upsampling and residual connection layer is used to add the fused multi-scale features, after activation function and upsampling processing, to the input feature map, and output the final features of the motion parsing module.

[0027] Furthermore, the jump rope counting detection model also includes a shallow feature extraction module, a middle feature extraction module, and a deep feature extraction module connected in sequence; wherein, the shallow feature extraction module contains two parallel branches, each branch containing multiple motion parsing layers in series; the middle feature extraction module and the deep feature extraction module each contain multiple feature extraction blocks in series, each feature extraction block having multiple parallel branches, each branch containing multiple motion parsing layers in series, and the outputs of each branch are summed.

[0028] The second aspect of the present invention is achieved through the following technical solution:

[0029] A jump rope counting system for implementing the method of the first aspect, the system comprising:

[0030] The data acquisition module is used to acquire video data containing rope jumpers and extract image frame sequences;

[0031] The key point detection module includes a pre-trained jump rope counting detection model, which is used to receive the image frame sequence and output the predicted point information of human skeleton key points.

[0032] The jump rope detection module includes a pre-trained target detection model, which is used to receive the image frame sequence and output the position information of the jump rope.

[0033] The status judgment module is used to determine whether the jumper is in a preparatory jumping state based on the predicted point information;

[0034] The action recognition and counting module is used to identify the process of the jump rope changing from the first state to the second state in continuous image frames based on the position information of the jump rope and the predicted position information of the key points of the human skeleton, and to complete the recognition and counting of effective jump rope actions in combination with the judgment of the preparatory jump state.

[0035] Furthermore, the jump rope counting detection model is obtained by: collecting and labeling training data containing key points of the human body and the position of the jump rope; using the labeled data to train a neural network structure containing a motion parsing module to obtain the jump rope counting detection model.

[0036] The third aspect of the present invention is achieved through the following technical solution:

[0037] A jump rope counting device, comprising:

[0038] At least one processor;

[0039] The memory is communicatively connected to the at least one processor;

[0040] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the method described in the first aspect.

[0041] The fourth aspect of the present invention is achieved through the following technical solution:

[0042] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0043] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0044] 1. This invention can completely replace manual counting. A single camera can achieve unattended automatic counting, which greatly improves testing efficiency and is suitable for large-scale synchronous testing.

[0045] 2. This invention uses a dedicated rope skipping counting and detection model (including a motion analysis module) to accurately locate key points on the human body. By combining the dual criteria of the preparatory state and the rope skipping state transition, it effectively filters out interference such as invalid swings, resulting in high counting accuracy.

[0046] 3. This invention only requires ordinary camera equipment, does not require students to wear any sensors, is easy to deploy, does not affect the exercise process, and provides a user-friendly experience.

[0047] 4. The entire counting process of this invention is based on algorithms and data, the results are objective, and process data (such as keyframes and status markers) can be recorded for post-event review and traceability, thus resolving scoring disputes. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the overall process of the jump rope counting method provided in the embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram of 17 key skeletal points of the human body as defined in this embodiment of the invention, and a schematic diagram of the preparatory state judgment.

[0051] Figure 3 This is a schematic diagram of the jump rope counting logic in an embodiment of the present invention, showing the two states of the jump rope relative to the overall area of ​​the human body: above and below.

[0052] Figure 4 This is a schematic diagram of the overall network structure of the jump rope counting detection model in this embodiment of the invention.

[0053] Figure 5 This is a detailed structural diagram of the motion analysis module in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0055] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0056] It should be understood that the invention can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0057] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising” and / or “including,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0058] To fully understand this invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by this invention. Optional embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.

[0059] See Figures 1-5 A method for counting jump ropes, the method comprising the following steps:

[0060] S1. Acquire video data containing rope skippers and extract a continuous sequence of image frames from it;

[0061] S2. Input the image frame sequence into the pre-trained rope skipping detection model to obtain the predicted point information of human skeletal key points in each image frame; the rope skipping detection model includes a motion analysis module for extracting multi-scale features of human key points;

[0062] S3. Synchronously input the image frame sequence into the pre-trained target detection model to obtain the position information of the jump rope in each image frame;

[0063] S4. Based on the predicted position information of the key points of the human skeleton, determine whether the jumper is in a preparatory jumping state.

[0064] S5. In a continuous image frame sequence, detect the process of the jump rope position changing from a first state to a second state relative to the whole human body region; wherein, the first state is that the jump rope is located above the whole human body region, and the second state is that the jump rope passes under the whole human body region.

[0065] S6. When a complete transition from the first state to the second state is detected, and the jumper was in the preparatory jump state before the transition, it is determined that a valid jump rope action has been completed, and the count is performed.

[0066] In step S2, the jump rope counting detection model obtains the predicted point information in the following way: the model outputs a heat map of a preset number of key points, analyzes the peak coordinates of each heat map, and takes the local maximum value point as the predicted point of the corresponding human skeletal key point.

[0067] In step S4, determining whether the jumper is in a preparatory jumping state specifically involves determining whether the following conditions are met simultaneously: the nose is higher than the left and right wrists; the left and right wrists are higher than the left and right hips; the left and right hips are higher than the left and right ankles; and the left and right wrists are closest to the hips on the same side.

[0068] The jump rope counting detection model is trained through the following steps:

[0069] S11. Collect images containing various human postures and rope skipping images to construct a training dataset;

[0070] S12. Mark the key points of human skeleton on the human images in the training dataset, and mark the jump rope position information on the jump rope images;

[0071] S13. Using the labeled dataset, train the jump rope counting detection model; wherein the jump rope counting detection model is trained based on a neural network structure that includes the motion parsing module.

[0072] The motion analysis module includes:

[0073] The first processing branch performs layer normalization on the input feature map and then multiplies it with the first learnable weight factor;

[0074] The second processing branch multiplies the input feature map by a second learnable weight factor;

[0075] A feature fusion layer is used to add the outputs of the first processing branch and the second processing branch, and then perform downsampling processing;

[0076] A multi-scale feature extraction layer is used to perform convolution operations with multiple different kernel sizes on the output of the feature fusion layer in parallel to obtain feature maps at multiple scales.

[0077] An adaptive weight fusion layer is used to assign learnable weights to the feature maps at multiple scales and perform weighted summation to obtain fused multi-scale features.

[0078] The upsampling and residual connection layer is used to add the fused multi-scale features, after activation function and upsampling processing, to the input feature map, and output the final features of the motion parsing module.

[0079] The jump rope counting detection model further includes a shallow feature extraction module, a middle feature extraction module, and a deep feature extraction module connected in sequence. The shallow feature extraction module contains two parallel branches, each containing multiple motion parsing layers in series. The middle and deep feature extraction modules each contain multiple feature extraction blocks in series. Each feature extraction block has multiple parallel branches, each containing multiple motion parsing layers in series, and the outputs of each branch are summed.

[0080] For example, for data collection and annotation, images and videos of students engaging in physical activities and skipping rope under different lighting, backgrounds, and clothing conditions can be collected extensively, and keyframes can be extracted to form an image dataset. Using annotation tools, 17 key points of the human skeleton—nose, left and right eyes, left and right ears, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left knee, and left and right ankles—can be accurately annotated on the images (see...). Figure 2 ) and the bounding box of the jump rope.

[0081] For training the object detection model, YOLOv5 or a similar framework can be used, employing images labeled with jump rope bounding boxes for training, to obtain a model capable of detecting the jump rope position in an image in real time; for training the jump rope counting detection model, images labeled with human key points can be used to train the dedicated network proposed in this invention (see...). Figure 4 The training loss function of this network can be a combination of Mean Squared Error (MSE) and heatmap loss of key points to optimize the accuracy of key point prediction.

[0082] For real-time video capture and preprocessing, in practical application scenarios (such as a playground), a high-definition camera can be fixed approximately 3-5 meters directly in front of the subject, adjusting the angle to ensure the image covers the entire body and the range of rope swing. After the test begins, the system captures a video stream at a rate of 30 frames per second (fps) and caches the image frames sequentially. It can perform preprocessing on images, such as standardization and resizing.

[0083] During parallel inference and feature extraction, for the current processing frame It can be input into an already loaded jump rope counting detection model, which ultimately outputs a value of size [missing information]. A heatmap tensor is generated, where each channel corresponds to the response heatmap of a keypoint. By performing a two-dimensional Gaussian peak lookup on the heatmap of each channel, the pixel coordinates of the 17 keypoints in the frame are obtained. .

[0084] At the same time, Input to the loaded target detection model. The model outputs the detection box for the jump rope. ,in The coordinates of the top left corner The coordinates are at the bottom right corner. This can be used to calculate the longitudinal center line or upper and lower boundaries of the jump rope for status assessment.

[0085] Assessing the body's preparatory state.

[0086] Based on the key point coordinate set Then, the preparatory state judgment algorithm is executed. A specific implementation example is as follows:

[0087] Calculate necessary coordinates: Obtain the nose , left and right wrists left and right buttocks left and right ankles of Coordinates (image coordinate system, downward is positive).

[0088] Check spatial hierarchy: Determine if the following conditions are met simultaneously:

[0089] (Nose higher than wrist)

[0090] (Wrist higher than the hip on the same side)

[0091] (Hips higher than the ankle on the same side)

[0092] Check distance relationships: Calculate the left wrist. Euclidean distances to all key points, confirming its distance to the left hip. Is the distance to the right wrist the minimum (or less than the threshold)? to the right hip Is the distance the minimum?

[0093] If all the above conditions are met, the jumper in that frame is determined to be in a "prepared jump state", and a status flag is set. ,otherwise .

[0094] See Figure 3 In the process of recognizing and counting the state of the rope skipping action, the system maintains two core state variables: the rope skipping state RopeState (initially None) and the valid count flag ValidCountFlag.

[0095] Determine the overall human body region: Based on the coordinates of all 17 keypoints in the current frame, calculate its minimum bounding rectangle (or convex hull), which is defined as the "overall human body region" of this frame. This can also be simplified to taking all key points. Minimum value of coordinates and maximum value It is used to characterize the vertical span of the human body.

[0096] Determine the relative position of the jump rope:

[0097] From the jump rope detection box In the middle, get the top of the jump rope coordinate (Right now ) and bottom coordinate (Right now ).

[0098] State 1 (above) judgment: If If the bottom of the jump rope is still above the highest point of the human body, then the jump rope is considered to be above the human body and is denoted as Candidate_State1.

[0099] State 2 (below) judgment: If If the top of the jump rope is below the lowest point of the body, then the entire jump rope is considered to have passed under the body, and this is recorded as a candidate state. .

[0100] During state transitions and counting, the system analyzes consecutive frame sequences (such as the most recent 10 frames) using a sliding window approach.

[0101] When detected in multiple consecutive frames And subsequently detected in the frame sequence If so, a potential "state transition event" is recorded.

[0102] Check if at least one frame in the frames preceding the "state transition event" (e.g., the three frames before the transition start point) contains such a event. The sign is .

[0103] If the above conditions are met, and the current jump rope cycle is not locked (to prevent repeated counting of a single action), then a valid jump rope action is determined to have occurred, and the global counter is activated. Increment by 1 and enter a short counting lock period (e.g., 0.5 seconds) to avoid repeated identification during continuous rope swinging.

[0104] During the result output and display process, after the test is completed, the system displays the final counting result on the interface. Meanwhile, key data (such as key points, detection boxes, status flags, and count timestamps for each frame) can be recorded to log files or databases, supporting process playback and result verification.

[0105] Jump rope counting detection model and its motion analysis module

[0106] like Figure 4As shown, the jump rope counting detection model adopts an encoder-decoder structure with multi-branch, hierarchical feature extraction.

[0107] Input and initial convolution: The input is an RGB image (e.g., 256x192x3). It first passes through two convolutional layers (...). ), batch normalization layer ( )and Standard convolutional blocks composed of activation functions are used to increase feature dimensions and perform preliminary feature extraction.

[0108] Motion resolution layer: This is one of the core components of the model. For example... Figure 5 As shown, each motion parsing layer consists of a standard convolutional block and a motion parsing module connected in series.

[0109] Motion analysis module, such as Figure 5 As shown, its workflow is as follows:

[0110] Dual weighted branch: Input feature map The process proceeds to two branches. Branch 1: First, it undergoes layer normalization (…). ), and then with a learnable weight vector (Channel-by-channel multiplication) effect. Branch 2: Directly with another learnable weight vector. Function. The results of the two branches are added together, aiming to adaptively fuse the normalized and original feature information.

[0111] Downsampling and Multi-Scale Convolution: The summed features are passed through a downsampling convolutional layer (stride 2) to reduce the resolution and increase the number of channels. Subsequently, the feature map is copied four times and fed into four parallel convolutional layers with kernel sizes of 7x7, 5x5, 3x3, and 1x1, respectively, to extract multi-scale features within different receptive fields. , , , .

[0112] Adaptive weight fusion: Set four learnable scalar weights , , , ,pass Normalizing the function yields , , , The multi-scale feature fusion result is as follows: This design allows the network to dynamically adjust its dependence on features at different scales.

[0113] Upsampling and residual connection: go through After activation, the image is restored to its original state using an upsampling layer (such as transposed convolution or interpolation). Same spatial dimensions. The upsampled result is the same as the original. Element-wise addition is performed to form residual connections, and the final features are output. This structure facilitates gradient flow and feature reuse.

[0114] Hierarchical Feature Extraction: The model includes shallow, medium, and deep feature extraction modules. Through stacking different numbers of motion parsing layers and branching, fusion, and downsampling operations on feature maps, multi-level features from details to semantics are gradually extracted and fused.

[0115] Output Header: Finally, after a series of upsampling and multi-scale feature fusion, the feature map is compressed to 17 channels through a 1x1 convolutional layer, generating a heatmap of 17 key points.

[0116] A jump rope counting system for implementing a jump rope counting method, the system comprising:

[0117] The data acquisition module is used to acquire video data containing rope jumpers and extract image frame sequences;

[0118] The key point detection module includes a pre-trained jump rope counting detection model, which is used to receive the image frame sequence and output the predicted point information of human skeleton key points.

[0119] The jump rope detection module includes a pre-trained target detection model, which is used to receive the image frame sequence and output the position information of the jump rope.

[0120] The status judgment module is used to determine whether the jumper is in a preparatory jumping state based on the predicted point information;

[0121] The action recognition and counting module is used to identify the process of the jump rope changing from the first state to the second state in continuous image frames based on the position information of the jump rope and the predicted position information of the key points of the human skeleton, and to complete the recognition and counting of effective jump rope actions in combination with the judgment of the preparatory jump state.

[0122] A jump rope counting system, wherein the jump rope counting detection model is obtained by: collecting and labeling training data containing key points of the human body and the position of the jump rope; and using the labeled data to train a neural network structure containing a motion analysis module to obtain the jump rope counting detection model.

[0123] A jump rope counting device, comprising:

[0124] At least one processor;

[0125] The memory is communicatively connected to the at least one processor;

[0126] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform a jump rope counting method.

[0127] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a jump rope counting method.

[0128] For example, a jump rope counting system, which can be deployed on a server or edge computing device, includes software modules including:

[0129] Data acquisition module: responsible for controlling the camera, receiving video streams, frame extraction, and image preprocessing.

[0130] Key point detection module: Loads and runs the jump rope counting detection model, receives image frames, and returns the coordinates of key points.

[0131] Rope skipping detection module: Loads and runs the target detection model, receives image frames, and returns the rope skipping position.

[0132] State determination module: Implements the preparatory state determination algorithm.

[0133] Action recognition and counting module: Implements the logic for detecting jump rope state transitions and determining valid actions, and maintains the counter.

[0134] User interface and data management module: Provides start / end test control, result display, data storage and query functions.

[0135] A jump rope counting device can be a dedicated terminal integrating a camera and a computing unit, or a general-purpose computer or server installed with the aforementioned system software. The device includes at least one processor (CPU / GPU), memory (RAM / ROM / hard disk), a camera interface, and necessary input / output devices. The memory stores the operating system and the application program implementing the method of this invention.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for counting jump ropes, characterized in that, The method includes the following steps: S1. Acquire video data containing rope skippers and extract a continuous sequence of image frames from it; S2. Input the image frame sequence into the pre-trained rope skipping detection model to obtain the predicted point information of human skeletal key points in each image frame; the rope skipping detection model includes a motion analysis module for extracting multi-scale features of human key points; The motion analysis module includes: The first processing branch performs layer normalization on the input feature map and then multiplies it with the first learnable weight factor; The second processing branch multiplies the input feature map by a second learnable weight factor; A feature fusion layer is used to add the outputs of the first processing branch and the second processing branch, and then perform downsampling processing; A multi-scale feature extraction layer is used to perform convolution operations with multiple different kernel sizes on the output of the feature fusion layer in parallel to obtain feature maps at multiple scales. An adaptive weight fusion layer is used to assign learnable weights to the feature maps at multiple scales and perform weighted summation to obtain fused multi-scale features. The upsampling and residual connection layer is used to add the fused multi-scale features, after activation function and upsampling processing, to the input feature map, and output the final features of the motion parsing module; The jump rope counting detection model further includes a shallow feature extraction module, a middle feature extraction module, and a deep feature extraction module connected in sequence; wherein, the shallow feature extraction module contains two parallel branches, each branch containing multiple motion parsing layers in series; the middle feature extraction module and the deep feature extraction module each contain multiple feature extraction blocks in series, each feature extraction block having multiple parallel branches, each branch containing multiple motion parsing layers in series, and the outputs of each branch are summed; S3. Synchronously input the image frame sequence into the pre-trained target detection model to obtain the position information of the jump rope in each image frame; S4. Based on the predicted position information of the key points of the human skeleton, determine whether the jumper is in a preparatory jumping state. S5. In a continuous image frame sequence, detect the process of the jump rope position changing from a first state to a second state relative to the whole human body region; wherein, the first state is that the jump rope is located above the whole human body region, and the second state is that the jump rope passes under the whole human body region. S6. When a complete transition from the first state to the second state is detected, and the jumper was in the preparatory jump state before the transition, it is determined that a valid jump rope action has been completed, and the count is performed.

2. The method for counting jump ropes according to claim 1, characterized in that, In step S2, the jump rope counting detection model obtains the predicted point information in the following way: the model outputs a heat map of a preset number of key points, analyzes the peak coordinates of each heat map, and takes the local maximum value point as the predicted point of the corresponding human skeletal key point.

3. The method for counting jump ropes according to claim 2, characterized in that, In step S4, determining whether the jumper is in a preparatory jumping state specifically involves determining whether the following conditions are met simultaneously: the nose is higher than the left and right wrists; the left and right wrists are higher than the left and right hips; the left and right hips are higher than the left and right ankles; and the left and right wrists are closest to the hips on the same side.

4. The method for counting jump ropes according to claim 2, characterized in that, The jump rope counting detection model is trained through the following steps: S11. Collect images containing various human postures and rope skipping images to construct a training dataset; S12. Mark the key points of human skeleton on the human images in the training dataset, and mark the jump rope position information on the jump rope images; S13. Using the labeled dataset, train the jump rope counting detection model; wherein the jump rope counting detection model is trained based on a neural network structure that includes the motion parsing module.

5. A jump rope counting system, characterized in that, The system for implementing the method as described in any one of claims 1 to 4 comprises: The data acquisition module is used to acquire video data containing rope jumpers and extract image frame sequences; The key point detection module includes a pre-trained jump rope counting detection model, which is used to receive the image frame sequence and output the predicted point information of human skeleton key points. The jump rope detection module includes a pre-trained target detection model, which is used to receive the image frame sequence and output the position information of the jump rope. The status judgment module is used to determine whether the jumper is in a preparatory jumping state based on the predicted point information; The action recognition and counting module is used to identify the process of the jump rope changing from the first state to the second state in continuous image frames based on the position information of the jump rope and the predicted position information of the key points of the human skeleton, and to complete the recognition and counting of effective jump rope actions in combination with the judgment of the preparatory jump state.

6. A jump rope counting system according to claim 5, characterized in that, The jump rope counting detection model is obtained by: collecting and labeling training data containing key points of the human body and the position of the jump rope; using the labeled data to train a neural network structure containing a motion analysis module to obtain the jump rope counting detection model.

7. A jump rope counting device, characterized in that, include: At least one processor; The memory is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the jump rope counting method as described in any one of claims 1 to 4.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the jump rope counting method as described in any one of claims 1 to 4.

Images