icSPORTS 2021 Abstracts

Area 1 - Computer Systems in Sports

Full Papers

Paper Nr:	6
Title:	Analysis of Deep Learning Action Recognition for Basketball Shot Type Identification
Authors:	Carlos Olea, Gus Omer, John Carter and Jules White
Abstract:	Recent technologies have been developed to track basketball shooting and provide detailed data on shooting accuracy for players. Further segmenting this shot accuracy data based on shot types allows a more detailed analysis of player performance. Currently this segmentation must be performed manually. In this paper, we apply a state-of-the-art action recognition model to the problem of automated shot type classification from videos. The paper presents experiments performed to optimize shot type recognition, a unique taxonomy for the labeling of shot types, and discusses key results on the task of categorizing three different shot types. Additionally, we outline key challenges we uncovered applying current deep learning techniques to the task of shot type classification. The NOAH system enables the capture of basketball shooting data by recording every shot taken on a court along with its shooter and critical statistics. We utilized videos of 50,000 practice shots from various players captured through NOAH system to perform the task of classifying shot type. These three second video clips contain the shooting action along with movements immediately preceding and following the shot. On the problem of shot type classification, the Temporal Relational Network achieved an accuracy of 96.8% on 1500 novel shots.
Download

Paper Nr:	15
Title:	Data-driven Support of Coaches in Professional Cycling using Race Performance Prediction
Authors:	Aleksei Karetnikov, Wim Nuijten and Marwan Hassani
Abstract:	In individual sports, the judgment of which training activity will lead to the best performance is mostly based on the expert knowledge of the coach. Recent advances in data collection and data science have opened up new possibilities for performing a data-driven analysis to support the coach in improving the training programs of the athletes. In this paper, we investigate several methods to do such analysis for professional cyclists. We provide the coach with a framework to predict the Maximum Mean Powers (MMPs) of a cyclist in an upcoming race based on the recently performed training sessions. This way the coach can experiment with several planned alternatives to figure out the best way for preparing the athlete for a race. We conduct multiple prediction models through an extensive analysis of a real dataset collected recently about the performance of professional riders with varying physiologies and temporal performance peaks. We show that the application of the hybrid model using XGBoost and CatBoost has clear advantages. Additionally, we show that the accuracy of our general model can be further increased by filtering according to the mountain stages. We have additionally validated the proposed framework using an openly available real dataset and the results were consistent with those of the collected data. We offer an open source implementation of our proposed framework.
Download

Paper Nr:	18
Title:	Passing to Win: Using Characteristics of Passing Information for Match Winner Prediction
Authors:	Taihu Li, Jeewoo Yoon, Daejin Choi and Jinyoung Han
Abstract:	Predictingthe football match results has received great attention both in sports industry and academic fields. Many researchers have studied on predicting the match outcome using the simple features such as the number of shots and passes. However, little attention has been paid to using pass interaction features, which can represent how players in a match interact to each other. To this end, we propose a win-lose prediction model that predicts a match result using the pass interaction and other features, achieving high accuracy of 79.5%. By conducting an ablation study, we find that the proposed interaction features play an important role in accurately predicting match results. We believe our work can provide important insights both for industry and academic researchers who want to understand the characteristics of winning teams.
Download

Short Papers

Paper Nr:	4
Title:	A Study of Machine Learning Models for Personalized Heart Rate Forecasting in Mountain Biking
Authors:	Xiaoxing Qiu, Jules White and Douglas C. Schmidt
Abstract:	Heart rate forecasting in cycling is most effective when it is personalized and course-specific to account for the influence of individual and terrain factors. This paper empirically assesses various personalized and course-specific heart rate forecasting models based on four machine learning models, including random forest, feed forward neural networks (FFNNs), recurrent neural networks (RNNs), and long short term memory (LSTM). The mean square error (MSE) is selected as the metric for model comparison. The results of our experiments show that despite the severely overfitted random forest models the LSTM models have the lowest MSE in the heart rate forecasting on our test dataset.
Download

Paper Nr:	9
Title:	Starting Position Analysis of Winners in Short Track Speed Skating Competitions during 2007-2019
Authors:	Lixin Sun, Kuan Tao, Tianxiao Guo and Fei Liu
Abstract:	Short track speed skating is a racing sport on ice, in which contestants compete against each other instead of the clock. Thus compared to the speed and technical skating ability, equally, if not more important is race tactics. This study analysed in-depth the starting positions of winners in 121 competitions during 2007-2019, where 173 female and 247 male champions were announced correspondingly from 4313 female and 5212 male individual races (preliminaries, heats, quarter-finals, semi-finals, and final in 500m, 1000m, 1500m, and 3000m classifications), to explore a pattern of effective tactical positioning strategy. The Kendall’s tau-b (rt) correlation between starting and finishing position decreases with race distance, which was highest and positive for all 500m races (0.347, P<0.05), which verified previous studies. Furthermore, starting position distributions of winners in each round and starting positions variations of champions along the rounds were analysed. Results show that skaters in the first track were inclined to win the rounds in 500m, 1000m and 1500m (28%, 28% and 22%, respectively) and the differences between starting and finishing positions for champions were minimized in semi-finals, indicating skaters should spare no effort in semi-final to achieve success.
Download

Paper Nr:	17
Title:	POOF: Efficient Goalie Pose Annotation using Optical Flow
Authors:	Brennan Gebotys, Alexander Wong and David Clausi
Abstract:	Due to the wide range of applications for human pose estimation including sports analytics and more, research has optimized pose estimation models to achieve high accuracies when trained on large human pose datasets. However, applying these learned models to datasets that are from a different domain (which is usually the goal for many real-world applications) usually leads to a large decrease in accuracy which is not acceptable. To achieve acceptable results, a large number of annotations is still required which can be very expensive. In this research, we leverage the fact that many pose estimation datasets are derived from individual frames of a video and use this information to develop and implement an efficient pose annotation method. Our method uses the temporal motion between frames of a video to propagate ground truth keypoints across neighbouring frames to generate more annotations to provide efficient POse annotation using Optical Flow (POOF). We find POOF achieves the best performance when used in different domains than the pretrained domain. We show that in the case of a real-world hockey dataset, using POOF can achieve 75% accuracy (a +15% improvement, compared to using COCO-pretrained weights) with a very small number of ground truth annotations.
Download

Paper Nr:	23
Title:	Testing Environment for Developing a Wireless Networking System based on Image-assisted Routing for Sports Applications
Authors:	Shiho Hanashiro, JunFeng Xue, Junya Morioka, Ryusuke Miyamoto, Takuam Hamagami, Kentaro Yanagihara, Yasutaka Kawamoto, Hiroyuki Okuhata, Hiroyuki Yomo and Tomohito Takubo
Abstract:	To improve the effectiveness of exercise a novel vital sensing system is under development. For real-time sensing of vital signs by a wireless network during exercise, image-assisted routing that enables dynamic routing of a multi-hop network according to sensor locations estimated by visual information. To develop the novel networking system, testing environment that enables runtime verification of dynamic routing based on image processing. Experimental results actual sensor nodes with AR markers showed that locations of sensor nodes obtained using a USB camera could be appropriately given to the control software of base station to manage routing information.
Download

Paper Nr:	24
Title:	“How Was the Match?”: Semantic Similarity between Electronic Media Commentary and Work Domain Analysis Key Phrases
Authors:	Gustavo Silva, Ricardo Ribeiro and Rui J. Lopes
Abstract:	Football player’s performance can be measured in an objective way (e. g. Goals scored, assists, interceptions), this being seldom a method to compare and rank the best players by categories. Over years of study, many other factors that can influence the players performance were discovered and studied, considering not only objective factors, but also subjective factors. Match commentary from different sources (e.g., social and formal media) also plays an important role on a more subjective performance assessment. By using semantic similarity analysis, this study aims to contribute to the understanding of the concepts that are used in this commentaries, notably to each extend key phrases associated to match processes are used in commentaries published in social and formal media.
Download

Paper Nr:	20
Title:	Modern Light Sport Training Systems: Critical Analysis of Their Construction and Performance Features
Authors:	Anton Ezhov, Anna Zakharova and Dmitriy Kachalov
Abstract:	Modern digital technologies help to provide the athlete progress in the majority of sport kinds. There are too many sport gadgets intended for different training aims. So, assessment of their advantages or disadvantages is difficult for a coach or other customers. The aim of our research was to provide a detailed comparison of the light sport training systems available with the identification of the operation features. Methods. FitLightTM, BlazePod and XLiGHT sport training systems were considered in terms of usability, features, performance and diagnostic possibilities. It has been found that FitLightTM is more suitable for sport diagnostics. It should be mentioned however that BlazePod and XLiGHT simulator are more affordable and can be high in demand by coaches.
Download

Paper Nr:	36
Title:	Machine Learning Applications to Sports Injury: A Review
Authors:	Hanna Sigurdson and Jonathan H. Chan
Abstract:	As sports injuries increase in frequency in adolescents, and injuries in professional athletes create a detrimental impact on the sports industry, research surrounding preventing sports injuries becomes more prevalent. The mechanism for sports injury is well defined and includes intrinsic (age, psychology etc.) and extrinsic risk factors (weather, training load etc.), and the inciting event. With the rise of machine learning (ML), a variety of ML techniques have been applied to various sports injury aspects. The purpose of this work is to assess the current applications of ML to sports injury and identify areas of growth by a systematic analysis of applications to each injury element: intrinsic factors, extrinsic factors, and the inciting event. Current underdeveloped areas are identified as: psychological effect, use of extrinsic factors, analysis of the inciting event, and application of the action recognition ability of videos and wearable technology. Future technical applications in these underdeveloped areas should be undergone to expand on and improve sports injury prevention technology.
Download

Area 2 - Health and Support Technology

Full Papers

Paper Nr:	21
Title:	Investigation of UX and Flow Experience in Sports Activities during the Covid-19 Pandemic: A Comparative Analysis of Cycling Apps
Authors:	Klemens Weigl, Sabrina Schuster and Andreas Riener
Abstract:	Since the onset of the Covid-19 pandemic, a dramatic increase in mHealth application (app) downloads has been documented. However, overall dwell retention for fitness apps is low, so gamification techniques are used within apps with the goal of positively influencing the user experience and ultimately the user’s motivation. The so-called flow, which is related to intrinsic motivation, has been little explored in the context of cycling apps. Therefore, we conducted a quasi-experimental cycling study with 34 cyclists (20 female, 14 male; 19 to 57 years old) who tested the adidas Running by Runtastic (Adidas Runtastic), Komoot, and Strava cycling apps during a 20-minute bike ride. After testing each cycling app, they completed the User Experience Questionnaire (UEQ) and the Flow State Scale-2 Short (FSS-2S). Our results showed no significant differences across the six factors of the UEQ, nor across the total score of the FSS-2S. Thus, we conclude that the three cycling apps Adidas Runtastic, Komoot, and Strava are perceived and rated almost equally by female and male cyclists.
Download

Short Papers

Paper Nr:	5
Title:	Criterion Validation of an Open-source Wearable Physiological Sensors Device
Authors:	Antoine Langevin, William Bégin, Martin Lavallière, Louis-David Beaulieu, Bob-Antoine J. Menelas, Sébastien Gaboury, Kevin Bouchard, Ghyslain Gagnon and Linda Paquette
Abstract:	Wearable sensors are very popular in monitoring sport performances and increasingly used in scientific research. However, several scientific and ethical issues regarding pricing, raw data accessibility, validity and commercial access to user’s data are linked with these devices. To address these limitations, an open-source device, called Emotibit, was designed through crowdfunding. The aim of this study is to evaluate the criterion validity of this new open-source device’s physiological components in resting position. To this end, heart rate (HR) and heart rate variability (HRV) via photoplethysmography (PPG) and electrodermal activity (EDA) were assessed and compared with a medical grade reference device, the FlexComp Infiniti. The Bland-Altman plot and ratio (BAr) results indicate a good validity for HR estimation with a BAr of 0.02. However, results suggest an insufficient validity for HRV, as well as EDA amplitude and number of activation events estimation. These results are comparable to other studies using PPG for HRV estimation, but the EDA components need adjustment in regard to the sensitivity of the device. We analyze the validity issues associated with open source technology, and conclude that further improvements are required to qualify its accuracy with statistical significance. This study also contributes to the wearable sensors studies by identifying and describing the many challenges associated with the democratization of access to biosensing technology.
Download

Area 3 - Signal Processing in Human Movement

Full Papers

Paper Nr:	7
Title:	Measuring Pressure in Different Layers of the Ski Boot to Estimate Skiing Movements
Authors:	Aljoscha Hermann, Patrick Carqueville, Melanie Baldinger and Veit Senner
Abstract:	A pressure sensing measurement ski boot or sock would allow to estimate body positions, skiing manoeuvres, and external loads on the foot. This information may be used for research, in consumer products or for intelligent safety systems like a mechatronic ski binding. To investigate the optimal placement of pressure sensors with respect to the foot and the number of sensors needed to detect six pre-defined loading conditions, three pressure sensor systems were developed measuring the pressure in three respective layers: between foot and sock, sock and liner, liner and shell. The prototypes were evaluated in a laboratory test. The participant performed a series of six simulated ski manoeuvres each held for 5 seconds. In this pilot test the system sock / liner shows the best overall performance due to pressure curves in the mid-range of the sensor characteristics. Though, with an optimized sensor design a measurement boot with sensors between inner boot and shell may be possible, which would increase the robustness of the system needed for a future customer product. As a result of this study, a recommendation for sensor positions for the determination of the loading conditions in alpine skiing is given.
Download

Area 4 - Sport Performance and Support Technology

Full Papers

Paper Nr:	11
Title:	A Machine Learning Model to Predict Player’s Positions based on Performance
Authors:	Zixue Zeng and Bingyu Pan
Abstract:	The prediction of the player's positions, or determining which position a player is suitable for based on sports performance and physiological indicators, plays a major role in association football. This research is based on the public dataset provided by Wyscout, from which player-related indicators are extracted and processed. Six indicators, including the accuracy of shot, the accuracy of simple pass, the accuracy of glb (Ground loose ball), the accuracy of defending duel,the accuracy of air duel, the accuracy of attacking duel, are selected according to the ANOVA (analysis of variance) test, and being imported into BP neural network for training. Since the neural network has three hyperparameters: training rate, iterations, and the number of neurons in the hidden layer, it is required to use the k-fold cross-validation to evaluate by which hyperparameter pair the model predict best. It is found that when the learning rate is set to 0.0125 and the hidden layer neuron is set to 6, the average accuracy of the cross-check is the highest, which is 73%. When iterations reach 300, the accuracy curve tends to converge. The final accuracy rate can reach 77%.
Download

Paper Nr:	34
Title:	Evaluating Player’s Passing Ability based on Passing Network
Authors:	Zhang Zhipei and Pan Bingyu
Abstract:	As one of the most basic techniques of modern football, passing not only involves coordination within the team and different passing routes but is also closely related to the opponent team and the situation of the game. How to evaluate players’ ability to pass in such a complex environment has been a popular issue in the field of sports performance. Based on social network analysis, this paper constructs passing network and calculates ten indicators such as degree centrality. Then, using classification algorithms –The Gradient Boosting Decision Tree where these ten indicators serve as features to train a model to predict the average goal of each player. After adjusting parameters, the accuracy of the model reaches 0.628 while the value of AUC is 0.709, which shows the prediction of the model is relatively high, and social network analysis is an effective way to evaluate players’ passing ability to some extent.
Download

Short Papers

Paper Nr:	3
Title:	Sport Sparks: Supporting Creative Thinking by Professional Coaches
Authors:	James Lockerbie, Neil Maiden, Alex Wolf and Konstantinos Zachos
Abstract:	This short paper reports a new digital tool that supports creative idea generation about possible solutions to these challenges. Exploratory design research resulted in a new digital tool that was designed for use by the coaches of elite athletes, to discover creative ideas with which to remove, overcome and mitigate the effects of concrete athlete under-performance. Furthermore, initial feedback from 22 professional sports practitioners revealed that use of the tool led to most of them understanding the challenge from new perspectives, exploring alternative options to solve the challenge, and influenced their decision-making about the challenge.
Download

Paper Nr:	16
Title:	Estimated Effect of Monofin Stiffness on Sports Performance in Marathon Swimmers
Authors:	Alexander Bolotin, Vladislav Bakayev and Pavel Dudchenko
Abstract:	The study conducted has confirmed that the techniques and methods for training marathon swimmers in fins, tailoring the monofin stiffness to the individual capabilities of the athletes, play a major role in increasing the efficiency of training for competitions. The athletes can reliably achieve better results in the training process and in competitions. Training sessions with monofin stiffness taken into account have been found to have a higher efficiency. Our studies established that the kinematic characteristics undergo certain changes, some of them rather considerable. Using the monofin with the stiffness corresponding to individual capabilities of marathon swimmers allowed increasing the intracyclic speed of the strongest athletes from 2.53 m/s to 3.01 m/s (at the end of the experiment), i.e., the speed increase amounted to 19%. This is also confirmed by the increased average speed of distance swimming: 3.02 m/s and 3.48 m/s respectively, which is an increase of 17%. The study showed that successfully training athletes for competitions largely depends on accounting for the individual characteristics of muscular activity in marathon swimmers and the stiffness of the monofin. We have discovered that higher elasticity of an athlete's muscles should correspond to smaller monofin stiffness. The article presents the results of assessing the effect of monofin stiffness on the performance of marathon swimmers.
Download

Paper Nr:	22
Title:	Visualization of Human Motion via Virtual Reality Interface and Interaction based on It
Authors:	Akihiro Sakurai, Yosuke Ikegami, Milutin Nikolić, Yoshihiko Nakamura and Ko Yamamoto
Abstract:	This paper presents the concept of ecological sports training and the initial developments based on biomechanics analysis, VR technologies, and visualization of interaction. Computation algorithms in robotics have been applied to biomechanics analyses including muscle force estimation. These technologies are widely used for the self-biomechanics training, in which an athlete accesses and evaluates the analyzed results on his/her own motions and will necessarily move toward the ecological training that considers interactions with the counterpart and the environment in a sport game. We develop a VR visualization system of musculoskeletal analysis that provides a realistic experience of the interactions to an athlete. We also report an initial evaluation of the interactions with the virtual counterpart in the virtual environments through the VR system.
Download

Paper Nr:	35
Title:	Feasibility Study of an Image-based Supporting System for Sprint Training
Authors:	Shiho Hanashiro, Motoki Takematsu and Ryusuke Miyamoto
Abstract:	This study focuses on developing a novel system to improve the performance of short-distance races, where stride length, stride frequency, and maximum velocity are important factors. To estimate stride length and stride frequency, color-based image processing is adopted to extract the feet of a runner, based on cosine similarity in the RGB color space. The experimental results indicate that the stride length and stride frequency could be estimated with negligible errors. To estimate the running velocity; visual object detection, and pose estimation based on state-of-the-art deep learning schemes were applied: RetinaNet for visual object detection, and OpenPose for pose estimation. The experimental results using the real image dataset, indicated that the estimation error of the velocity by the proposed scheme was quite negligible.
Download