TELKOMNIKA Telecommunication, Computing, Electronics and Control

Vol. 18, No. 4, August 2020, pp. 1926~1933

ISSN: 1693-6930, accredited First Grade by Kemenristekdikti, Decree No: 21/E/KPT/2018
DOI: 10.12928/TELKOMNIKA.v18i4.14730  1926

Deep learning in sport video analysis: a review

Keerthana Rangasamy1, Muhammad Amir As’ari2, Nur Azmina Rahmad3,

Nurul Fathiah Ghazali4, Saharudin Ismail5
1,2,3,4,5School
of Biomedical Engineering and Health Sciences, Faculty of Engineering,
Universiti Teknologi Malaysia, Malaysia
2Sport Innovation and Technology Center, Institute of Human Centered Engineering,

Universiti Teknologi Malaysia, Malaysia

Article Info ABSTRACT

Article history: Sport is a competitive field, where it is an element of measurement for
a countries development. Due to this reason, sport analysis has become one
Received Dec 2, 2019 of the major contribution in analysing and improving the performance level
Revised Mar 2, 2020 of an athlete. Video-based modality has become a crucial tool used in sport
Accepted Mar 13, 2020 analysis by coaches and performance analysis. There were wide variety
of techniques used in sport video analysis. The main purpose of this review
paper is to compare and update review between traditional handcrafted
Keywords: approach and deep learning approach in sport video analysis based on human
activity recognition, overview of recent study in video based human activity
Deep learning recognition in sport analysis and finally concluded with future potential
Human activity recognition direction in sport video analysis.
Sport video analysis

This is an open access article under the CC BY-SA license.

Corresponding Author:
Muhammad Amir As’ari,
Department of Biomedical Engineering,
School of Biomedical Engineering and Health Sciences,
Faculty of Engineering,
Universiti Teknologi Malaysia,
Johor Bahru, Malaysia.
Email: [email protected]

1. INTRODUCTION
Sport is an activity that involves the active movement of an athlete either in the form of a team
or individual to complete with the opposite team [1]. It is one of the elements of measurement for a countries
development as it brings a good reputation to a country if its team win in the game [2]. In order to achieve
high performance in sport by the athletes, coaches and sports professionals play a vital role to evaluate
and train their athletes [3]. Among various methods, sport video analysis is one of the methods to evaluate
the performance level of athletes and to enhance training techniques [4]. A process to assess the performance
level of an athlete is known as performance analysis [5]. Performance analysis can be divide into two which
are technical analysis and tactical or notational analysis [6]. Through technical analysis, we would get
an answer to the question: how does the game is performed by the athletes. On the other hand, through
tactical or notational analysis the question of what activity is performed would be answered [7].
Over the past few years, in the field of computer vision, different approaches have been
implemented to analyse sport videos. During the early stage traditional handcrafted approaches were
proposed by many researchers for sport video analysis [8]. After the development of technology,
deep leaning approach has become the current topic of interest in sport video analysis due to its successful

Journal homepage: http://journal.uad.ac.id/index.php/TELKOMNIKA

TELKOMNIKA Telecommun Comput El Control  1927

performance in other computer vision areas such as human-computer interface, handwriting recognition
and speech recognition [9]. Generally, sport video analysis using such approach can be divide into several
directions such as tracking players as well as ball, trajectory detection and action recognition. This paper
aims to review current trends in deep leaning approach in human activity recognition-based sport video
analysis. The remainder of this paper is organized as section 2 illustrates traditional handcrafted architecture.
Next section 3 presents deep learning architecture. Section 4 discusses the drawback of state of the art works
and Section 5 present conclusion and future works to address.

2. TRADITIONAL HANDCRAFTED ARCHITECTURE

In this section, overview of handcrafted architecture and its application in sport video analysis was
studied based on prior research.

2.1. Overview of handcrafted architecture

Before the advancement of technology and the emergence of deep learning architecture, sport video
analysis is mainly performed using handcrafted features [9]. Traditional handcrafted features are human
design features for specific problems [10]. It consists of feature descriptors and feature extractors that used to
manually extract the important information from the sport video for further analysis. However,
these traditional handcrafted features could only execute low-level features extraction [11]. Further in this
section would like to present the existing traditional handcrafted architecture that was used widely
in computer vision to recognize activity in video.
During the prehistoric ages of sport video analysis, action recognition was performed using
manually human design features to extract and represent features for action recognition by which this
technique is called a handcrafted approach. At first, low-level action features such as histogram of optical
flows (HOF), histogram of gradients (HOG), sparse spatial-temporal interest points (STIP) features were
designed to extract features in videos for analysis [9]. Then, those sparse features were embellished to dense
spatio-temporal features [12]. After some time, improved dense trajectories (iDT) were innovated which
comprises optical flow and speeded up robust features (SURF) [13]. To further enhance iDT multi-layer
staked fisher vector (FV) was developed [14]. Multi-skip Feature staking was also used for feature extraction
of action recognition in videos which improve the performance of action recognition [15]. Although,
these handcrafted features were evolved with time and also improved in performance but it could only
perform in a specific problem. It is not flexible to be used in other datasets. Table 1 shows some
of the existing popular handcrafted features used in action recognition for video analysis such as sport
video analysis.

Table 1. List of popular handcrafted features used for action recognition in videos
Handcrafted features Reference
Histogram of Optical Flow (HOF) and [16, 17]
Histogram of Gradients (HOG)
Sparse Spatio-temporal interest points [18]
Improved Dense Trajectories (iDT) [13]
Fish Vector (FV) [14]
Multi-layer Feature Stacking (MLFS) [15]

2.2. Sport video analysis using handcrafted features

During the early stages of video analysis in computer vision such as in sports, handcrafted features
were used as mentioned earlier. For instance, Abdulmunem et al. [19] were proposed a method for action
recognition based on salient object spotting using local and global descriptors in KTH and UCF-Sports
datasets. In this method, they were using handcrafted 3D-SIFT-HOOF (SGSH) features which were pass
through SVM after encoding with bag of visual words approach. Besides, Carnonneau et al. [20] presented a
novel technique to identify play break events in hockey videos. They used bag-of-words event detectors to
identify the key events like line-change and play-break STIP were used in final decision making by creating
context descriptor. Figure 1 the shows schematic block diagram of the event detector used in [20].
Zhu et al. [7] introduced a novel method to detect an event in a soccer games to extract tactical
information. In this research work, they employ multi-object trajectory and field locations of the event shots
to recognize the semantic event in the soccer games. Moreover, Lee et al. [21] implement pattern matching
techniques to automatically recognize events in soccer games using a multi-object tracking unit and motion
recognizer. Their proposed block diagram is shown in Figure 2.
Lien et al. [22] studied scene-based event detection for baseball videos by implementing various
handcrafted features such as image-based features, object-based features and global motion to capture

Deep learning in sport video analysis: a review (Keerthana Rangasamy)

1928  ISSN: 1693-6930

semantic information from the baseball sport video. Then those, extracted features were fed into hidden
Marvon model (HMM) to classify the detected events. Chen et al. [23] developed a novel handcrafted model
based on player trajectories reconstruction method which includes court detection, player detection, player
tracking and homography transformation in broadcasted basketball videos.

Figure 1. Schematic block diagram of event detectors in [20]

Figure 2. Block diagram of soccer event detection system [21]

Lu et al. [24] constructed a system to automatically track various hockey players as well as
recognize their activity concurrently in broadcasted hockey video. In this study, HOG descriptors were
utilised to detect the players and to model the appearance of the players a probabilistic framework was
designed with a mixture of local subspaces. Additionally, boosted particle filter (BPF) was employed in this
tracking system. Mukherjee et al. [25] proposed a novel descriptor based on improved dense trajectories in
human action and event recognition. Their design also used Fisher Vector (FV) and the proposed novel
approach was trained with binary support vector machine (SVM). The dataset used for this research was UCF
sports, CMU Mocap and Hollywood2 datasets for event and action recognition.

3. DEEP LEARNING
In this section, overview of deep learning archicture and its application in sport video analysis was
reviewed based on previous studies.

3.1. Overview of deep learning architecture

Deep learning architecture works automatically by directly classifying raw input images or video
frames through multiple processing layers so as to learn and represent data [26]. Unlike traditional
handcrafted architecture, it does not requires any feature descriptors or feature extractors. For instance,
deep learning architecture uses local perception, down pooling, weight sharing, a multi-convolution kernel,
etc. to automatically learn local features from just a segment of an image rather than whole image [27].
Deep learning techniques able to classify high-level or complex action recognition which attracts huge
research of interest [28]. Examples of widely used deep learning models is convolutional neural network
(CNN), recurrent neural network (RNN) and long short-term memory (LSTM).
Excellent performance with overwhelming accuracy of deep learning in visual task inspired
the exploitation of deep learning in video analysis. Initially, CNN works independently to extract information
from the still images [29]. However, 2D-CNN fails to extract temporal information in video sequences.
In order to overcome this issue, 3D-CNN is then constructed to extract both spatial and temporal information
of video frames [30]. Following this RNN was used in action recognition. RNN based method effectively

TELKOMNIKA Telecommun Comput El Control, Vol. 18, No. 4, August 2020: 1926 - 1933
TELKOMNIKA Telecommun Comput El Control  1929

capture temporal information whereby its current prediction is based on both current observations as well as
past information [31]. But in spite of that, RNN architecture only has short term memory, which could not
apply in the real-world scenario. To alleviate this problem, LSTM model was proposed. This model able to
extract temporal information from sequential video data. LSTM model has a memory unit that decides when
to remember and forget hidden states [32]. Due to its superiority, the LSTM model broadly used in computer
vision applications such as action recognition. Table 2 show a comparison between deep learning models.

Table 2. Comparison between deep learning model

Model Advantage Drawback
2D-CNN  Automatically capture  Could not able to capture the temporal
the spatial information in  information in video data
the image patches
3D-CNN  Automatically capture both  Expensive model due to 3D
spatial and temporal
information
RNN  Automatically capture  Has short memory ability, could not
the temporal information apply in real situation
in sequential data  Gradient explosion
LSTM  Automatically capture  NIL
the temporal information
in sequential data

3.2. Deep learning architecture in sport video analysis

Despite the astonishing performance of deep learning approach in various computer vision
application such as voice recognition, text recognition it also achieves outstanding results in sport video
analysis in recent years. Although it is still in the early stage of application in sport video analysis and only
very few research has been done, yet so far its performance is more accurate as compared to traditional
approaches and it’s getting more attention presently [33]. Tora et al. [34] proposed a novel deep learning
approach in classifying puck possession events in ice hockey. They used pre-trained CNN to first extract
the features then use LSTM for classification of the five types of events which are dump in, dump out,
loose puck recovery, pass and shot. Sozykin et al. [35] presented a 3D CNN based action recognition system
for multi-class imbalanced in ice hockey. They first extract features from both single image and a slice
of frames using CNN. Then, to solve the multi-class imbalance, they introduce two different strategies that
can compare its performance separately which are the ensemble of k independent single-label learning
networks and a single multi-label k-output network. It is found that as compared to the ensemble model,
single multi-label k output model achieves high performance. Longteng et al. [36] designed a joint
framework comprising of both athlete tracking and action recognition in sport videos using scaling
and occlusion robust tracker with spatial pyramid pooling network (SPP-net) [37] and LSTM to extract
motion, spatial and temporal features.
Jiang et al. [38] introduce deep learning based automatic soccer video event detection by using CNN
and RNN. This paper focused on 4 types of event which are goal, goal attempt, corner and card. CNN model
was used to extract image features. And RNN structure was used to capture temporal relation. However,
in this paper three different types of RNN structures (traditional RNN model, LSTM model and gated
recurrent unit (GRU) model) were used to determine the optimum model. LSTM is the best-performed model
among those three RNN structures. Hong et al. [39] delivered deep transfer learning based end-to-end soccer
video scene and event classification. Scene classification includes long view and close-up view while event
classification includes corner, free-kick, goal and penalty. Their classification model used CNN models as
well as transfer learning method. Whereas, Yu et al. [40] designed a deep learning-based soccer event
detection that includes event detection as well as story generation refer to Figure 3 which begins with event
clips and end with replay event. In this design CNN and LSTM were used.
Besides, in tennis sport, Mora et al. [41] have proposed a domain-specific deep learning action
recognition method by utilising pre-trained CNN with three-layered LSTM model. This paper was aim to
recognised fine-grained action in tennis sports. The deep LSTM network used in this research is able to learn
high-level structures and provides high accuracy. Apart from that, Ibrahim et al. [42] innovated a deep model
to extract dynamic temporal information in volleyball using LSTM models. The game state was deduced by
capturing players’ state through the two-stage LSTM model. Ramanathan et al. [43] established a deep
leaning method to detect and categorised basketball events using RNN. This work used an attention model to
detect key players from multi-person videos first. Then used CNN and LSTM to extract feature and detect an
event in basketball.

Deep learning in sport video analysis: a review (Keerthana Rangasamy)

1930  ISSN: 1693-6930

Figure 3. Proposed soccer story detection framework [40]

4. DISCUSSION
For the past few decades, there was quite a lot of research focusing on sport video analysis based on
handcrafted architectures. However, this handcrafted architecture which consists of feature descriptors
and extractors are problem-dependent. In other words, those handcrafted features can only apply to specific
problems. It could not be used for other datasets or problems which is irrelevant to apply in a real scenario.
Apart from that, the traditional handcrafted architecture used for sport video analysis could only extract
low-level features, it is very challenging in capturing for high-level semantic information [44].
Thus, the advancement in technology leads to the emergence of deep learning-based sport video analysis.
Since deep learning-based sport video analysis is still a new and growing research field, there were only
a few studies found. Most of the studies are focusing on soccer games, tennis, baseball and basketball [45].
Only limited resources are found on the other sports such as hockey and badminton. However, the main
drawback of using deep learning models are its data-hungry meaning to learn features automatically from
raw inputs data it is depends on thoudsands of input data [46]. Besides, to elevate the performance level
of the deep learning model it needs high-performance GPUs [47]. But, the recent advancement in technology
and the growth in big data have overcome these issues.
Previously, CNN model which is one of the deep learning approach has shown tremendous success
in image recognition, speech recognition audio recognition, etc [48]. However, in the analysis of video input
data researchers face many challenges because video sequences dynamically evolve with time. It is difficult
to extract temporal information. With the continuous study in video analysis using deep learning, cause
the establishment of a sequential models such as RNN and LSTM. These models are able to extract temporal
information in video input data. There were some researches work on the combination of both CNN
and LSTM model to extract spatio-temporal information. But only a few researches were found in extraction
high-level semantic information in sport video analysis. Despite astonishing performance of deep learning
based archicture, the advancement achieves in image classification have not been reached in certain field like
video classification or sport video analysis [49]. It is still an open issue in deep learning-based research in
which many researchers try to solve and it is an ongoing research work [50].

5. CONCLUSION
This paper contributes a comprehensive survey on sport video analysis by comparing both
handcrafted and deep learning approach. In summary, deep learning approach has overcome the limitations
encountered by traditional methods in activity recognition of sport video analysis. However, only a few
research has focused on sport video analysis. So, in future studies, the researchers can focus on extraction
high-level semantic information in sport analysis which will be used by coaches and sports professionals in
evaluating players’ tactical performance in the game. Moreover, future research should also concentrate more
on sport videos that are apart from soccer games, tennis, baseball and basketball as almost 80% of prior
research had been focused on those sports.

ACKNOWLEDGEMENTS
The authors would like to express their appreciation to Universiti Teknologi Malaysia (UTM)
for endow this research and the Minister of Higher Education (MOHE), Malaysia for supporting this research
work under Zamalah Scholarship and Research Grant No. Q.J130000.2545.19H88.

TELKOMNIKA Telecommun Comput El Control, Vol. 18, No. 4, August 2020: 1926 - 1933
TELKOMNIKA Telecommun Comput El Control  1931

REFERENCES
[1] Lexico, “sport | Definition of sport in English by Oxford Dictionaries,” [Online]. Available:
https://en.oxforddictionaries.com/definition/sport. Accessed: 31 January 2019.
[2] D. Gu, “Analysis of tactical information collection in sports competition based on the intelligent prompt automatic
completion algorithm,” Journal of Intelligent and Fuzzy Systems, vol. 35, no. 3, pp. 2927-2936, 2018.
[3] N. Homayounfar, S. Fidler, and R. Urtasun, “Sports field localization via deep structured models,” 2017 IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), vol. 2017-January, pp. 4012–4020, 2017.
[4] M. Stein et al., “Bring It to the Pitch: Combining Video and Movement Data to Enhance Team Sport Analysis,”
IEEE Transactions on Visualization and Computer Graphics, vol. 24, no. 1, pp. 13–22, 2018.
[5] E. E. Cust, A. J. Sweeting, K. Ball, and S. Robertson, “Machine and deep learning for sport-specific movement
recognition: a systematic review of model development and performance,” Journal of Sports Sciences, vol. 37,
no. 5, pp. 568–600, 2019.
[6] N. A. Rahmad, M. A. As’ari, N. F. Ghazali, N. Shahar, and N. A. J. Sufri, “A survey of video based action
recognition in sports,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 11, no. 3,
pp. 987–993, 2018.
[7] G. Zhu et al., “Event tactic analysis based on broadcast sports video,” IEEE Transactions on Multimedia, vol. 11,
no. 1, pp. 49–67, 2009.
[8] A. Kar, N. Rai, K. Sikka, and G. Sharma, “AdaScan: Adaptive scan pooling in deep convolutional neural networks
for human action recognition in videos,” 2017 IEEE Conference on Computer Vision and Pattern Recognition
(CVPR), pp. 3376–3385, 2017.
[9] B. Meng, X. Liu, and X. Wang, “Human body action recognition based on quaternion spatial-temporal convolutional
neural network,” Yi Qi Yi Biao Xue Bao/Chinese J. Sci. Instrum., vol. 38, no. 11, pp. 2643–2650, 2017.
[10] H. Yang et al., “Asymmetric 3D Convolutional Neural Networks for action recognition,” Pattern Recognit.,
vol. 85, pp. 1–12, 2019.
[11] Z. Huiqun, W. Hui, and W. Xiaoling, “Application research of video annotation in sports video analysis,” 2011
International Conference on Future Computer Science and Education, pp. 62–66, 2011.
[12] S. Dollár, P., Rabaud, V., Cottrell, G. and Belongie, “Going deeper into action recognition : A survey,” Image and
Vision Computing, vol. 60, pp. 4–21, 2017.
[13] H. Chen, J. Chen, R. Hu, C. Chen, and Z. Wang, “Action recognition with temporal scale-invariant deep learning
framework,” China Communications, vol. 14, no. 2, pp. 163–172, 2017.
[14] X. Peng, C. Zou, Y. Qiao, and Q. Peng, “Action Recognition with Stacked Fisher Vectors,” European Conference
on Computer Vision - ECCV 2014: Computer Vision – ECCV 2014, pp. 581–595, 2014.
[15] Zhenzhong Lan, Ming Lin, Xuanchong Li, A. G. Hauptmann, and B. Raj, “Beyond Gaussian Pyramid: Multi-skip
Feature Stacking for action recognition,” 2015 IEEE Conference on Computer Vision and Pattern Recognition
(CVPR), pp. 204–212, 2015.
[16] N. Dalal, B. Triggs, and C. Schmid, “Human Detection Using Oriented Histograms of Flow and Appearance,”
European Conference on Computer Vision (ECCV ’06), pp. 428–441, 2006.
[17] N. Dalal, B. Triggs, “Histograms of oriented gradients for human detection,” 2005 IEEE Computer Society Conference
on Computer Vision and Pattern Recognition (CVPR'05), 2005.
[18] P. Dollár, V. Rabaud, G. Cottrell, S. Belongie,“Behavior recognition via sparse spatio-temporal features,” 2005 IEEE
International Workshop on Visual Surveillance and Performance Evaluation of Tracking and Surveillance, 2005.
[19] A. Abdulmunem, Y. K. Lai, and X. Sun, “Saliency guided local and global descriptors for effective action
recognition,” Computational Visual Media, vol. 2, pp. 97–106, 2016.
[20] M. A. Carbonneau, A. J. Raymond, E. Granger, and G. Gagnon, “Real-time visual play-break detection in sport
events using a context descriptor,” 2015 IEEE International Symposium on Circuits and Systems (ISCAS),
pp. 2808–2811, 2015.
[21] J. Lee, D. Nam, S. Moon, J. Lee, and W. Yoo, “Soccer Event Recognition Technique based on Pattern Matching,”
2017 Federated Conference on Computer Science and Information Systems (FedCSIS), pp. 643–646, 2017.
[22] C. C. Lien, C. L. Chiang, and C. H. Lee, “Scene-based event detection for baseball videos,” Journal of Visual
Communication and Image Representation, vol. 18, no. 1, pp. 1–14, 2007.
[23] L. H. Chen, C. W. Su, and H. A. Hsiao, “Player trajectory reconstruction for tactical analysis,” Multimedia Tools
and Applications, vol. 77, pp. 30475–30486, 2018.
[24] W. L. Lu, K. Okuma, and J. J. Little, “Tracking and recognizing actions of multiple hockey players using the
boosted particle filter,” Image and Vision Computing, vol. 27, no. 1, pp. 189–205, 2009.
[25] S. Mukherjee and K. Karan, “Human action and event recognition using a novel descriptor based on improved
dense trajectories,” Multimedia Tools and Applications volume, vol. 77, pp. 13661–13678, 2018.
[26] A. Voulodimos, N. Doulamis, A. Doulamis, and E. Protopapadakis, “Deep Learning for Computer Vision: A Brief
Review,” Computational Intelligence and Neuroscience, vol. 2018, pp. 1-13, 2018.
[27] A. Sargano, P. Angelov, and Z. Habib, “A Comprehensive Review on Handcrafted and Learning-Based Action
Representation Approaches for Human Activity Recognition,” Appllied Science, vol. 7, no. 1, pp. 110, 2017.
[28] A. Elboushaki, R. Hannane, K. Afdel, and L. Koutti, “MultiD-CNN : A multi-dimensional feature learning
approach based on deep convolutional networks for gesture recognition in RGB-D image sequences,” Expert
Systems with Applications, vol. 139, pp. 112829, 2020.
[29] B. Meng, X. Liu, and X. Wang, “Human action recognition based on quaternion spatial-temporal convolutional neural
network and LSTM in RGB videos,” Multimedia Tools and Applications, vol. 38, no. 11, pp. 26901-26918, 2018.

Deep learning in sport video analysis: a review (Keerthana Rangasamy)

1932  ISSN: 1693-6930

[30] M. Asadi-aghbolaghi et al., “A Survey on Deep Learning Based Approaches for Action and Gesture Recognition in Image
Sequences,” FG 2017 - 12th IEEE Conference on Automatic Face and Gesture Recognition, pp. 476–483, 2017.
[31] X. Yang, P. Molchanov, and J. Kautz, “Multilayer and Multimodal Fusion of Deep Neural Networks for Video
Classification,” MM '16: Proceedings of the 24th ACM international conference on Multimedia, pp. 978–987, 2016.
[32] J. Y. H. Ng, M. Hausknecht, S. Vijayanarasimhan, O. Vinyals, R. Monga, and G. Toderici, “Beyond Short
Snippets: Deep Networks for Video Classification,” 2015 IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), 2015.
[33] N. A. Rahmad, N. A. J. Sufri, N. H. Muzamil, and M. A. As’ari, “Badminton player detection using faster region
convolutional neural network,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 14, no. 3,
pp. 1330–1335, 2019.
[34] M. R. Tora and J. J. Little, “Classification of Puck Possession Events in Ice Hockey,” IEEE Conference on
Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 147-154, 2017.
[35] K. Sozykin, S. Protasov, A. Khan, R. Hussain, and J. Lee, “Multi-label class-imbalanced action recognition in
hockey videos via 3D convolutional neural networks,” arXiv:1709.01421, 2018.
[36] L. Kong, D. Huang, J. Qin, and Y. Wang, “A Joint Framework for Athlete Tracking and Action Recognition in Sports
Videos,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 30, no. 2, pp. 532–548, 2020.
[37] J. He, K., Zhang, X., Ren, S. and Sun, “Spatial Pyramid Pooling in Deep Convolutional Networks for Visual
Recognition,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 37, no. 9, pp. 1904–1916, 2015.
[38] H. Jiang, Y. Lu, and J. Xue, “Automatic soccer video event detection based on a deep neural network combined
CNN and RNN,” 2016 IEEE 28th International Conference on Tools with Artificial Intelligence (ICTAI),
pp. 490–494, 2017.
[39] Y. Hong, C. Ling, and Z. Ye, “End-to-end soccer video scene and event classification with deep transfer learning,”
2018 International Conference on Intelligent Systems and Computer Vision (ISCV), pp. 1–4, 2018.
[40] J. Yu, A. Lei, and Y. Hu, “Soccer Video Event Detection Based on Deep Learning,” International Conference on
Multimedia Modeling - MMM 2019: MultiMedia Modeling, Springer, vol. 8936, pp. 377-389, 2019.
[41] S. V. Mora and W. J. Knottenbelt, “Deep Learning for Domain-Speficis Action Recognition in Tennis,” 2017 IEEE
Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 170–178, , 2017.
[42] M. G. Ibrahim MS, Muralidharan S, Deng Z, Vahdat A, “A hierarchical deep temporal model for group activity
recognition,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1971–1980, 2016.
[43] V. Ramanathan, J. Huang, S. Abu-el-haija, A. Gorban, K. Murphy, and L. Fei-fei, “Detecting events and key actors
in multi-person videos,” 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.
[44] J. Lee, D. W. Nam, J. S. Lee, S. Moon, K. Kim, and H. Kim, “A study on composition of context-based soccer
analysis system,” 2017 19th International Conference on Advanced Communication Technology (ICACT),
pp. 886–889, 2017.
[45] H. C. Shih, “A Survey of Content-Aware Video Analysis for Sports,” IEEE Transactions on Circuits and Systems
for Video Technology, vol. 28, no. 5, pp. 1212–1231, 2018.
[46] P. Felsen, P. Agrawal, and J. Malik, “What will Happen Next? Forecasting Player Moves in Sports Videos,” 2017
IEEE International Conference on Computer Vision (ICCV), pp. 3362–3371, 2017.
[47] Z. Xu, Y. Yang, and A. G. Hauptmann, “A discriminative CNN video representation for event detection,” 2015
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1798–1807, 2015.
[48] Y. Lecun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015.
[49] I. Rodríguez-Moreno, J. M. Martínez-Otzeta, B. Sierra, I. Rodriguez, and E. Jauregi, “Video activity recognition:
State-of-the-art,” Sensors (Switzerland), vol. 19, no. 14, pp. 1–25, 2019.
[50] A. Khan, A. Sohail, U. Zahoora, and A. S. Qureshi, “A Survey of the Recent Architectures of Deep Convolutional
Neural Networks,” arXiv:1901.06032, pp. 1–62, 2019.

BIOGRAPHIES OF AUTHORS

Keerthana Rangasamy received B.E (Hons) in Biomedical Engineering from Universiti

Teknologi Malaysia (UTM) in 2018. Currently, she pursuing her PhD in Biomedical
Engineering under School of Biomedical Engineering and Health Science, Faculty of
Engineering in Universiti Teknologi Malaysia. Her research interests are artificial
inteligence, computer vision and deep learning.

TELKOMNIKA Telecommun Comput El Control, Vol. 18, No. 4, August 2020: 1926 - 1933
TELKOMNIKA Telecommun Comput El Control  1933

Dr. Muhammad Amir Bin As'ari holds a PhD in Biomedical Engineering from Universiti
Teknologi Malaysia. His Ph’s work was in the field of assistive technology, computer vision
and image processing and his work focused on developing a novel 3D shape descriptors for
recognizing the activities of daily living based on Kinect-like depth image.He pursued his
master degree and bachelor degree at Faculty of Electrical Engineering, Universiti Teknologi
Malaysia, majored in Electrical Engineering. Currectly, he is working on the development of
signal and image processing approaches with intelligent intervention in assistive technology
and rehabilitation as well as sport performance technology especially in automated human
movement recognition

Nur Azmina Rahmad received B.E (Hons) in Biomedical Engineering from

Universiti Teknologi Malaysia (UTM) in 2017. Currently, she pursuing her PhD in
Biomedical Engineering under School of Biomedical Engineering and Health
Science, Faculty of Engineering in Universiti Teknologi Malaysia. Her research
interests are artificial inteligence, computer vision and deep learning.

Nurul Fathiah Ghazali received Diploma in Electrical Engineering (Power) from

Universiti Teknologi Malaysia, Kuala Lumpur (2013) and Bachelor of Engineering in
Biomedical from Universiti Teknologi Malaysia, Johor Bahru (2017). She is
currently working on her PhD studies in Biomedical Engineering at Universiti
Teknologi Malaysia, Johor Bahru and her current research is about activity
recognition study in badminton sport using inertial sensor which collaborated with
Sport Innovation and Technology Centre (SITC) and Institute of Human Centered
Engineering (iHumEn) from UTM JB. Her research interest includes activity
recognition studies, signal processing, and machine learning.

Saharudin Ismail graduated from Bachelor of Occupational Therapy (Honours),

UKM in 2008. He pursued his Master of Philosophy in Rehabilitation Technology
on 2013 and attained his PHD in Health Science on 2019 from Universiti Teknologi
Malaysia. Most of his work is related to Assistive Devices, Rehabilitation,
Biostatistics, Motor Control, Motor Learning and Kinesiology through the application
of the inertial sensors.

Deep learning in sport video analysis: a review (Keerthana Rangasamy)