CC BY
1DIGITAL TRANSFORMATION IN SPORT
The effectiveness of the training process of skiers using computer vision methods
UDC 796.012.37
PhD N.B. Novikova1 A.N. Novikov2 I.G. Ivanova1 A.N. Belyova1
1Saint-Petersburg scientific-research institute for physical culture, St. Petersburg 2ITMO University, St. Petersburg
Corresponding author: [email protected]
Received by the editorial office on 01.07.2024
Abstract
Objective of the study was to experimentally validate the use of computer vision techniques to enhance the efficiency of ski racing training management.
Methods and structure of the study. The examination of video footage of athletes gliding on roller skis at their top speed was conducted using a specially designed software that incorporates a motion-detection system for athletes, powered by the Alpha Pose neural network.
Results and conclusions. The advanced software enables precise identification of ski racers' movements in training and competition settings, providing visual representation of angular features and velocities in joints. Additionally, it can generate videograms automatically. The data obtained can be used to assess the effectiveness of skiing techniques and detect technical flaws that may go unnoticed using conventional methods.
Keywords: highly skilled cross-country skiers, technical training, technique of simultaneous one-step skating, neural network, computer vision methods, scientific and methodological support, video analysis of sports movements
Introduction. The performance of high-class cross-country skiers largely depends on the accuracy of training effects and timely control of physical and technical training. The main problems of pedagogical control in cross-country skiing are the lack of objective indicators of special training, due to the variability of the conditions of training and competitive activities and the lack of modern technologies and instrumental methods. One of the current areas of improving scientific and methodological support in cross-country skiing is the search for testing tools and methods that ensure rapid provision of objective and significant information to the coach, available for use in field conditions, not affecting the training process, fairly cheap and not requiring imported consumables. It was previously shown that the values of joint angles in key phases of the ski run depend on the effort exerted and change as a result of a controlled change in the intensity of
movement or an unintentional decrease in speed due to fatigue, which allows using these indicators to assess the level of training in races and intensive training [1]. A significant drawback of biomechani-cal research methods is the labor intensity of video recording processing, since even in advanced video analysis programs, annotation of key points and measurement of most indicators is done manually. Currently, solutions for markerless motion capture are being developed using complex models of the human body, computer vision and machine learning algorithms [2]. Most often, optoelectronic marker systems are used for motion analysis, but their use outdoors is limited due to the uncontrolled impact of environmental factors, the presence of other people or insufficient lighting. Publications by German scientists describe the experience of developing programs for recognizing ski jump videos [3], however, in cross-country skiing, the measurement of kine-
36
Theory and Practice of Physical Culture I teoriya.ru I October I № 10-2024
DIGITAL TRANSFORMATION IN SPORT
matic characteristics from videos made by 2D cameras outdoors has so far been performed manually.
Objective of the study was to experimentally validate the use of computer vision techniques to enhance the efficiency of ski racing training management.
Methods and structure of the study. All measurements were performed in a specially created computer program1, including a system for recognizing athletes' movements based on the Alpha Pose neural network, calculating the values of the required joint angles, and constructing graphs and cinograms. The video was filmed using a fixed Sony HDR-CX 730 EV video camera installed perpendicular to the direction of the athletes' movement.
The kinematic data were filtered using a smoothing function by averaging the values in a fixed-length window that moves through the data. The selection
of points for angle measurements was performed according to traditional positions [1].
Results of the study and discussion. The developed computer program calculated the values of joint angles in the ankle, knee and hip joints in a continuous cycle of simultaneous one-step skating stroke of athletes of the Russian national team when moving at maximum speed on roller skis. The acceleration time on a gentle slope was 18-20 s. The measurements of the joint angle values were visualized in the form of graphs, examples of which are presented in figure 1.
The start of measurements corresponds to the moment of placing the poles, the decrease in the curves corresponds to squatting, the first rise of the curves (increase in the angles in the knee and hip joints) corresponds to pushing off. The second rise of the lines corresponds to the extension of the
Athlete 1
Athlete 2
0 15Q 0
CT 0
"O 120' C0~
CT c <C
f
s
\ \ Y/
- ■-- --
-T- --- -r-r-1- - --
175 150 125 tOO 75 50
o'.o oj ù.2 û.i (M o's ai o'e o'n i"n i!i ô i*4 i!s
Athlete 3
Athlete 4
co seo
o
0
CT .g 12t)'
CO 0
CT 80'
c
<C
40-
0.0 0.1 0.2 0.3 0.4 0.5 OB 0.7 O.S 0.9 1.0 1.1 1.2 1.3
Time, s
iae
120
£0
40
0.0 0.1 0.2 0.3 0.4 0.5 O.G 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4
Time, s
= shin tilt — knee — hip
Figure 1. Examples of graphs of the dynamics of the angular characteristics of a simultaneous singlestep stroke
1Certificate of state registration of the computer program "Program for automatic determination of biomechanical characteristics of skiing technique based on 2D video recordings" (SPbNIIFK. SKI-COMPUTER VISION) No. 2024668312 dated 06.08.2024.
DIGITAL TRANSFORMATION IN SPORT
Athlete 1
Athlete 2
0-ú tú û!â ùa û's û'ê a'.? û.ë ù'è i !û 1.1
5.0 2.5
■2,5
m ô.i û.i o.a û.s û.s û.b a.? ô.a o.® 1.D 1.1 i.i i.a \a
Athlete 3
Athlete 4
5.0-
CO
T3 5
m
-t-» 0 0-
0
0
0 > -2,5-
cc
-5.0-
O)
r
<
q'o ai Q.2 cls va o"ä o> o!s û!& t!o i!n 1.2 Time, s
-3
m o.\ 0.2 o'à o 'A o"s o> o'e O!B IO Ti i"2 1.3 Time, s
■ shin tilt
knee
hip
Figure 2. Examples of graphs of the dynamics of angular velocities of movement in joints in a simultaneous single-step movement
legs before the next placing of the poles. The resulting graphs allow us to evaluate the consistency and amplitude of movements in the joints, compare the duration of squatting and pushing off.
For example, athlete 3 demonstrates the beginning of leg flexion before placing the poles on the support - an error that often appears when moving
at maximum speed, but is not detected when measuring angular characteristics only in key positions. The deepest squatting (decrease in the curve of the angle in the knee) is determined in athlete 1, however, the angle in the ankle joint increases (rise in the red line), which leads to a shift in the center of gravity back and can prevent the timely transfer of
Figure 3. Example of a videogram of a simultaneous single-step stroke, performed in automatic mode
Theory and Practice of Physical Culture I teoriya.ru I October I № 10-2024
DIGITAL TRANSFORMATION IN SPORT
body weight. The efficiency of ski racing technique is determined not so much by the magnitude as by the precision and timeliness of the applied efforts with maximum use of the repulsion inertia in the sliding step. It is impossible to evaluate the real dynamic characteristics of the repulsion without interfering with the training process, but measuring continuous indicators of angular velocities of movement in the joints can provide significant information on the distribution of efforts (Figure 2).
The simultaneous increase in the knee and hip joint extension velocity in athletes 2 and 4 apparently indicates a high push-off efficiency due to the coordinated work of large muscle groups. The increase in angular velocity during the push-off of athlete 3 is expressed to a lesser extent. Further studies are needed to analyze the association of the angular velocity gradient of the push-off with the athlete's forward velocity. The discrepancy in the hip and knee joint extension velocities during the swing movement also requires attention and, possibly, correction.
Consideration of the dynamics of angular characteristics and angular velocities simultaneously with the analysis of kinograms (Figure 3) can provide a complete picture of the technical features of each athlete.
Сonclusions. The developed computer program based on the application of the computer vision method allows with sufficient accuracy to recognize video recordings of cross-country skiers made in conditions of training and competitive activities, to visualize the dynamics of angular characteristics and angular velocities of movement in joints, and to build videograms in automatic mode. The obtained
data allow to evaluate the efficiency of skiing technique and to identify technical errors that are difficult to notice when using traditional methods of technique analysis. The advantages of the method include reduced time and increased accuracy of measurements, the ability to process large amounts of data and automatic visualization of results. There are some limitations in using the developed computer program, for example, increased requirements for server and video card performance, Linux OS, the inability to recognize the movements of skiers moving in parallel in one video sequence. Further research is needed to improve the methodology for assessing technical preparedness using the developed program.
References
1. Novikova N.B., Ivanova I.G., Belyova A.N. In-formativnost biomekhanicheskikh kriteriyev v otsenke sorevnovatelnoy effektivnosti lyzhnik-ov-gonshchikov vysokoy kvalifikatsii. Teoriya i praktika fizicheskoy kultury. 2024. No. 5. pp. 34-37.
2. Colyer S.L., Evans M., D.P. Cosker, Salo A. A Review of the Evolution of Vision-Based Motion Analysis and the Integration of Advanced Computer Vision Methods Towards Developing a Markerless System. Sports Med Open. 2018. Vol. 4. No 4 (1). 24 p.
3. Ludwig K. Lienhart R., Muller S., Kreibich S. Optimierung der voll automatischenzeit kontinuierlichen Erkennung der Korperpose und Skiposition von Skispringern in Videoaufnahmen. BISp-Jahrbuch Forschungs f rderung. 2021/22. pp.361-364.
