KUMITRON: A Multimodal Psychomotor Intelligent Learning System to Provide Personalized Support when Training Karate Combats

Abstract

Martial arts discipline can contribute to build multimodal psychomotor intelligent learning systems due to the nature of the movements involved to learn the corresponding motor skills: they are predefined and governed by the laws of physics. In this context, we are developing KUMITRON, an artificial intelligence system to support Karate practice. In particular, KUMITRON monitors the physical activity of the karatekas during a kumite (Karate combat) in real time using diverse sensors (physiological and inertial) as well as a drone camera in order to provide multisensorial personalized feedback computed with machine learning and computer vision algorithms. This multimodal monitoring aims to anticipate the movements that a fighter is going to perform on the mat, providing added value for training and offering personalized advice on the type of strategy to follow in order to win the kumite.

Keywords

• Wearable
• IoT
• kumite
• drone
• modelling human behavior
• OpenCV
• karate
• Artificial Intelligence
• Smart Dojo

-

Bloom's taxonomy defines a series of educational purposes from the behaviorist perspective [ 1 ] identifying three domains: cognitive, affective and psychomotor. Recent research has shown that classical computer-based support systems for learning can be extended with sensors to provide formative assessment in those three domains, paying special attention to their use as feedback tools [ 2 ]. In our current research we focus on psychomotor learning which deals with physical movement, coordination and the use of the motor-skill areas [ 3 ]. In particular, psychomotor learning starts from the low level of motor ability (recognition, involuntary actions, imitation, etc.), up to the high level (internalized performance and automated, sophisticated choreography, creation of new patterns or art, etc.) [ 4 ]. The acquisition of motor skills could benefit from individualized personalized instruction and support [ 5 ], just as learning from cognitive skills.

Martial arts discipline can contribute to build multimodal psychomotor intelligent learning systems due to the nature of the movements involved to learn the corresponding motor skills: they are predefined and governed by the laws of physics [ 6 ]. The teaching methods that exist today in martial arts are based on attending a class with a teacher (or sensei in the martial arts jargon), who in many occasions performs some movements that must be copied and imitated by the learners. In addition, there are a large number of audiovisual media that allow individual training, but this type of material is of little use for collaborative learning (e.g., to support combat practice) and in no case this approach provides a guide to whether the movements are correctly executed or not by the learner, and how to improve them (and avoid injuries). In fact, there is an opportunity to delve into the personalization of motor skills learning for martial arts practitioners [ 7 ]. In that work several parameters to deliver feedback that helps the practitioner to improve the execution of the technique are identified, such as breathing, energy consumed, explosiveness, acceleration and correct position. Furthermore, that paper analyses the work done so far to personalize the learning of martial arts movements and highlights two challenges: 1) to improve the modelling of movements in pairs, for example, in combat, since currently research focuses on postural detection or striking movements, but not on the execution of techniques between two practitioners; and 2) to improve the interactive design to make the virtual environment more realistic, building intelligent environments that offer multisensory feedback that includes other channels beyond the visualization of the techniques, such as auditory (through voice commands) and tactile (through vibration or physically guiding movements directly).

Thus, there are already several scientific articles that discuss the need for intelligent tools applied to the teaching of psychomotor skills [ 4 ][ 5 ]. In the same sense, a recent systematic review [ 8 ] on intelligent psychomotor tutoring systems shows the lack of this type of tools, as only nine psychomotor intelligent systems were found. This outcome is being revised in an on-going review of the psychomotor learning field [ 9 ]. 2

Related works

In our current review of the literature, no intelligent psychomotor systems have been found that are useful for martial arts training in pairs. On the one hand, there are video based applications to show how to perform Aikido techniques in pairs [ 11 ] and virtual reality based systems that presents physically simulated opponents [ 12 ]. In addition, the applicability of sensors for performance analysis in combat sports has also being reviewed [ 13 ]. On the other hand, there are educational approaches based on STEAM (Science, Technology, Engineering, Arts and Maths) education that uses sensors and computer vision tools to develop educational material for scientific teaching that involves the execution of martial arts movements in pairs, such as the kote-gaeshi (projection technique - or immobilization - by twisting of the wrist) defense in Aikido martial art to show the concept of physics torsion [ 14 ], but it does not directly support the learning of the motor skills involved. In addition, neural networks have been integrated in a mobile app to learn American Kenpo Karate movements on-line, but currently only serves for arm movements performed individually [ 10 ].

For video recording, works that present advances in relation to the use of fixed cameras have been analyzed, which propose the use of drones. Drones (as in [ 15 ] and [ 16 ]) improve the video taking since i) they provide more flexibility to capture movements, either by fixing the recording point (for example, the front of the knee) although the user moves and turns, and ii) they can direct the point of interest to any part of the body at any time simply by moving the drone around the participant (at a sufficient distance that it does not disturb him or her) [ 17 ]. In this context, computer vision [ 18 ] is being widely used for facial recognition, identification of objects or identification of movements. There are computer vision algorithms [ 19 ] capable of recreating the movement that a human is performing, recreating the skeleton of the human body and imitating the movements that an individual performs during their activity. Some systems use virtual reality environments (as in [ 20 ] and [ 12 ]) to make the training experience of martial arts more immersive, but do not delve into the creation of intelligent training tools in real time.

Although there are works dealing with the monitoring and sensing of martial arts, the objectives set out in our research are not fulfilled: to provide multisensory personalized feedback when training physical activity in pairs that is monitored with multimodal data coming from different types of capture devices. For this reason, we have designed KUMITRON (KUMITe+dRONe), an Artificial Intelligence (AI) system for martial arts learning that supports personalized training in pairs during a karate combat (kumite) using drones for image capture (among other sensing technologies). 3

KUMITRON System

In this section we describe the system we are building to monitor the interaction between various practitioners and provide intelligent advice in real time in the combat practice of martial arts in general, and Karate in particular, where the combat practice is called kumite. The development of KUMITRON [ 21 ][ 22 ] was inspired by the state of the art: sensors to collect motion related data [ 13 ][ 23 ], existing karate systems [ 12 ][ 24 ] and sports [ 25 ][ 26 ], drones for dynamic video gathering [ 15 ][ 16 ][ 17 ], and computer vision algorithms that combine video and inertial information [ 19 ] [ 27 ]. 3.1

Designing the KUMITRON system

For the design of the system, user centered design methods have been applied, that is, the system design has taken into account the information received from expert users (i.e., martial art teachers or senseis) and learners. In this way, the points of view of the training needs of both those who are going to receive the teaching of kumite practice and those who are going to teach it were gathered with a questionnaire (Learners: N=15; male=7/female=8; age range=10-65 years; Teachers: N= 5; all male; age range = +25 years). The main outcomes that were used as input to design the system were: 1) the majority (teachers = 100% and learners = 93.3%) of the respondents are positive about the use of new technologies in training, and 2) the following features were considered of interest: i) inertial, physiological and video data should be shown in real time; ii) demand analysis of the kumite in real time and its storage in a database for post-analysis; iii) learners want to get feedback in real time, and iv) both teachers and learners are interested in movements, technique, sensor data, and kumite strategy.

In addition to the questionnaire, the four stages framework proposed in [ 5 ] to build intelligent psychomotor systems has been applied in KUMITRON, as shown in Fig 1.

KUMITRON system current implementation

As shown in Fig. 2, KUMITRON uses a drone to capture the video with the best viewing angle of both karatekas, which is processed with OpenCV filters. Karatekas wear an Arduino board that collects inertial (accelerometer and gyroscope) and physiological (pulse meter) signals that are synchronized and displayed in real-time. The sensei visualizes this data in real time, so he or she can give audio instructions to the karatekas during the kumite training. The application is connected to WEKA data mining suit via sockets to generate automatic expert recommendations, but data collection (following the GDPR [ 28 ]) needs to be performed to train the algorithms.

KUMITRON expected benefits

Karate has several benefits on the individual [ 29 ][ 30 ]. As discussed in [ 21 ], KUMITRON takes advantage of Karate practice and enriches it with intelligent support, which expects to provide the following benefits: • Innovation in martial arts teaching: KUMITRON introduces new technologies in the dojo (the gym where martial arts are practiced), improving and transforming it into a “smart dojo”. In this sense, KUMITRON supports the analysis and evaluation of psychomotor learning through the collection of data from the sensors worn by the learner, comparing the learner’s current data with historical data and expert datasets, so that the activity and development of the learner can be measured. In fact, the application is supported by ubiquity and mobility capabilities to be able to track the motion activity in real time and thus provide appropriate formative multisensorial personalized feedback (currently visual and aural). In addition the sensei can use KUMITRON to record teaching material (both of the movements involved in the kumite to be learn as well as to learn physics concepts as in Phy+Aik [ 14 ]) using the services on the web to upload MOOC-type didactic material to support the learner. • Physical condition improvement and peripheral vision: Physiological values (pulse and calculated effort) are tracked to improve fatigue management during combat. Applying analytics and AI algorithms can improve effort management on a mat. Despite the fact that KUMITRON is a tool for training in pairs, it also allows individual training to improve skills such as peripheral vision, as explored in [ 31 ]. • Kumite tactic improvement: KUMITRON applies computer vision filters in real time to the video captured with the drone in order to analyze the movements applied in the combat and provide multisensorial personalized feedback. The inertial data received from the sensors is visualized through a motion vector that illustrates the direction of movement of the practitioners, being able to generate warnings of change of direction and anticipate movements of the fighters. Thus, AI is used for the analysis of the video images in real time and to obtain expert advice through predictive algorithms. In addition, learners can access the recorded sessions to analyze the technical errors made during the kumite training. • Kumite strategy improvement: AI applied to historical data can provide expert knowledge to select the strategy to be applied during the performance of a kumite based on various factors such as physical characteristics, type of the most effective technique, historical data of victories, etc. 3.4

Outcomes from studies using KUMITRON

KUMITRON brings multiple improvements not only to kumite training, but also to combat skills. As an example, in [ 31 ] we report an intra-subject study exploring if KUMITRON can support individual peripheral vision training using OpenCV motion filters. The experiment was carried out with two subjects, a female (46 years old) white belt and a male (46 years old) green belt. The experiment was performed through the analysis of video images, comparing the filtered images with OpenCV (using the Tracking Mode algorithm) with the raw images recorded with the drone.

The experiment was as follows: one of the karateka threw tsukis (fist punches) from either left and right sides in several series of 60 blows from a room while recorded by a mini-drone that sent the image (using KUMITRON infrastructure) to the computer in another room, where the other karateka watched the videos (either raw video or filtered) and had to anticipate the origin of the attack/blow (left or right) in the shortest possible time. The order of raw and filtered videos was changed among series and between the participants to avoid effects in the order of the data collection.

The results (shown in Fig. 3) suggest that the use of some motion filters of OpenCV library can increase the success rate and decrease the time required to detect the direction of the opponent’s punch. As discussed in [ 31 ], this findings are of relevance since the training of body reading in anticipation of the opponent’s movements is not only useful in Karate performance but it is likely that peripheral vision ability can be trained with this practice.

Exploring Personalized support using Multimodal Data

In Karate there are three main forms of kumite [ 32 ]: i) Jihu Ippon kumite, which consists in a preset match, ii) Jihu kumite, which is a free combat, and iii) Shiai kumite, which is a regulated or competitive combat. KUMITRON is a tool mainly aimed at preparing karateka for Shiai kumite, or regulated combat. With this objective, KUMITRON is being designed to analyze the strategies of the karatekas and try to give expert advice during combat at a physical, tactical and strategical level, as commented in section 3.3.

In turn, there are three main strategies in a kumite: i) Sen no Sen (Taking the initiative) where both karatekas initiate movement simultaneously, ii) Sen Sen no Sen (Superior Initiative or Anticipate Initiative), where both karatekas are ready and determined to attack, and iii) Go No Sen (Defense Initiative), where each karateka remain calm and carefully observe the opponent.

Different tactics [ 33 ] are developed on these general strategies, which should be trained by learners according to their psychomotor abilities. A specific tactic that can be implemented in KUMITRON is counterattack, a very effective technique [ 34 ] as it can be seen in Fig. 4. The idea is that when a learner is training against a defensive karateka, KUMITRON warn the learner when he or she is going to receive a counterattack from the opponent.

To predict a counterattack, KUMITRON can process the inertial and physiological data obtained from the sensors. For this, we are exploring the heart rate and heart rate variability during training [ 35 ][ 36 ]. An example of heart rate variability processed with KUMITRON is shown in Fig. 5, where it can be seen the distance between beats.

Moreover, KUMITRON can analyze the opponent's direction through a direction vector (shown in Fig. 6) obtained from the inertial sensors. In this way, it can show both the directions of the karateka who is following a defensive strategy (backs away) represented by the blue vector (left image) and, the directions of the karateka who is tackling (advances) represented by the orange vector (right image).

With the above information, the feedback in KUMITRON can be implemented to warn that a counterattack is about to be launched. In particular, we hypothesize that a counterattack is being prepared when the karateka goes backward and at the same time a pulse change occurs (since stressful emotional parameters and sudden movement changes can cause the pulse to increase, so a counterattack is expected to generate a spike in heart rate). Currently we are designing a user study to validate this hypothesis. For the experiment, KUMITRON will launch an on-screen message of “Warning!!!” as shown in Fig. 7 and speak it aloud to the practitioner by synthesize voice through the earphones. In the user study we will also explore if these are the most appropriate multisensorial channels to deliver the feedback. Finally, in order to predict the movement of the opponent in the counterattack, we are also exploring (see Fig. 8) how to combine deep learning computer vision algorithms such as OpenPose [ 37 ] that builds a skeleton of the movement, with motion indicators trained with Weka. Moreover, we are adding OpenPose face detection algorithms to analyze the change in the emotional state of the opponent during the kumite as we hypothesize that this change might also reflect the intention of the opponent regarding the next movement to be executed. In this way, we are working on processing the multimodal data collected using diverse sensors (physiological and inertial) as well as a drone camera in order to provide multisensorial personalized feedback computed with machine learning and computer vision algorithms. This multisensorial feedback currently combines visual and aural information and is aimed to anticipate the movements that a fighter is going to perform on the mat, in such a way that it can provide added value for training and offer expert personalized advice on the type of strategy to follow in order to win the kumite.

Conclusions and On-going works

KUMITRON is an AI-based tool that aims to provide personalized intelligent support in Karate training. Computer vision algorithms are being explored to train peripheral vision and thus, help to improve the anticipation of the opponent's attacks during a kumite. In addition, the different sensors send information (inertial and physiological) to be processed in combination, and thus can provide expert knowledge about the reactions of the human body during high-demand exercise that is being performed in pairs. In particular, we have explored the personalized support that can be provided with the multimodal data collected to prepare for a counterattack in a regulated kumite considering the physiological information obtained with the pulse meter (i.e., heart rate and heartrate variability) and the opponent's direction obtained with the inertial sensors. Moreover, we are also exploring OpenPose deep learning computer vision algorithm in two ways: i) to predict the movement of the opponent in the counterattack by considering the motion indicators trained with Weka within the OpenPose processing, and ii) to analyze the change in the emotional state of the opponent’s face during the kumite as this change might reflect the intention of the opponent regarding the next movement to be executed. In this way, warning signals are to be identified to train karatekas in anticipating to the opponent’s movements. In this context, the integration of virtual glasses could extend KUMITRON for individual combat training with imaginary and simulated opponents, where learners could train without the supervision of the sensei, but with the personalized intelligent support provided by the system.

Besides the on-going developments mentioned previously and the consideration of other computer vision algorithms and machine learning techniques to obtain more relevant information for the feedback (as well as to consider tactile feedback in addition to visual and aural) and build the corresponding models for the adaptation purpose (i.e., user, expert and domain), we are also interested in extending the sensors of the system with a chest wearable that includes a breathing sensor to measure the respiratory rate and hence, enrich the physiological data obtained in KUMITRON (which currently only includes data collected with a pulse meter). This sensor can complement the affective information obtained with OpenPose face detection algorithms. The rationale behind is to extend the personalized support to consider another Bloom’s domain (i.e., the affective one) in addition to the psychomotor and thus, analyze the impact of emotions during the kumite training. In this way, it could be possible to model certain emotional states such as concentration and relaxation, or at least, changes between different emotional states that provide relevant information to be used in the feedback delivery.

Moreover, we are also exploring the technological support provided in psychomotor teaching in other arts such as dance where technologies for motion analysis such as motion capture and whole-body interaction have been applied and provide some annotation mechanism that helps to define a ground truth against to which compare the results of different implementations ([ 38 ], [ 39 ] and [ 40 ]). In addition wearable technology has also been used in dance systems to build user models that can personalize dance learning [ 41 ].

Acknowledgements

Sincerely thank all the people involved in the development of the project, without whose contribution and support it would not have been possible to bring it to fruition. Special thanks to the martial community in general and Karate in particular for their participation in answering the different questionnaires sent. Specifically, we would like to thank Victor Sierra and Marcos Martín at Idavinci. The work is also partially supported by the project INT2AFF funded under Grant PGC2018-102279-B-I00 (MCIU/AEI/FEDER, UE) by the Spanish Ministry of Science, Innovation and Universities, the Spanish Agency of Research and the European Regional Development Fund (ERDF).

References

1. W. D. J. Rohwer and K. Sloane , Bloom's taxonomy : a forty-year retrospective . Chicago: Chicago, IL: The University of Chicago Press, 1994 .

2. J. Schneider , D. Börner , P. Van Rosmalen , and M. Specht , “ Augmenting the senses: A review on sensor-based learning support,” Sensors (Switzerland) , vol. 15 , no. 2 , pp. 4097 - 4133 , Feb. 2015 , doi: 10.3390/s150204097.

3. A. Harrow , A Taxonomy of the Psychomotor Domain: A Guide for Developing Behavorial Objectives . New York, NY, USA: David McKay Co., 1972 .

4. O. C. Santos , “ Artificial Intelligence in Psychomotor Learning: Modeling Human Motion from Inertial Sensor Data,” Int. J. Artif . Intell. Tools , vol. 28 , no. 4 , Jun . 2019 , doi: 10.1142/S0218213019400062.

5. O. C. Santos , “ Training the Body: The Potential of AIED to Support Personalized Motor Skills Learning,” Int. J. Artif . Intell. Educ., vol. 26 , no. 2 , pp. 730 - 755 , Jun. 2016 , doi: 10.1007/s40593-016-0103-2.

6. O. C. Santos and A. Corbi , “ Can Aikido Help with the Comprehension of Physics? A First Step towards the Design of Intelligent Psychomotor Systems for STEAM Kinesthetic Learning Scenarios ,” IEEE Access, vol. 7 , pp. 176458 - 176469 , 2019 , doi: 10.1109/ACCESS. 2019 . 2957947 .

7. O. C. Santos , “ Psychomotor learning in martial arts: An opportunity for user modeling, adaptation and personalization,” in UMAP 2017 - Adjunct Publication of the 25th Conference on User Modeling, Adaptation and Personalization, Jul. 2017 , pp. 89 - 92 , doi: 10.1145/3099023.3099107.

8. L. M. Neagu , E. Rigaud , S. Travadel , M. Dascalu , and R. V. Rughinis , “ Intelligent tutoring systems for psychomotor training - a systematic literature review , ” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) , Jun. 2020 , vol. 12149 LNCS, pp. 335 - 341 , doi: 10.1007/978-3- 030 -49663-0_ 40 .

9. O. C. Santos and A. Casas , Chapter II-8. Intelligent Systems for Psychomotor Learning , in Handbook of Artificial Intelligence in Education. Edward Elgar Publishing (in progress) , 2021 .

10. A. Casas-Ortiz and O. C. Santos , “ KSAS: An AI Application to learn Martial Arts Movements in on-line Settings,” in 22nd International Conference on Artificial Intelligence in Education (AIED 2021 ), 2021 , Interactive event . Available at: https://aied2021.science.uu.nl/wp-content/uploads/2021/05/ KSAS-An-AI-Application-tolearn-Martial-Arts-Movements-in-on-line- Settings .pdf, 2021 , pp. 1 - 4 .

11. Aspamya Dreams Ltd, “ Aikido3D - Home,” 2003 . https://www.aikido3d.com/ (accessed Mar. 28 , 2021 ).

12. T. M. Takala , Y. Hirao , H. Morikawa , and T. Kawai, “ Martial Arts Training in Virtual Reality with Full-body Tracking and Physically Simulated Opponents,” in 2020 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW) , May 2020 , pp. 858 - 858 , doi: 10.1109/vrw50115. 2020 . 00282 .

13. M. Worsey , H. Espinosa , J. Shepherd , and D. Thiel , “ Inertial Sensors for Performance Analysis in Combat Sports: A Systematic Review ,” Sports, vol. 7 , no. 1 , p. 28 , Jan . 2019 , doi: 10.3390/sports7010028.

14. A. Corbi , O. C. Santos , and D. Burgos , “ Intelligent Framework for Learning Physics with Aikido (Martial Art) and Registered Sensors,” Sensors, vol. 19 , no. 17 , p. 3681 , Aug . 2019 , doi: 10.3390/s19173681.

15. P. H. Han, Y. S. Chen , Y. Zhong , H. L. Wang , and Y. P. Hung , “ My Tai-Chi coaches: An augmented-learning tool for practicing Tai-Chi Chuan ,” in ACM International Conference Proceeding Series, Mar. 2017 , pp. 1 - 4 , doi: 10.1145/3041164.3041194.

16. J. La Delfa , M. A. Baytas , R. Patibanda , H. Ngari , R. A. Khot , and F. “Floyd” Mueller, “Drone Chi: Somaesthetic Human-Drone Interaction ,” in Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, Apr . 2020 , pp. 1 - 13 , doi: 10.1145/3313831.3376786.

17. X. Zhou , S. Liu, G. Pavlakos , V. Kumar , and K. Daniilidis , “ Human Motion Capture Using a Drone,” Proc. - IEEE Int. Conf. Robot . Autom., pp. 2027 - 2033 , Apr. 2018 , Accessed: Jan. 16 , 2021 . [Online]. Available: http://arxiv.org/abs/ 1804 .06112.

18. D. G. Lowe, “ Distinctive image features from scale-invariant keypoints,” Int. J. Comput . Vis., vol. 60 , no. 2 , pp. 91 - 110 , Nov. 2004 , doi: 10.1023/B:VISI. 0000029664 .99615.94.

19. C. Malleson , J. Collomosse , and A. Hilton , “ Real-Time Multi-person Motion Capture from Multi-view Video and IMUs,” Int. J. Comput . Vis., vol. 128 , no. 6 , pp. 1594 - 1611 , Jun. 2020 , doi: 10.1007/s11263-019-01270-5.

20. F. Mayweather , “Mayweather Boxing + Fitness,” https://mayweather.fit/, 2018 . https://mayweather.fit/ (accessed Mar. 28 , 2021 ).

21. J. Echeverria and O. C. Santos , “ KUMITRON: Learning in Pairs Karate related skills with Artificial Intelligence support ,” in 22nd International Conference on Artificial Intelligence in Education (AIED 2021 ), Interactive event . Available at: https://aied2021.science.uu.nl/wp-content/uploads/2021/05/Kumitron.pdf, 2021 , pp. 1 - 3 .

22. J. Echeverria and O. C. Santos , “KUMITRON: Artificial intelligence system to monitor karate fights that synchronize aerial images with physiological and inertial signals,” in International Conference on Intelligent User Interfaces , Proceedings IUI , Apr. 2021 , pp. 37 - 39 , doi: 10.1145/3397482.3450730.

23. V. Camomilla , E. Bergamini , S. Fantozzi , and G. Vannozzi, “ Trends Supporting the In-Field Use of Wearable Inertial Sensors for Sport Performance Evaluation: A Systematic Review ,” Sensors, vol. 18 , no. 3 , p. 873 , Mar . 2018 , doi: 10.3390/s18030873.

24. T. Hachaj, M. R. Ogiela , and M. Piekarczyk , “ The open online repository of karate motion capture data: A tool for scientists and sport educators ,” in 2017 IEEE Symposium Series on Computational Intelligence , SSCI 2017 - Proceedings, Feb. 2018 , vol. 2018-Janua , pp. 1 - 5 , doi: 10.1109/SSCI. 2017 . 8285270 .

25. J. Tholander and C. Johansson , “ Design qualities for whole body interaction - Learning from golf, skateboarding and bodybugging,” Nord. 2010 Extending Boundaries - Proc. 6th Nord. Conf . Human-Computer Interact ., pp. 493 - 502 , 2010 , doi: 10.1145/1868914.1868970.

26. M. P. Wozniak et al., “Subtletee: Augmenting Posture Awareness for Beginner Golfers,” Proc. ACM Human-Computer Interact ., vol. 4 , no. ISS , pp. 1 - 24 , Nov. 2020 , doi: 10.1145/3427332.

27. N. Haralabidis , D. J. Saxby , C. Pizzolato , L. Needham , D. Cazzola , and C. Minahan , “ Fusing Accelerometry with Videography to Monitor the Effect of Fatigue on Punching Performance in Elite Boxers ,” Sensors, vol. 20 , no. 20 , p. 5749 , Oct . 2020 , doi: 10.3390/s20205749.

28. CE, “ I (Legislative acts) REGULATIONS REGULATION (EU) 2016/679 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repeali ,” 2016 .

29. G. Funakoshi, MY WAY OF LIFE . Kodansha America, Inc, 1981 .

30. F. Polat , “ Inclusion in education: A step towards social justice,” Int. J. Educ . Dev., vol. 31 , no. 1 , pp. 50 - 58 , Jan. 2011 , doi: 10.1016/j.ijedudev. 2010 . 06 .009.

31. J. Echeverria and O. C. Santos , “ Punch Anticipation in a Karate Combat with Computer Vision,” in 29th ACM Conference on User Modeling, Adaptation and Personalization (ACM UMAP 2021 ), 2021 , ( accepted ), doi: 10.1145/3450614 .3461688.

32. Real Federación Española de Karate, “RFEK Normativa de Competición Kumite y Kata, ” Www.Rfek.Es , vol. 2019 , no. Enero , p. 107 , 2019 , [Online]. Available: http://www.rfek.es/images/pdf/normativas/NORMATIVA-ARBITRAJE2019.pdf.

33. F. Torres Baena , “ El karate deportivo, estudio praxiológico; análisis del kumite deportivo en la modalidad de shiai-kummite : (estudio histórico y kinantropométrico ),” Universidad de Las Palmas de Gran Canaria , 1997 .

34. Y. Ramasamy and E. Suwarganda, “ Attacking sequences by Winners and Losers in Karate Kumite during the Sea Games , 2013 .,” 2014 , no. November, doi: 10.13140/2 .1.4658.9122.

35. S. Laborde , A. Brüll , J. Weber , and L. S. Anders , “ Trait emotional intelligence in sports: A protective role against stress through heart rate variability?,” Pers. Individ. Dif. , vol. 51 , no. 1 , pp. 23 - 27 , 2011 , doi: 10.1016/j.paid. 2011 . 03 .003.

36. J. Achten and A. E. Jeukendrup , “ Heart rate monitoring applications and limitation ,” vol. 33 , no. 7 , pp. 517 - 538 , 2003 .

37. Z. Cao , G. Hidalgo, T. Simon , S.-E. Wei , and Y. Sheikh , “ OpenPose: Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields,” IEEE Journals Mag . | IEEE Xplore, Art. no. Vol. 43 , 2021 , Accessed: Jun. 06 , 2021 . [Online]. Available: https://ieeexplore.ieee.org/abstract/document/8765346.

38. K. El Raheb , G. Tsampounaris , A. Katifori , and Y. Ioannidis , “ Choreomorphy: A wholebody interaction experience for dance improvisation and visual experimentation , ” May 2018 , doi: 10.1145/3206505.3206507.

39. K. El Raheb , A. Kasomoulis , A. Katifori , M. Rezkalla , and Y. Ioannidis , “ A web-based system for annotation of dance multimodal recordings by dance practitioners and experts ,” Jun. 2018 , doi: 10.1145/3212721.3212722.

40. K. El Raheb , M. Stergiou , A. Katifori , and Y. Ioannidis , “ Dance interactive learning systems: A study on interaction workflow and teaching approaches,” ACM Computing Surveys , vol. 52 , no. 3. Association for Computing Machinery, Jun. 01 , 2019 , doi: 10.1145/3323335.

41. A. D. P. Dos Santos , K. Yacef , and R. Martinez-Maldonado , “ Let's Dance: How to build a user model for dance students using wearable technology,” in UMAP 2017 - Proceedings of the 25th Conference on User Modeling, Adaptation and Personalization , Jul. 2017 , pp. 183 - 191 , doi: 10.1145/3079628.3079673.