Table A4.

Overall view of remarkable papers on the topic of HPE in SPE and the used data.

Paper	Topic	Dataset/Data Source
Estimation of Gait Parameters from 3D Pose for Elderly Care [20]	Analysis of gait parameters (i.e., cadence, step length and step duration) of elderly people using HPE.	-RGB images + depth -Output: 3D -Own not publicly available data of gait using Kinect.
Discriminative hierarchical part-based models for human parsing and action recognition [22]	Human body parsing and action recognition.	-RGB images -Output: 2D UIUC (University of Illinois Urbana-Champaign) [71], annotated by hand, and a sports image dataset collected from the Internet in [72] (the annotation process is not specified). Both are publicly available in https://vision.cs.uiuc.edu/humanparse/ (last date accessed: 6 September 2021)
Athlete pose estimation by non-sequential key-frame propagation [12]	HPE from uncalibrated unconstrained monocular TV sports footage.	-RGB images -Output: 2D -Three sequences from the publicly available dataset HumanEva-I (ground truth obtained using a MoCap system). -Five TV quality sports sequences with different camera angles, zoom, and motion, which are not publicly available (own data). Annotated by hand, the occluded parts are not included in the error calculation as they are prone to human error.
HPE of Diver Based on Improved Stacked Hourglass Model [24]	HPE of divers.	-RGB images -Output: 2D Publicly available datasets MPII and LSP.
Pose Estimation of Complex Human Motion [25]	HPE of “complex human motion”, including a lot of sports activities (not managing properly occlusions and character inversion)	-RGB images -Output: 2D Publicly available COCO dataset.
AI Coach: Deep HPE and Analysis for Personalized Athletic Training Assistance [27]	Development of an AI Coach using HPE to analyze the pose of the athlete and detect “bad” poses, focused on Freestyle Skiing (athlete detection and tracking, HPE, bad pose detection).	-RGB images -Output: 2D (+ “correctness” of the pose) -Tracking tested onrgf publicly available VOT2018-LT and sports video dataset from LaSOT. -HPE tested on publicly available Penn Action and sub-JHMDB (manual annotation using Amazon Mechanical Turk).
Real-time dance evaluation by markerless human pose estimation [30]	A framework that evaluates dance performance by markerless HPE, with a special focus on correct detection in full-body rotation and self-occlusion situations.	-RGB images + depth -Output: 3D Publicly available datasets: EVAL (recorded using Kinect) for accuracy, and SMMC-10 (ground truth from PhaseSpace MoCap system) for error. Own publicly available K-Pop (true positions labeled using a marker-based MoCap system) (https://goo.gl/NoVDm4 link provided but not working at the last accessed date: 6 September 2021).
Human Pose Estimation-Based Real-Time Gait Analysis Using Convolutional Neural Network [13]	Approach that uses HPE to detect abnormalities in gait patterns with 5 possible outputs: normal, abnormal left toe, abnormal left foot, abnormal right toe, abnormal right foot.	-RGB images -Output: 2D (+ gait output category) Own not public dataset of RGB images of people walking in different situations using markers for the hip, knee, and ankle (no HPE data is specified as ground truth, the walking category is labeled by hand)
Can Markerless Pose Estimation Algorithms Estimate 3D Mass Centre Positions and Velocities during Linear Sprinting Activities? [17]	Test the capacity of estimating the 3D mass center positions and velocities during linear sprinting activities using 3D HPE. (in such actions in which skeleton is pushing, current HPE methods show quite high error for the objective of the paper, at least for the proposed method)	-RGB images + depth -Output: 3D Own not public dataset created using maker-based MoCap system Qualysis and markerless OpenPose system to record sprints.
Human Posture Recognition and Estimation Method Based on 3D Multiview Basketball Sports Dataset [14]	3D HPE using multiview basketball sports dataset.	-RGB images + depth -Output: 3D -ModelNet40 (CAD models with category label) -Own basketball dataset which is not publicly available (the annotation process is not indicated)
A Mobile Application for Running Form Analysis Based On Pose Estimation Technique [15]	2D HPE applied for running form analysis using a phone.	-RGB images -Output: 2D (+running performance data) Own not public dataset created using a motion capture system by Vicon Motion Systems.
HyperStackNet: A Hyper Stacked Hourglass Deep Convolutional Neural Network Architecture for Joint Player and Stick Pose Estimation in Hockey [35]	HPE in combination with stick estimation applied to hockey players.	-RGB images -Output: 2D -First half of the network was trained with the public dataset MPII. -The whole training and testing have been performed using the dataset HARPE (Hockey Action Recognition Pose Estimation) (from the source paper [60] it is interpreted that manual annotation has been used, but it is not expressed explicitly) from another paper, which at this moment is not publicly available.
Kinematic Pose Rectification for Performance Analysis and Retrieval in Sports [37]	HPE of athletes using the example of swimming, with images from a single camera which records inside and out the water at the same time (additionally, implements its own method of improving the estimation by inserting the swimming style by hand).	-RGB images -Output: 2D -Pretrained with publicly available dataset LSP. -- Tested on own dataset not publicly available of swimming videos using one camera that records the athlete inside and outside the water at the same time, annotated by a human expert.
Estimation of Center of Mass for Sports Scene Using Weighted Visual Hull [18]	Estimation of the CoM in sports using 3D HPE information as input.	-RGB images (+ output of HPE using a method from other paper) -Output: 3D position of the CoM Own not public data using 5 GoPro cameras (used to reconstruct the 3D position, no pose data is stored) and a force plate, being this last one element the one that gives the position of the CoM to compare with the result of the system.
Development of a markerless optical motion capture system for daily use of training in swimming [38]	Estimation of the pose and rotation and velocity of joints of swimmers, and fluid force simulation.	-RGB images -Output: 2D Own not public data recorded using a single static camera underwater recording the swimmer performing butterfly stroke. The segments of the body are annotated manually.
Athlete pose estimation by a global-local network [39]	HPE of athletes using a global-local approach.	-RGB images -Output: 2D Publicly available datasets: LSP for quantitative and qualitative HPE evaluation and UCF for qualitative evaluation, as this last dataset is used for sports action recognition, so, it does not include any joint position annotation.
Human Body Parts Estimation and Detection for Physical Sports Movements [41]	HPE for physical sports movements	-RGB images -Output: 2D Publicly available KTH Multiview Football and UCF Sports Actions (it is not an HPE dataset, but it is interpreted from the paper that the joints have been annotated manually for testing) datasets.
Robust Estimation of Flight Parameters for SKI Jumpers [42]	HPE and flight parameter estimation for ski jumpers during the flight phase.	-RGB images -Output: 2D Own not public dataset of images of different skiers in different conditions performing jumps, with joint and ski annotations. From the paper, it is interpreted that the annotation process has been manual, but it is not explicitly expressed.
Synthetic Image Translation for Football Players Pose Estimation [43]	HPE applied to football using cameras placed far from the field.	-RGB images -Output: 2D Publicly available COCO for training and comparison of results, and own not publicly available dataset created using four high-view and high-class wide-view cameras located far from the field.
FuturePose—Mixed Reality Martial Arts Training using Real-time 3D Human Pose Forecasting with an RGB Camera [46]	HPE applied to martial arts using a single 720p camera and combined with a pose forecasting method and VR technology.	-RGB: images -Output: 3D Publicly available MPI-INF-3D and Human3.6M for pre-training and validation. Own not publicly available dataset of martial arts practitioners and professionals doing boxing and kicking actions gathered from the Internet.