hao/tennis-ai-research
1
0
派生 0
代码 工单 合并请求 工作流 软件包 项目 发布 百科 活动

feat: 全网网球视频教程、AI特征库与训练资料调研完整合集

浏览代码 
- 新增 docs/01_Video_Tutorials_Research.md: YouTube和Bilibili高播放量网球视频教程调研报告
- 新增 docs/02_GitHub_AI_Tennis_Research.md: GitHub网球项目、训练图片与AI特征库调研报告
- 新增 research_notes/: YouTube频道、Bilibili视频、AI综述论文原始调研数据
- 新增 github_tennis_projects.csv: 12个GitHub网球相关开源项目详细数据
- 新增 ai_tennis_research.csv: 8个AI网球研究方向详细数据
- 更新 README.md: 完整项目说明和核心发现摘要
这个提交包含在：
hao
2026-03-14 14:21:50 -04:00
父节点 43506ccd82
当前提交 f78e84cba8
共修改 8 个文件，包含 493 行新增和 2 行删除
下载 Patch 文件 下载 Diff 文件 展开所有文件 折叠所有文件

13
github_tennis_projects.csv 普通文件
查看文件

@@ -0,0 +1,13 @@
Subject,Project Name,Stars,Description,Tech Stack,Dataset Info,Main Features,Has Pretrained Model,Last Update,Error
https://github.com/HaydenFaulkner/Tennis,HaydenFaulkner/Tennis,121,A Tennis dataset and models for event detection & commentary generation.,"Keras, MXNet, Gluon, Python",The dataset consists of 5 matches with manually annotated temporal events and commentary captions. It includes 746 points with captions and an additional 10817 unaligned captions.,"Event detection, commentary generation, dense fine-grained event recognition, localisation and description.",True,2025-06,
https://github.com/yastrebksv/TennisCourtDetector,yastrebksv/TennisCourtDetector,239,一个用于从广播视频中检测网球场关键点的深度学习网络。该网络基于热图，可以检测14个网球场关键点，并集成了经典的计算机视觉方法以增强球网检测效果。,"PyTorch, TrackNet, OpenCV, NumPy, Matplotlib",包含8841张图片，分为训练集（75%）和验证集（25%），每张图片有14个标注点，分辨率为1280x720，覆盖硬地、红土和草地三种场地类型。,从广播视频中检测网球场关键点；基于热图的深度学习网络可检测14个关键点；通过经典的计算机视觉技术进行后处理以优化球网预测。,True,2023-08,
https://github.com/chow-vincent/tennis_action_recognition,chow-vincent/tennis_action_recognition,38,Using deep learning to perform action recognition in the sport of tennis.,"Keras, TensorFlow, Inception V3, LSTM, optical flow","1980 RGB videos sized 640 x 480. In each clip, a player performs one of 12 possible tennis strokes.","Classify videos of players performing tennis strokes (e.g. forehand, backhand, service).",True,2021-05,
https://github.com/andrenatal/AI_Tennis_Coach,andrenatal/AI_Tennis_Coach,5,"A deep learning-based computer vision system to analyze tennis matches in real time, classify strokes, and assess similarities with reference players using pose estimation. This extensible AI platform aims to improve performance, enhance technique, and prevent injuries for both professional and recreational players.","Movenet, OpenPose, Keras, Tensorflow.js, ONNX, scikit-learn, Numpy, Math.js, npyjs.js, absl-py, astunparse, certifi, charset-normalizer, coloredlogs, contourpy, cupy, cycler, fastrlock, flatbuffers, fonttools, gast, google-pasta, grpcio, h5py, humanfriendly, idna, imageio, joblib, kiwisolver, libclang, Markdown, markdown-it-py, MarkupSafe, matplotlib, mdurl, ml-dtypes, mpmath, namex, onnxruntime, opencv-python, opt-einsum, optree, packaging, Pillow, protobuf, py-cpuinfo, PyYAML, requests, rich, scipy, seaborn, six, tensorboard, tensorboard-data-server, tensorflow, tensorflow-estimator, tensorflow-io-gcs-filesystem, termcolor, typing-extensions, urllib3, Werkzeug, wrapt",Youtube footage of professional players and recreational players was extracted and annotated. A custom tool for annotating tennis data was built and is incorporated in this repo.,"Real-time analysis of tennis matches, stroke classification, and similarity comparison with reference players. It utilizes pose estimation from video footage and is extensible to other sports for performance improvement, technique enhancement, and injury prevention.",True,2024,
https://github.com/ArtLabss/tennis-tracking,ArtLabss/tennis-tracking,649,"Open-source Monocular Python HawkEye for Tennis. The main objectives are to track the ball, detect court lines, and detect the players in a tennis match.","TrackNet, ResNet50, YOLOv3, sktime, TensorFlow","The project provides sample input videos and CSV files (bigDF.csv, df.csv) containing data for training the models.","Track the ball, detect court lines, detect the players, provide a dynamic mini-map with player and ball positions, and predict bounce points.",True,2022,
https://github.com/ShadowMasterAJ/Tennis-Pose-Estimation-Detection-Classification,ShadowMasterAJ/Tennis-Pose-Estimation-Detection-Classification,0,"TennisPose: A Multi-Task Deep Learning Approach for Player Action Detection, Pose Estimation, and Classification in Tennis","YOLO, EfficientNet, Squeeze-and-Excitation, Multi-scale Fusion, LSTM, SpatioTemporal Attention",包含YOLO和原始数据集，分别带有图片和标注文件,该项目通过深度学习方法，实现了对网球运动员的动作检测、姿态估计和击球分类。主要功能包括运动员检测、关键点姿态估计以及基于姿态的动作分类。,True,2025-02,
https://github.com/antoinekeller/tennis_shot_recognition,antoinekeller/tennis_shot_recognition,42,该项目使用密集神经网络（Dense NN）和循环神经网络（RNN）识别网球击球动作。它利用Movenet进行人体姿态估计，并提供了从视频下载、标注、特征提取到模型训练和推理的完整流程。,"TensorFlow, Keras, GRU, Movenet",数据集通过YouTube视频创建，使用annotator.py进行标注，生成包含击球类型和帧ID的CSV文件。每个击球动作被视为持续30帧（约1秒）的姿态序列。,1. 使用Movenet进行实时人体姿态估计。 2. 提供视频标注工具，用于标记正手、反手、发球等动作。 3. 将网球击球动作提取为人体姿态序列特征。 4. 分别使用全连接网络和RNN（GRU）进行训练和分类。 5. 支持对视频进行逐帧或基于RNN的平滑分类和计数。,True,2022-03,
https://github.com/KalinLai-void/TennisPoseTrainer,KalinLai-void/TennisPoseTrainer,1,A tennis stroke training application is developed based on OpenPose.,"OpenPose, Python, opencv-python, Pillow, numpy, tk, tkvideoplayer",The repository does not contain a dataset. It uses a camera to capture user pose in real-time.,"This application is a sports tool based on OpenPose that allows users to train their tennis strokes (forehand and backhand). Users can select their dominant hand and see their swing status, pose, and trajectory after each swing.",True,2024-10,
https://github.com/jasnwag/tennis_serve_dataset,jasnwag/tennis_serve_dataset,2,一个全面的网球发球数据集，来自 2024 年美国网球公开赛，提供 3D 关键点跟踪、发球分析和性别分类数据。,"Python, Pandas, Numpy","包含 6,370 个网球发球数据，来自 2024 年美国网球公开赛。数据为 CSV 格式，内嵌 JSON 格式的 3D 关键点数组（每帧 17 个关节）。",提供详细的发球数据（速度、方向、结果）、17个关节的3D关键点跟踪、以及球员人口统计信息。可用于生物力学分析、机器学习（如性别分类、结果预测）和体育分析。,True,2025-09,
https://github.com/julieemmerson/opensim_tennis_model,julieemmerson/opensim_tennis_model,3,"This repository contains a full-body OpenSim model designed to be suitable for tennis, along with the geometry files. The model combines the scapulothoracic joint of Seth et al. (2016), the upper body of Cazzola et al. (2017) and the lower body of Rajogopal et al. (2016). The inclusion of the scapulothoracic joint allows for realistic shoulder movement, even during tennis serves. Segment mass and inertial properties are based on de Leva (1996). This model has no muscles and is currently only suitable for inverse kinematics analysis.",OpenSim,Contains geometry files for a full-body OpenSim model.,"A full-body OpenSim model for tennis, enabling realistic shoulder movement for serves and suitable for inverse kinematics analysis.",,2024-12,
https://github.com/JeffSackmann/tennis_atp,JeffSackmann/tennis_atp,1.5k,该项目提供了ATP网球排名、比赛结果和统计数据。这个仓库包含了主要的ATP球员文件、历史排名、比赛结果和比赛统计数据。,CSV,该数据集包含ATP球员信息、历史排名以及从1968年至今的详细比赛结果和统计数据，以CSV格式提供。它涵盖了巡回赛级别、挑战者赛和希望赛，以及双打比赛。,提供全面且历史悠久的ATP网球排名、比赛结果和球员统计数据集。它包括单打和双打比赛的数据，涵盖了从20世纪60年代至今的各种级别的锦标赛。,,2024,
https://github.com/abdullahtarek/tennis_analysis,abdullahtarek/tennis_analysis,807,该项目通过分析视频中的网球运动员来测量他们的速度、击球速度和击球次数。项目使用YOLO检测球员和网球，并利用CNN提取场地关键点。这是一个非常适合提升机器学习和计算机视觉技能的实践项目。,"YOLO, CNN, Pytorch, OpenCV, Pandas, Numpy",未提供具体的数据集信息，但项目包含用于训练网球检测器和场关键点提取器的Jupyter Notebook。,分析网球运动员的速度、击球速度和击球次数；使用YOLO进行球员和网球检测；使用CNN提取场地关键点。,True,2022-07,
1 Subject Project Name Stars Description Tech Stack Dataset Info Main Features Has Pretrained Model Last Update Error
2 https://github.com/HaydenFaulkner/Tennis HaydenFaulkner/Tennis 121 A Tennis dataset and models for event detection & commentary generation. Keras, MXNet, Gluon, Python The dataset consists of 5 matches with manually annotated temporal events and commentary captions. It includes 746 points with captions and an additional 10817 unaligned captions. Event detection, commentary generation, dense fine-grained event recognition, localisation and description. True 2025-06
3 https://github.com/yastrebksv/TennisCourtDetector yastrebksv/TennisCourtDetector 239 一个用于从广播视频中检测网球场关键点的深度学习网络。该网络基于热图，可以检测14个网球场关键点，并集成了经典的计算机视觉方法以增强球网检测效果。 PyTorch, TrackNet, OpenCV, NumPy, Matplotlib 包含8841张图片，分为训练集（75%）和验证集（25%），每张图片有14个标注点，分辨率为1280x720，覆盖硬地、红土和草地三种场地类型。 从广播视频中检测网球场关键点；基于热图的深度学习网络可检测14个关键点；通过经典的计算机视觉技术进行后处理以优化球网预测。 True 2023-08
4 https://github.com/chow-vincent/tennis_action_recognition chow-vincent/tennis_action_recognition 38 Using deep learning to perform action recognition in the sport of tennis. Keras, TensorFlow, Inception V3, LSTM, optical flow 1980 RGB videos sized 640 x 480. In each clip, a player performs one of 12 possible tennis strokes. Classify videos of players performing tennis strokes (e.g. forehand, backhand, service). True 2021-05
5 https://github.com/andrenatal/AI_Tennis_Coach andrenatal/AI_Tennis_Coach 5 A deep learning-based computer vision system to analyze tennis matches in real time, classify strokes, and assess similarities with reference players using pose estimation. This extensible AI platform aims to improve performance, enhance technique, and prevent injuries for both professional and recreational players. Movenet, OpenPose, Keras, Tensorflow.js, ONNX, scikit-learn, Numpy, Math.js, npyjs.js, absl-py, astunparse, certifi, charset-normalizer, coloredlogs, contourpy, cupy, cycler, fastrlock, flatbuffers, fonttools, gast, google-pasta, grpcio, h5py, humanfriendly, idna, imageio, joblib, kiwisolver, libclang, Markdown, markdown-it-py, MarkupSafe, matplotlib, mdurl, ml-dtypes, mpmath, namex, onnxruntime, opencv-python, opt-einsum, optree, packaging, Pillow, protobuf, py-cpuinfo, PyYAML, requests, rich, scipy, seaborn, six, tensorboard, tensorboard-data-server, tensorflow, tensorflow-estimator, tensorflow-io-gcs-filesystem, termcolor, typing-extensions, urllib3, Werkzeug, wrapt Youtube footage of professional players and recreational players was extracted and annotated. A custom tool for annotating tennis data was built and is incorporated in this repo. Real-time analysis of tennis matches, stroke classification, and similarity comparison with reference players. It utilizes pose estimation from video footage and is extensible to other sports for performance improvement, technique enhancement, and injury prevention. True 2024
6 https://github.com/ArtLabss/tennis-tracking ArtLabss/tennis-tracking 649 Open-source Monocular Python HawkEye for Tennis. The main objectives are to track the ball, detect court lines, and detect the players in a tennis match. TrackNet, ResNet50, YOLOv3, sktime, TensorFlow The project provides sample input videos and CSV files (bigDF.csv, df.csv) containing data for training the models. Track the ball, detect court lines, detect the players, provide a dynamic mini-map with player and ball positions, and predict bounce points. True 2022
7 https://github.com/ShadowMasterAJ/Tennis-Pose-Estimation-Detection-Classification ShadowMasterAJ/Tennis-Pose-Estimation-Detection-Classification 0 TennisPose: A Multi-Task Deep Learning Approach for Player Action Detection, Pose Estimation, and Classification in Tennis YOLO, EfficientNet, Squeeze-and-Excitation, Multi-scale Fusion, LSTM, SpatioTemporal Attention 包含YOLO和原始数据集，分别带有图片和标注文件 该项目通过深度学习方法，实现了对网球运动员的动作检测、姿态估计和击球分类。主要功能包括运动员检测、关键点姿态估计以及基于姿态的动作分类。 True 2025-02
8 https://github.com/antoinekeller/tennis_shot_recognition antoinekeller/tennis_shot_recognition 42 该项目使用密集神经网络（Dense NN）和循环神经网络（RNN）识别网球击球动作。它利用Movenet进行人体姿态估计，并提供了从视频下载、标注、特征提取到模型训练和推理的完整流程。 TensorFlow, Keras, GRU, Movenet 数据集通过YouTube视频创建，使用annotator.py进行标注，生成包含击球类型和帧ID的CSV文件。每个击球动作被视为持续30帧（约1秒）的姿态序列。 1. 使用Movenet进行实时人体姿态估计。 2. 提供视频标注工具，用于标记正手、反手、发球等动作。 3. 将网球击球动作提取为人体姿态序列特征。 4. 分别使用全连接网络和RNN（GRU）进行训练和分类。 5. 支持对视频进行逐帧或基于RNN的平滑分类和计数。 True 2022-03
9 https://github.com/KalinLai-void/TennisPoseTrainer KalinLai-void/TennisPoseTrainer 1 A tennis stroke training application is developed based on OpenPose. OpenPose, Python, opencv-python, Pillow, numpy, tk, tkvideoplayer The repository does not contain a dataset. It uses a camera to capture user pose in real-time. This application is a sports tool based on OpenPose that allows users to train their tennis strokes (forehand and backhand). Users can select their dominant hand and see their swing status, pose, and trajectory after each swing. True 2024-10
10 https://github.com/jasnwag/tennis_serve_dataset jasnwag/tennis_serve_dataset 2 一个全面的网球发球数据集，来自 2024 年美国网球公开赛，提供 3D 关键点跟踪、发球分析和性别分类数据。 Python, Pandas, Numpy 包含 6,370 个网球发球数据，来自 2024 年美国网球公开赛。数据为 CSV 格式，内嵌 JSON 格式的 3D 关键点数组（每帧 17 个关节）。 提供详细的发球数据（速度、方向、结果）、17个关节的3D关键点跟踪、以及球员人口统计信息。可用于生物力学分析、机器学习（如性别分类、结果预测）和体育分析。 True 2025-09
11 https://github.com/julieemmerson/opensim_tennis_model julieemmerson/opensim_tennis_model 3 This repository contains a full-body OpenSim model designed to be suitable for tennis, along with the geometry files. The model combines the scapulothoracic joint of Seth et al. (2016), the upper body of Cazzola et al. (2017) and the lower body of Rajogopal et al. (2016). The inclusion of the scapulothoracic joint allows for realistic shoulder movement, even during tennis serves. Segment mass and inertial properties are based on de Leva (1996). This model has no muscles and is currently only suitable for inverse kinematics analysis. OpenSim Contains geometry files for a full-body OpenSim model. A full-body OpenSim model for tennis, enabling realistic shoulder movement for serves and suitable for inverse kinematics analysis. 2024-12
12 https://github.com/JeffSackmann/tennis_atp JeffSackmann/tennis_atp 1.5k 该项目提供了ATP网球排名、比赛结果和统计数据。这个仓库包含了主要的ATP球员文件、历史排名、比赛结果和比赛统计数据。 CSV 该数据集包含ATP球员信息、历史排名以及从1968年至今的详细比赛结果和统计数据，以CSV格式提供。它涵盖了巡回赛级别、挑战者赛和希望赛，以及双打比赛。 提供全面且历史悠久的ATP网球排名、比赛结果和球员统计数据集。它包括单打和双打比赛的数据，涵盖了从20世纪60年代至今的各种级别的锦标赛。 2024
13 https://github.com/abdullahtarek/tennis_analysis abdullahtarek/tennis_analysis 807 该项目通过分析视频中的网球运动员来测量他们的速度、击球速度和击球次数。项目使用YOLO检测球员和网球，并利用CNN提取场地关键点。这是一个非常适合提升机器学习和计算机视觉技能的实践项目。 YOLO, CNN, Pytorch, OpenCV, Pandas, Numpy 未提供具体的数据集信息，但项目包含用于训练网球检测器和场关键点提取器的Jupyter Notebook。 分析网球运动员的速度、击球速度和击球次数；使用YOLO进行球员和网球检测；使用CNN提取场地关键点。 True 2022-07