Quantification of Visual Search Motion During Basketball

in VR Simulation

Shinya Ishikawa

, Hidehiko Shishido

2,* a

, Kenji Yoshida

and Yoshinari Kameda

Graduate School of Systems and Information Engineering, University of Tsukuba, Japan

Center for Computaitonal Sciences, University of Tsukuba, Japan

Institute of Health and Sports Sciences, University of Tsukuba, Japan

Keywords: Vision Search, Behaviour Analysis, Gaze Detection, Head Mounted Display.

Abstract: In basketball, players must be able to make good decisions in complex and rapidly changing situations. An

important element of situational judgment is visual search motion. We propose to quantify the visual search

motion of a basketball player during play. In the proposed method, the experience of playing basketball is

realized on a VR simulation. The player wears a head-mounted display to experience the playing situation.

We prepared scenarios that considered the characteristics of the basketball game and the characteristics of the

head-mounted display and measured the visual search motion on the prototype system. Experiments were

conducted on defensive players, and the results show that the measurement results of visual search motion are

useful for discussing the characteristics of players.

1 INTRODUCTION

In basketball, players are required to make accurate

judgments in response to complex and rapidly chang-

ing surroundings (Bjurwill, 1993)

(Albernethy,

1993). One of the most important elements for mak-

ing judgments is the visual search movement.

Visual search movements are performed when it

is necessary to grasp the positions of multiple objects,

such as a ball or other players. If visual search move-

ments can be quantified, it will be possible to discuss

whether or not the players acted appropriately to the

situation together with their movements based on vis-

ual search movements.

In this study, we propose to quantify the visual

search motion of basketball players during play.

For quantification of visual search motion, it is es-

sential to obtain the following information. The basic

information is the position and posture of the athlete's

head. Information on the direction of the player's line

of sight and the object he/she is gazing at are also nec-

essary. In order to obtain this information in a real en-

vironment, it is necessary to know all the circum-

stances surrounding the players.

https://orcid.org/0000-0001-8575-0617

https://orcid.org/0000-0001-6776-1267

Now at Soka University, Japan

The proposed method realizes the experience of

playing basketball on a VR simulation. The VR sim-

ulation allows us to manage all the situations sur-

rounding the players. The player wears a head-

mounted display (HMD) to experience the playing sit-

uation. By using an HMD that can measure the posi-

tion and posture of the player's head and the direction

of the player's line of sight, the system can obtain all

information during the play.

The experience of playing on a VR simulation

with an HMD cannot match the experience of playing

in reality. Therefore, it is necessary to carefully con-

sider the scenario to be used when measuring visual

search motion, taking into account the characteristics

of the basketball game and the field of view of the

HMD. For this purpose, we will develop scenarios

with the advice of an experienced head coach of a bas-

ketball team.

By quantifying visual search movements, it is pos-

sible to investigate the characteristics of players in a

playing situation. As an example, in a scenario where

defensive movements are required, we calculate bas-

ketball-specific motions from the results of visual

search motion measurements. From the results, we

Ishikawa, S., Shishido, H., Yoshida, K. and Kameda, Y.

Quantiﬁcation of Visual Search Motion During Basketball in VR Simulation.

DOI: 10.5220/0013018600003828

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 12th International Conference on Sport Sciences Research and Technology Support (icSPORTS 2024), pages 225-232

ISBN: 978-989-758-719-1; ISSN: 2184-3201

225

show that it is possible to discuss the characteristics

of the players.

2 RELATED WORKS

In an effort to quantify the visual search motion of

basketball players in a real environment, measure-

ment of rotational head and eye movements using a

gyro sensor and an IR camera has already been pro-

posed

(Vickers, 2019) (Maarseveen, 2017). They

quantified the differences in visual search motion for

each selection when offensive players wearing eye-

tracking sensors were in possession of the ball in

Pick-and-Roll. In other sports, for example, in soccer,

Aksum et al. (Aksum, 2021) quantified the visual

search motion for each Ball Action of players wearing

eye tracking sensors in soccer.

In such efforts in real environments, it is difficult

to analyze the relationship between visual search

movements and the surrounding situation because the

surrounding situation is sometimes unknown.

As an approach to visual search motion using vir-

tual environments, Ferrer et al. proposed a method to

quantify visual search motion for passing in soccer

games (Ferrer, 2020). In this method, a playing situa-

tion is prepared on a VR simulation and the direction

of the head of a player wearing an HMD is measured.

Because of the short time between the visual search

motion and the passing action, the analysis of the vis-

ual search motion has not been addressed. (Wood,

2020) reported the effect of practice with visual

search motion on a VR simulation.

3 QUANTIFICATION OF VISUAL

SEARCH MOTION

In this study, we quantify the visual search motion of

basketball players during play. The following infor-

mation is used in the quantification of visual search

motion.

The basic information is the position and posture

of the player's head. This information is expressed in

the court coordinate system. Next is the direction of

the player's line of sight. This is represented by the

average of the direction vectors of both eyes starting

from the average of the eye positions of both eyes.

If there is a ball or another player at the gazing

point, it becomes the gazing object. The walls and

floors of the building where the basketball game is

played are also considered as gazing objects. If the

gaze vectors of the two eyes are nearly parallel, the

object whose gaze vectors collide first is considered

to be the gazing object.

In order to obtain these values, we use an HMD

that is capable of six-degree-of-freedom position and

posture estimation. Furthermore, by using an HMD

with an eye tracking function, we obtain the eye vec-

tors of the left and right eyeballs.

For the VR simulation, an international standard

basketball court is set up in the virtual space, and the

ball and other players necessary for the game are

placed in the court. Players wear HMDs to experience

the VR simulation. In order to ensure a sense of im-

mersion, a 3D model of the gymnasium is prepared

that resembles a real gymnasium.

In the VR simulation, labels are assigned to the

CG model to obtain the labels of objects that collide

with the eye vectors. Players and balls other than the

person experiencing the simulation are made to move

according to the simulation program.

4 SCENARIO STUDY

4.1 Prerequisite

In order to quantify visual search motion, players

wear HMDs while playing basketball.

Two limitations of wearing HMDs are listed be-

low. First, the HMD with the eye tracking function

must be firmly fixed on the head. Second, the field of

view of the HMD is narrower than the normal field of

view. Therefore, it is necessary to consider the influ-

ence of the difference in the field of view on the visual

search motion.

Under these limitations, the player's experience

should be a meaningful basketball play.

Based on the above, and based on the opinions of

experienced basketball players and coaches, this study

focuses on a scenario in which the players defend a

goal. In goal defense, the defending player observes the

movements of the attacking player through visual

search movements and acts accordingly. In this type of

action, there is almost no movement such as jumping,

so the possibility of HMD misalignment is low. The

difference in the field of vision can be mitigated by ap-

propriately positioning the initial position of the de-

fending players in relation to the attacking players.

4.2 Scenario

Taking advice from Kenji Yoshida, head coach of the

University of Tsukuba men's basketball team, we will

prepare three different scenarios.

icSPORTS 2024 - 12th International Conference on Sport Sciences Research and Technology Support

226

The common assumption for all three scenarios is

that there are three players, two offensive players and

one defensive player, as shown in Figure 1. The par-

ticipant in the experiment participates as a defensive

player. At the beginning of the play, the attacking

player who has the ball and is on the right side of the

top, as seen from the direction with his back to the

goal, is the ball man. The other attacking player on the

left wing is the marksman. The arrangement of the

two attackers and the defender is the same at the be-

ginning of every scenario. It is difficult for the de-

fender to see the two attackers at the same time.

In all scenarios, the defensive player should be

close to the ball man at the start of the game in order

to block a shot by the attacking player. The defender

may try to anticipate and cut off the pass. In this case,

the movement is a denial of the ball man reaching for

the path of the pass from the ball man to the marks-

man in the initial position. On the other hand, since

the two attackers are standing far from each other, the

defender needs either a visual search motion by shak-

ing his head from side to side or a visual improvement

motion by trying to get them in his field of vision by

moving back toward the goal.

Three scenarios for the attacker to score are de-

scribed below. A scenario is divided into events along

the time axis. The criteria for the division will also be

described.

4.2.1 Scenario A

The first is scenario a, in which the ball man moves

outside and shoots. Scenario a can be subdivided into

three events. The first event is event 1, when the

marksman moves outside the boundary of the field of

view along the 3-point line. The next event is event 2

when the ball man passes to the marksman. Finally,

the marksman shoots, which is the final Event 3. Fig-

ure 1 shows the scene at the end of Event 3. In each

event, the defender is considered to either maintain

his initial position or retreat toward the goal. There-

fore, for each of the three events, there are two possi-

ble outcomes: either the defender stays in front of the

goal without retreating, or he retreated. As a result,

Figure 1: Scenario a.

there are six types of events. We label these events a1-

front, a1-back, a2-front, a2-back, a3-front, a3-back.

4.2.2 Scenario B

The second is scenario b, in which the marksman

shoots on the spot. Scenario b can be subdivided into

two events. The first event is that the ball man passes

to the marksman while the marksman is stationary on

the spot. The second event is when the marksman

shoots. Figure 2 shows the scene at the end of Event

2. As in scenario a, each event differs depending on

whether the defender retreats to the goal or not. As a

result, there are four different events. As in scenario

a, we label the events.

Figure 2: Scenario b.

4.2.3 Scenario C

The third is scenario c in which the ball man dribbles

and shoots. Scenario c can be subdivided into three

events. The first event is until the marksman begins to

move. The next event is event 2, where the marksman

starts to move and the ball man starts to drive. The last

event, event 3, is until the marksman drives and

shoots. Figure 3 shows the situation at the end of

Event 3. If we take into account whether the defender

retreats or not, we end up with six different events.

The labeling is the same as in scenario a.

Figure 3: Scenario c.

Quantiﬁcation of Visual Search Motion During Basketball in VR Simulation

227

5 ACTION INDEX

This section describes a method for obtaining the in-

dices necessary to observe the visual search move-

ments of the defensive players in the scenarios de-

scribed in the previous section.

5.1 Event Switch

The factors of event switching can be divided into

those based on the motion of the ball and the attacking

players, and those based on the retreat of the defend-

ing players toward the goal. The former can be speci-

fied in time based on the description of the motion in

the VR simulation. For the latter, the head position of

the HMD is assumed to be the standing position of the

defender, and when the distance from the standing po-

sition to the goal (goal distance) becomes less than a

certain value, the player is judged to have retreated.

5.2 Head Shake

When the defender remains in the initial position and

continues to defend, he cannot see the two attackers

at the same time, and therefore, shaking his head to

the left and right is a visual search motion. We call

this action head shake.

We define the number of the head turns as follows

to indicate the amount of head shake. When the head

rotation angle returns to the original direction in a short

period of time, if the variation of the head rotation an-

gle exceeds a threshold value, one head turn is counted.

6 IMPLEMENTATION

To realize the proposed method, a VR simulation sys-

tem is constructed. In addition, a system for experi-

encing the scenarios will be constructed.

6.1 VR Simulation System

The VR simulation system is built using Unity, a

3DCG integrated development environment. For the

floor, ceiling, and four walls of the VR gymnasium,

we used the texture of a photo of the interior of a real

gymnasium used by the participants in the experiment

to provide a sense of immersion. The attacking player

is assumed to be 170 cm tall, assuming the height of

an average Japanese male. The size of the ball and the

court are based on the standards of the International

Basketball Federation (FIBA, 2024). An overview of

the VR gymnasium from a subject viewpoint in the

simulation system is shown in Figure 4.

We will explain the mechanism of the movement

required for the visual search exercise. We used Very

Animation, a Unity Asset that allows the creation and

editing of animations of objects. Several typical poses

are prepared for the attacker's movements. A Unity

feature called "Animation Blend" was used to com-

plement the movements between the poses to repro-

duce the real game behavior. Timeline, a Unity fea-

ture that allows multiple objects to be controlled sim-

ultaneously, was used to accurately describe the lay-

out of the positions of multiple players and the ball

and their movements with respect to time.

Figure 4: A view of VR gymnasium from a subject viewpoint.

6.2 User Experience System

VIVE PRO EYE from VIVE Technologies, Inc. will

be used as the HMD. Four base stations are used to

estimate the position and orientation of the HMD. The

wireless function is used to facilitate the movement of

the participants in the experiment. This ensures that a

10m square area is available for the participants to

move around. In order to provide an immersive expe-

rience for the participants, the system is set up in the

gymnasium where the participants usually practice.

Figure 5 shows an overview of the experience system.

Figure 5: Overview of the proposed experience system in

the university gymnasium at University of Tsukuba.

icSPORTS 2024 - 12th International Conference on Sport Sciences Research and Technology Support

228

6.3 Measurement

The accuracy of the measurement of the motion of the

participants in the experiment is determined by the ac-

curacy of the VIVE PRO EYE. The position and ori-

entation of the HMD are measured at 70 Hz or higher,

but the data are recorded at 30 Hz for synchronization.

The horizontal position of the HMD can be measured

with an accuracy of 1.8 × 10



𝑚 (Holzwarth,

2021). The direction vector of the HMD can be meas-

ured with an accuracy of 0.8° (Bauer, 2021). The eye

detection in the HMD can be measured at 120Hz, but

is recorded at 30Hz for data synchronization. The data

is recorded at 30Hz. The eye vectors can be measured

with an average accuracy of 6.21° (SD 0.77°) (Si-

patchin, 2021).

As for the gazing objects, the eye vectors and col-

lision judgments are performed for the floor, ceiling,

and four walls of the VR gymnasium, as well as for

the ball and the attacking player, and the target labels

and the time when the collision occurred are recorded.

For the attacking player, collision judgments are

made separately for each joint, so it is possible to ex-

amine which part of the opponent's body the player

was looking at.

Next, we will explain how to prepare the partici-

pants for safe defensive actions. In the experiment, the

movable area is 5.45m in depth and 7.00m in width.

The boundaries of the movable area are displayed dur-

ing the experience in a way that does not spoil the

sense of realism.

Next, the experimental subject is given time to

practice until he/she can perform the desired defen-

sive action in the actionable area of the VR gymna-

sium. During the practice, the players of the prepared

scenario are reversed in the center of the court. During

the practice time, the two attacking players exchange

passes without moving from their initial positions.

The experimental subjects are asked to "follow the of-

fensive player on the left one-on-one. Practice to

move smoothly to position 1 or 2, etc." They are told:

"Please assume that the offensive player is aiming at

the goal as in a game. Position 1 is the position where,

when the marksman has the ball, the defenseman is in

a straight line between the marksman and the goal and

can immediately go for a block if the marksman

moves to shoot. . The experimental subjects continued

to practice until they self-reported that they were able

to perform the defensive action within the area of pos-

sible movement. Thereafter, six trials are conducted.

7 EXPERIMENTS

7.1 Procedure

First, explain that there are two attacking players in

the experiment and that the goal of the experiment is

for the participant to interfere with the attacking shot

as a defensive player. Explain that there will be six

trials. We do not tell them how many different offen-

sive scenarios the attackers have prepared. The exper-

imenter is allowed two trials for each of the three sce-

narios so that the number of trials is not biased toward

any of the three scenarios. The order in which the sce-

narios are presented is determined for each participant

so as to counteract the order effect.

Next, we calibrate the eye gaze measurement for

each participant in the experiment. In order to check

the accuracy of the eye tracking, the participants are

asked to look at a basketball in the VR gymnasium

and erase it. The participant's gazing point is indicated

by a pink sphere. When the sphere collides with the

ball, the ball disappears.

7.2 Execution

Ten members of the University of Tsukuba men's bas-

ketball team, ranging in age from 19 to 21, partici-

pated in the experiment as participants A through J.

The experience system was set up in the gymnasium

used by the basketball team for practice (Figure 5).

The time required for the trials of scenarios a, b, and

c was 8.20, 4.90, and 8.30 seconds, respectively. In all

scenarios, the initial positions of the two attackers and

the defender are the same, and the attacker stands

93.2° open from the defender.

Based on the positioning of the players in the pre-

pared scenarios, we set the threshold value of the piv-

oting judgment to 20°. When the goal distance of the

participant in the experiment was less than 4.8m, the

participant in the experiment was considered to have

retreated, and an event switch was considered to have

occurred. Figure 6 shows the measurement of the

head rotation angle for experimental participant A in

scenario c. In this example, the number of pivots is 6.

Figure 7 shows the goal distance for participant E in

scenario b.

The measurement record of each trial can be visu-

alized. Figure 8 shows the scene. The position of the

participant's head is indicated by the gray dots in the

figure. The blue line indicates the frontal direction of

the head, and the white and green lines indicate the

edges of the HMD's field of view. The red line indi-

cates the direction vector of the gaze, and the light

blue dots indicate the gazing position. As shown in

Quantiﬁcation of Visual Search Motion During Basketball in VR Simulation

229

the subfigures, the gaze estimation is precise enough

to examine the objects that the subject checked.

Figure 6: Head rotation angle for experimental participant

A in scenario c.

Figure 7: Goal distance in scenario b for experimental par-

ticipant E.

7.3 Discussion

The characteristics of players are discussed based on

the results of visual search movement measurements.

The co-author of the discussion, Kenji Yoshida, is the

head coach of the University of Tsukuba basketball

team.

7.3.1 Adaptation to Multiple Scenarios

In terms of adaptation to the multiple scenarios in this

study, focusing on a single participant, there are two

possible directions: either the participant stays in the

initial position and increases the number of pivots as

the trial progresses, or the participant moves back to-

ward the goal and performs pivots. The order of sce-

narios is scenario c, scenario a, and scenario b, in or-

der of the number of pivots that should occur.

In Figure 9, we can see that participant B in the

first trial had three shakes, but in the sixth trial, he had

six shakes. It can be said that the participants did not

know what kind of situation would come in the first

trial, but as they repeated the trials, they became

aware that there were multiple scenarios, and this re-

inforced the visual search movement.

In Figure 10, we observe that participant E was in

the backward state for 2 out of 4 pivots in the first trial,

but all 8 pivots were in the backward state in the sixth

trial. It can be said that the subjects realized that there

were multiple scenarios as they continued the trials,

and performed the visual improvement exercise of re-

treating.

These are not defensive movements that should be

judged as correct. They should be regarded as the

manifestation of the characteristics of each player.

7.3.2 Unique Defence Motion

(a) At the beginning.

(b) Checking the left target.

Figure 8: Measurement visualization of visual search mo-

tion. Th time flows from (a) to (c). Note that the light blue

balls (gazing object) are placed properly.

icSPORTS 2024 - 12th International Conference on Sport Sciences Research and Technology Support

230

Looking at the cross-sectional results of the experi-

ment, we can find players who perform characteristic

defensive actions. The number of pivoting move-

ments for all 10 participants for scenario a is shown

in Figure 11. Participant J performs the pivoting with-

out retreating. From this, the head coach states that we

can read that this participant is faithfully performing

the deny, the action that prevents the path of the ball.

Figure 9: Number of head shakes per scenario for Partici-

pant B.

Figure 10: Number of head shakes per scenario for partici-

pant E.

Figure 11: Distribution of head shake frequency in scenario a.

8 CONCLUSIONS

We proposed a method for quantifying the visual

search motion of players during the game. We showed

how to construct a VR simulation system that can

manage all the situations surrounding the players and

their experience system, so that the visual search mo-

tion can be measured.

By preparing a scenario that takes into account the

characteristics of the basketball game and the charac-

teristics of the HMD, we showed that the measure-

ment of visual search motion is effective even on a

VR system.

As an index to investigate the characteristics of

visual search motion, the number of left and right piv-

ots at each segment in the scenario was obtained. We

conducted an experiment with members of the men's

basketball team of the University of Tsukuba, and

were able to discuss the characteristics of the relation-

ship between the players' visual search movements

and their defensive actions based on the results of the

measurement of visual search movements.

This study was partially supported by JSPS KA-

KENHI 23K21685/21H03476 and it was originated

by (Ishikawa, 2022).

REFERENCES

Bjurwill, C. (1993). “Read and React: The Football For-

mula,” Perceptual & Motor Skills, vol.76, no.3,

pp.1383-1386.

Albernethy, B., Thomas, K. T., Thomas, J. R. (1993). “Strat-

egies for Improving Understanding of Motor Expertise

(or Mis-takes We Have Made And Things We Have

Learned!),” Advances in Psychology, vol.102, pp.317-

358, 1993.

Vickers, J. (2019). “The role of quiet eye timing and loca-

tion in the basketball three-point shot: A new research

paradigm,” Frontiers in Psychology, vol.10, pp. 1-16.

Maarseveen, M. (2017). “In situ examination of decision-

making skills and gaze behaviour of basketball players,”

Human Movement Science, vol.57, pp.205-216.

Aksum, K. M., Brotangen, L., Bjørndal, C. T., Lukas M.,

Geir, J. (2021). “Scanning activity of elite football play-

ers in 11 vs. 11 match play: An eye-tracking analysis on

the duration and visual information of scanning,” Plos

one, vol 16.

Ferrer, C. D. R., Shishido, H., Kitahara, I. and Kameda, Y.

(2020). “Read-the-game: System for skill-based visual

exploratory activity assessment with a full body virtual

reality soccer simulation,” PloS one, vol 15(3).

Wood, G., Wright, D. J., Harris D., Pal A., Franklin Z. C.,

Vine S. J. (2021). “Testing the construct validity of a

soccer-specific virtual reality simulator using novice,

Quantiﬁcation of Visual Search Motion During Basketball in VR Simulation

231

academy, and professional soccer players,” Virtual Re-

ality, vol 25(1), pp.43-51.

International Basketball Federation (FIBA). (2024). “2024

Official basketball rules; basket rules & basketball

equipment,” p.12.

Holzwarth, V., Gisler, J., Hirt, C., Kunz, A. (2021). “Com-

paring the accuracy and precision of SteamVR tracking

2.0 and oculus quest 2 in a room scale setup,” ACM In-

ternational Conference Proceeding Series, pp.42-46.

Bauer, P., Lienhart, W., Jost, S. (2021). “Accuracy investi-

gation of the pose determination of a VR system,” Sen-

sors, vol 21(5), pp.1-17.

Sipatchin, A., Wahl, S., Rifai, K. (2021). “Accuracy and

precision of the HTC VIVE PRO eye tracking in head-

restrained and head-free conditions,” Investigative

Ophthalmology & Visual Science, vol 61.

Ishikawa, S., Shishido, H., Yoshida, K., Kameda, Y. (2022).

“Quantification of visual exploration activities on expe-

riencing the basketball game using VR simulation,”

IEICE Tech. Rep., vol. 121, no. 423, MVE2021-93, pp.

284-289.

icSPORTS 2024 - 12th International Conference on Sport Sciences Research and Technology Support

232