Professional Documents
Culture Documents
Artificial Intelligence Approach in
Artificial Intelligence Approach in
Biomechanics, Central
Biomechanics of Gait and Sport: A Tehran Branch, Islamic
Azad University, Tehran,
Systematic Literature Review Iran
2
Department of Sport In-
juries and Corrective Ex-
ercises, Sport Sciences
Rozhin Molavian (PhD Candidate)1 , Ali Fatahi (PhD)1* , Research Institute, Teh-
Hamed Abbasi (PhD)2, Davood Khezri (PhD)3 ran, Iran
3
Department of Sport
Biomechanics and Tech-
nology, Sport Science
ABSTRACT Research Institute, Teh-
Background: Artificial neural network helps humans in a wide range of activities, ran, Iran
such as sports.
Objective: This paper aims to investigate the effect of artificial intelligence on
decision-making related to human gait and sports biomechanics, using computer-
based software, and to investigate the impact of artificial intelligence on individuals’
biomechanics during gait and sports performance.
Material and Methods: This review was conducted in compliance with the
PRISMA guidelines. Abstracts and citations were identified through a search based on
Science Direct, Google Scholar, PubMed, Elsevier, Springer Link, Web of Science,
and Scopus search engines from 1995 up to 2023 to obtain relevant literature about the
impact of artificial intelligence on biomechanics. A total of 1000 articles were found
related to biomechanical characteristics of gait and sport and 26 articles were directly
pertinent to the subject.
Results: The extent of the application of artificial intelligence in sports biomechan-
ics in various fields. In addition, various variables in the fields of kinematics, kinetics,
and the field of time can be investigated based on artificial intelligence. Conventional
computational techniques are limited by the inability to process data in its raw form.
Artificial Intelligence (AI) and Machine Learning (ML) techniques can handle com-
plex and high-dimensional data.
Conclusion: The utilization of specialized systems and neural networks in gait
analysis has shown great potential in sports performance analysis. Integrating AI into
this field would be a significant advancement in sport biomechanics. Coaches and
athletes can develop more precise training regimens with specialized performance
prediction models.
Citation: Molavian R, Fatahi A, Abbasi H, Khezri D. Artificial Intelligence Approach in Biomechanics of Gait and Sport: A Systematic Literature
Review. J Biomed Phys Eng. 2023;13(5):383-402. doi: 10.31661/jbpe.v0i0.2305-1621.
Keywords
Artificial Intelligence; Biomechanics; Machine Learning; Performance; Sports
A
Introduction *Corresponding author:
Ali Fatahi
rtificial Intelligence (AI) and related technologies have signifi- Department of Sport Bio-
cantly impacted the scientific exploration of sports and human mechanics, Central Teh-
performance in sports biomechanics, whereby computerized ran Branch, Islamic Azad
scrutiny has been commercially viable [1]. The level of processing University, Tehran, Iran
E-mail:
capability and technological advancement has significantly increased ali.fatahi@iauctb.ac.ir
over the past twenty years, as evidenced by the current extent to Received: 19 May 2023
which computers are utilized for the biomechanical analysis of sports Accepted: 7 September 2023
388
David [33]
previous and current frame.
Artificial Intelligence in Sports Biomechanics
389
Data used to develop The most optimum model exhibited a precision level of 98%, showing the artifi-
To conduct a survey on volleyball team ranking
Tumer et al. [34] obtained from 2013 to 2015 cial neural network model has the capacity to forecast the hierarchical order of
prediction using ANN.
league chart teams in a league chart with remarkable precision.
To present a cascaded CNN that satisfies all three
Basketball and soccer The approach achieves high accuracy on gaming platforms with fewer param-
Lu et al. [35] requirements, including accuracy, efficiency, and
players eters compared to traditional CNNs, making it a lightweight alternative.
low memory consumption.
Women runners have a higher risk for running-related injuries compared to
To employ PCA and SVM to analyze differences in men. A research study revealed that a significant portion of the data's vari-
Phinyomark et 483 subjects including 263
running gait kinematics between male and female ability was accounted for by a majority of the principal components. Further-
al. [36] females and 220 males
runners. more, the study showed that classifying female and male runners using SVM
resulted in an accuracy of 86.34%.
To identify and evaluate the most proficient player An evaluation method to assess the performance of the proposed active
Pobar et al. [37] in handball by detecting all players present on the Handball players player detection was proposed, which was successfully tested on a custom
court. handball dataset.
Novatchkov et To illustrate the potential of Artificial Intelligence 7 Female, 8 Male/ 20-31 Artificial intelligence can evaluate weight training equipment performance and
al. [8] (AI) techniques in sports years provide athletes with quick feedback.
The simulation results of Weka software indicate that the classifier can accu-
To enhance the adaptability control of exoskel-
rately assess lower limb motion using gait analysis data. Furthermore, cluster
Zhao et al. [38] etons, leading to successful human-computer …
analysis can group lower limb motion for various users in different environ-
coupling.
ments and at different times.
…
Gau [44] ods in generating robot walking patterns. specific clearance, with a desired step length, and walking at a particular speed.
To utilize artificial intelligence techniques to elevate Establishment of an expert system catering to a precise and distinct sports tech-
Lapham et al. [1] the role of computers in decision-making processes … nique would present a noteworthy progression in the field of sports biomechan-
within the discipline. ics.
390
Artificial Intelligence in Sports Biomechanics
concurrently eliminating the requirement of an Lu et al. [35] a light CNN architecture is em-
all-inclusive communication level that consol- ployed to detect football and basketball play-
idates all functionalities into a solitary mod- ers in Television camera footage. RGB im-
ule, thereby expediting access to data [52]. ages are employed in the process of training
Some authors used ANN for the study of gait and testing the Convolutional Neural Network
and sport performance [8, 13, 28, 31, 34]. ANNs model. Lehuger et al. [46] announced that it
facilitate the learning process of computers is recommended to employ two convolutional
by means of experience and analogy. These layers and Grayscale images in order to de-
computer programs endeavor to construct a tect football players via a television camera.
mathematical framework of the neurons pres- In another comprehensive study, Acuna [33],
ent in the human brain. An ANN refers to a announced especially focuses on a perfor-
complex network of simple and adaptable pro- mance RGB image-based system and You
cessing elements, commonly known as nodes Only Look Once (YOLO) network for real-
[31]. The CNN is predominantly employed time detection of basketball players. Pobar
for the examination of visual imagery [53]. and Ivasic-Kos [37] showed that the YOLO
Specialized in the processing of pixel data, was used for optical flow to detect active play-
these technologically advanced systems are ers during handball matches. Moreover, Buric
specifically intended for employment in im- et al. [32] evaluated the performance of CNNs
age recognition and processing. Dorschky et to detect handball players. Sah et al. [26] also
al. [45] improved a measured inertial sensor evaluated the efficacy of recent and estab-
dataset by adding simulated data to train con- lished techniques for detecting players in field
volutional neural networks for estimating joint sports. The proposed CNN structures aimed
angles, moments, and ground reaction forces to evaluate the performance of the player uti-
during walking and running. The addition of lizing various images. According to the find-
simulated data reduced the root mean square ings, a CNN structure, which is employed in
error of the test set for joint angles and mo- tandem with image representations, attains the
ments by up to 27% and 6%, respectively, and most optimal outcomes. Hao Ren et al. [27]
for Ground Reaction Forces (GRFs) by up developed a system for detecting sports video
to 6%. However, the accuracy of the biome- athletes using deep learning for experimental
chanical model limited the simulation-aided testing. Additionally, the accuracy of detec-
estimation of joint moments and GRFs. In an- tion improved by nearly 10%, as compared to
other study, the estimation of lower limb joint conventional convolution network recognition
angles and moments was analyzed using a algorithms, which underscores the recognition
feedforward neural network with Inertial Mea- advantages of the system.
surement Unit (IMU) data. The dataset used Based on the ANN and CNN have shown
for this purpose includes both optical motion that the proposed approach can also evaluate
capture data from previous research and newly the accuracy of parameters, predict perfor-
collected IMU data. The addition of synthe- mance, and perform clustering.
sized data significantly improved joint angle
prediction, while the inclusion of noise led to Supervised learning
improved prediction accuracy, indicating the Supervised learning is considered a goal-
potential benefits of augmentation techniques oriented methodology, presenting the in-
in the context of biomechanical data for put and the desired output. A mathematical
machine learning applications [54]. framework is formed to depict the inputs in
CNNs have gained as a viable and feasible relation to the anticipated outcomes, leading
method to detect the athlete’s performance. to precisely classifying unobserved data and
effectively assess lower limb motion by between the actual and predicted outcomes
employing data derived from gait analysis. produced by all four NN models [41].
In the study conducted by Tou et al. [31], the
Performance-Prediction artificial neural network model offered a more
Athletic performance refers to the ability of accurate representation of the functional rela-
athletes to excel in sports, which is influenced tionship between exercise indices of a specific
by specific physical attributes that impact attribute in shot put athletes and their special-
their proficiency in various sports skills. The ized athletic performance. Furthermore, the
relationship between sports performance and model predicts the unique performance of
its indicators is intricate and nonlinear. Neu- shot-put athletes with greater precision. The
ral networking, a significant branch of mod- study also revealed that the specific physical
ern nonlinear science, has experienced rapid attribute of shot-put athletes plays a crucial
advancements and widespread application in role as a foundation for their specialized ath-
sports and gait analysis. Researchers have ex- letic ability.
amined the influence of artificial intelligence The results of the research by Silva et al.
on athletic performance and gait prediction, [41] demonstrated that the implementation
serving as the motivation for this study. Con- of a feed-forward neural network led to non-
versely, predicting the outcomes of sports linear analysis, which facilitated the creation
competitions poses a complex computational of four models for predicting performance.
challenge due to the multitude of uncertain Consequently, the utilization of neural net-
factors affecting the result of a match. Proba- work tools could affect addressing intricate
bilities are assigned to every conceivable state, issues, such as performance prediction model-
which the game could end, when predictions ing and talent identification in various sports,
about such outcomes that this task is generally most notably swimming. The manuscript
considered more difficult due to the increased authored by Eskofier et al. [7] reports on a
number of potential outcomes. study that outlines initial findings pertaining
In the study conducted by Novatchkov et to the establishment of a data-driven pattern
al. [8], similar patterns can be identified when recognition method that can successfully dis-
analyzing force characteristics, indicating that cern the key predictors of golf swing technique
supervised machine learning methods are suit- that are essential to enhancing performance.
able for classification purposes. This research In the study conducted by Schrapf et al. [28],
aimed to highlight the significance of AI ap- on 289 rallies of the 2nd Austrian Volleyball
proaches in sports, particularly in the context League women, the researchers assessed the
of resistance training. Additionally, the model- agreement between predicted and actual val-
ing results obtained in this study revealed fa- ues by measuring the proportion of accurate
vorable performance and accurate prediction predictions. The quality of predictions was
outcomes, showing the practicality and ef- evaluated by comparing the results produced
fectiveness of AI techniques in automatically by the ANN with those provided by expert
assessing athletic performance on resistance volleyball coaches. The ANN exhibited a pre-
training equipment and providing athletes diction accuracy of 68.1% for the designed
with timely guidance. area and 79.2% for the setting speed, which
In a study conducted on individuals engaged significantly surpassed random chance. Ad-
in swimming, the use of feed-forward NNs ditionally, the ANN outperformed the coaches
enabled the development of four models for in predicting target areas by 2.8% and showed
predicting performance. The findings of this a remarkable 14.6% improvement in predict-
investigation revealed a small mean difference ing setting speed compared to the coaches’
age-based sub-groups. The findings showed accuracy, heightened sensitivity and specific-
that forty principal components explained ity, as well as reduced training duration [11].
84.74% of the variability in the data. Addi- Silva et al. [41] showed that utilizing neu-
tionally, the SVM achieved an accuracy of ral network models is a viable approach to
86.34% in accurately distinguishing between address complex challenges, such as perfor-
female and male runners [36]. mance modeling and talent identification in
Tou et al. [31] indicated that the employment various athletic activities, with a specific focus
of a neural network prediction model results in on swimming.
significantly decreasing error between the cal- In the study conducted by Eskofier et al.
culated value and actual value of the special [7], classification experiments were employed
results, as compared to the utilization of the to evaluate the golf swing, and the findings
multiple linear regression models. This obser- could potentially enhance the understanding
vation underscores the enhanced capacity of of the key factors that influence the successful
the neural network model to fit the quality. execution of a golf swing.
Moreover, the fitting accuracy of the function- The findings of the research conducted by
al relationship is notably high between train- Tou et al. [31], demonstrated that employing
ing level and special performance. an artificial neural network model provides
According to examination and detailed anal- a better representation of the functional rela-
ysis of the diverse findings from the fields of tionship between training indices related to the
AI and ML, these techniques possess the capa- unique qualities of shot-put athletes and their
bility not only to predict performance but also specific athletic performance. Moreover, this
to assess the accuracy of parameters. advanced model exhibits a higher level of ac-
curacy in predicting the special performance
Sports Domain output exhibited by shot put athletes.
Sports has experienced significant growth The main objective of the study conducted
over the past two decades, leading to a piv- by Liu et al. [30] is to enhance professional
otal driver for numerous economies and a pro- sports training by leveraging digitalization and
found impact on our social and cultural fabric. intellectualization techniques. This is achieved
The complexities involved in predicting and through the implementation of motion capture
enhancing athletic performance present formi- technology based on AI and HCI, leading to
dable challenges that have traditionally been the creation of enriched application scenarios.
addressed by subject matter experts, such as The findings of this research indicate a signifi-
coaches, managers, scouts, and sports health cant improvement in the accuracy of pose res-
professionals using rudimentary analytical toration among athletes during sports training.
techniques. A comprehensive examination of the multi-
The results of the study by Novatchkov et tude of factors contributing to athletic perfor-
al. [8] revealed that AI methods possess a mance holds significant potential for the ben-
high degree of effectiveness in the automat- efit of coaches, researchers, and athletes. By
ic evaluation of weight training equipment thoroughly exploring the intricacies and nu-
performance and the provision of timely guid- ances of these variables, a more comprehen-
ance to athletes. sive understanding of their impact on athletic
Experiments with diverse training and outcomes can be achieved. This enhanced
test datasets have demonstrated that utiliz- understanding can then inform and guide the
ing curve-fitted data can result in enhanced development of more effective training pro-
performance metrics for athletes. These im- grams, performance assessments, and over-
provements encompass increased predictive all athletic strategies. Therefore, investing in
the analysis of performance prediction and in gait data. This approach addresses the limi-
the accuracy of parameters can yield substan- tations of manual engineering. Although the
tial dividends in terms of athletic success and available dataset encompasses a wide range of
improvement. walking environments, it remains insufficient
for reliable gait analysis [17].
Gait domain The examination of human walking patterns,
Research pertaining to the locomotion of commonly known as gait studies, has made
humans is referred to as gait analysis. The noteworthy contributions to the development
methodology of gait analysis involves the of prosthetic limbs and orthotic devices for
quantification of elements that govern the func- amputees. Moreover, this research has served
tionality of the lower extremities, thereby ex- as a source of inspiration for the creation of
posing the underlying mechanisms of human artificial locomotor controllers, which are
movement. The study and analysis of human utilized in exoskeletons and robotics. Juang
locomotion, commonly known as gait analy- [44] has suggested a model for synthesizing
sis, has a multitude of applications in various the gait of robots by employing kinematics
fields such as medical diagnosis, safety, vir-
principles and utilizing the fuzzy neural net-
tual reality, trade, and physical activity knowl-
work. Kim and Lee [40] used a Hybrid genetic
edge [15, 63]. Dysfunctional ambulation may
algorithm and neural network to determine
manifest as a result of either acute or chronic
joint angles according to the accelerator
injury, or due to inappropriate biomechanics.
data. Analysis of gait is a fundamental aspect
Healthcare professionals can derive signifi-
of studying the natural walking patterns of
cant assistance from different research types
human-like robots. Various algorithms that
in analyses of gait variability, utilizing kine-
utilize artificial intelligence techniques have
matics and kinetic parameters, both in predict-
ing the onset of a situation and monitoring the been developed to facilitate this process
recovery profile of patients in clinical fields [64-66]. Heinen and Osório et al. [67] have
[17]. A significant cohort of scholars, who are explained a scientific method to optimize
captivated by this field, have suggested inno- the gait of humanoid robots. The approach is
vative methodologies. This particular group designed to create and control consistent
concentrates on the scrutiny of walking pat- gaits for legged robots within a physically
terns with the aim of deliberating on the utili- based simulation environment. Wawrzy ´nski
zation of machine learning approaches and the [68] delves into the domain of enhancing the
influence it has had on the study of gait. walking pattern of humanoid robots through
It has been determined that, presently, the the application of reinforcement learning with
most extensively employed and compre- experience. Rozumalski et al. [39] in crouch
hended methods of analysis for gait involve gait, which can help guide treatment decisions
computational techniques. The data pertaining and outcomes assessment.
to gait is characterized by marked heterogene- The incorporation of novel methods has
ity, high dimensionality, temporal dependence, provided clinicians with the opportunity to
and variability. Proceeding with this data utilize instrumental gait analysis as a stan-
presents a challenge. Moreover, conventional dard clinical practice in order to evaluate
computational techniques are limited by their the condition, recuperation, and progress
inability to process natural gait data in their raw of patients with intricate neurological and
form. However, deep learning architectures musculoskeletal disorders, ultimately mitigat-
offer a wide range of opportunities for identi- ing the potential propagation of injury through
fying, classifying, and detecting abnormalities gait analysis.