Developing the Thrower Detection System for Seamless

基礎論文 Paper

Developing the Thrower Detection System for Seamless

Player-Ball Interaction in Augmented Dodgeball

Kadri Rebane*1, David Hoernmark*1, Ryota Shijo*1,

Sho Sakurai*1, Koichi Hirota*1 and Takuya Nojima*1

Abstract --- Getting enough movement is crucial for people of any age group. In the modern world,

the lifestyle has become increasingly sedentary, so getting enough physical activity in a day relies

on the free will of people during their leisure time. Augmenting sports is a way to make physical

activities more exciting, fun, and enjoyable by virtue of computer game concepts. However, to

achieve these augmented sports, a system to detect events in the physical world and send them to

a computer game engine is required. Thus, the reliability of such systems has a significant impact

on the player experience. Especially in augmented sports with a ball, player-ball interaction is

critical for detecting physical events. This paper discusses the hardware design implementations

that are used in developing the augmented dodgeball thrower detection system. We discuss the

system’s design iterations from the prototype with a helmet and RFID tags to the current glove

system using a Hall sensor and magnets on the ball. We overview the overall mechanical design

considerations to make the system light and portable, details about calibrating, and sensor

placement. The described sensor calibration and placement improved the accuracy of detecting

the ball near the sensor and therefore improved recognizing the player-ball interaction event

during the game.

Keywords: Augmented Sports, Augmented Dodgeball, Player-ball Interaction

1 Introduction

Getting enough physical movement in a day is essential for

maintaining physical and mental health [1], [2]. However,

according to the World Health Organization (WHO) statistics [3],

many people among adolescents and adults do not get the

recommended amount of physical movement in a day. Sedentary

work is common [4], [5] and the growth of remotely accessible

services means that people do not need to move as much as they

used to in their daily life. This also means that physical exercise

has become a more precious activity done voluntarily during

leisure time. Sports participation depends on the perceived fun,

health benefits, and confidence in one’s skills [6], [7].

Augmented sports focus on existing activities or exertions and

enhance them with computer game concepts [8]. They are a

potential way to create novel and fun sports with sophisticated

rules and handicaps for players with different skills and gaming

expectations. This can be made by tracking the game actions

with technology and creating a game engine with all the desired

rules. The desired rules are the additions to the game that game

designers have created to realize the augmented game. As an

early stage of augmented sports, we mainly focus on augmenting

ball sports due to their variety and familiarity.

When developing augmented sports, the original rules of a

physical sport are modified to use virtual parameters, which are

used to balance physical ability among players. For example, in

traditional dodgeball player is out of the game when they get hit

by the ball. In augmented dodgeball, their virtual parameter

called “life points” is decreased. The amount of “life points”

decreased depends on the virtual parameter named “attack power”

assigned to the thrower and the virtual parameter called “defense

power” given to the player who got hit. This virtual parameter

method can empower a player when playing augmented sports,

even if the player is physically weak [9].

To work this method, the physical interaction between players

and the ball, which we call “player-ball interaction” in this paper,

must be provided to the game engine to update each player’s

virtual parameters for game matching.

1.1 Player-ball interaction

In augmented sports, it can be said that the game takes place

in two different dimensions, in the physical world and the virtual

world. Actions in the physical world should produce reactions in

the virtual world and vice versa. For example, when playing

augmented dodgeball [10], a player’s virtual life points will be

modified when the ball physically hits the player. Virtual life

points in augmented dodgeball are similar to “life points” or

“stamina” in computer games. The player can continue the game

*1 The University of Electro-Communications

TVRSJ Vol.26 No.2 pp.129-138, 2021

―129―

Transactions of the Virtual Reality Society of Japan Vol.26, No.2, 2021

when they have enough “stamina or “life-points”. In augmented

dodgeball, when the player loses all the virtual life points, the

player becomes an outfield player. In ball games, the player-ball

interaction is the center of attention and the source of actions. All

stakeholders of the game need to know the state of the ball during

playtime. In most ballgames, it is essential to know which player

is interacting with the ball at a particular moment and how and

where in the field is the ball.

These action/reaction pairs are expected to be logical, as

designed, explained by the rules, and almost immediate. The

virtual game engine, which is the essential part of augmented

sports that handles virtual parameters, must work seamlessly and

reliably for a good user experience. When considering this

feature, it is clear that the system to detect player-ball interaction

should have a considerable impact on playing augmented sports.

In this research, our goal is to develop a player-ball interaction

detection device capable of detecting which player, when, and

how is interacting with the ball. This paper describes two

different prototypes developed for playing augmented dodgeball,

capable of detecting a limited number of player-ball interactions.

Besides, we also describe the calibration method to increase

system reliability.

2 Related works

Augmenting ballgames and creating unique technology for

them has been done before. Most of these are trying to create

new game elements and/or reduce players’ skill gap(s). For

example, TAMA [11] is a ball that can change its trajectory using

injected gas from a gas tank placed inside the ball. Shepherd Pass

[12] is a ball-shaped quadcopter that can adjust its speed and

trajectory based on the player’s skill level. ACTUATE Racket

[13] can change the angle of a table tennis racket’s striking

surface. SomaticBall [14] can make the ball stick to the player’s

hand using a magnetic force. These projects created active

devices that can be used to introduce new game elements and

handicaps between players. However, the equipment design is

closely related to the game design and cannot be easily changed

or altered.

Augmenting something means making it greater in some way.

In games, diminishing technology has also been used to provide

unique and novel experiences. For example, D-Ball [15] uses a

head-mounted display to diminish all the environment except for

the markers on the ball and players. By doing this, passing the

ball between players changes significantly, creating a new ball-

catching game. Imaginary Reality basketball [16] uses a virtual

ball that is not visible to the players. The players are only

provided with some auditory feedback about the ball and should

watch how other players act, to understand where is the ball.

When augmenting games by uniting the virtual and the

physical world, it is essential to reflect actions and the

corresponding reactions in these worlds. Moreover, a monitoring

system is necessary. In projects like ShepherdPass [12],

Catching the Drone [17], and Sports Support System [18],

motion trackers are used to recognizing the players and ball(s),

so a unique environment and setup are needed for these systems

to work. For detecting the impact between the player and the ball,

sensors can also be used. For example, Piezo elements have been

used to calculate impact localization on table tennis rackets [19].

Gyro sensors attached to the players have been used to recognize

the type of beach volleyball serve [20]. In football, pressure

sensors integrated into the shoe have been used to detect and

analyze the interaction between the player’s foot and the ball

[21]. These systems are great for analyzing the kind of impact

the player and the ball had but are not suited to recognize any

longer holds of the ball. Dodgeball is one of the ball games

where holding the ball is permitted, and by knowing who is

holding the ball, the game can be augmented [10]. Now we

describe the solutions that have enabled us to detect the player

interacting with the ball.

3 Thrower detection system for augmented dodgeball

Augmented dodgeball [10] uses virtual parameters to balance

among players with different skill levels. To achieve this kind of

game, it is necessary to provide information such as who throws

the ball and who gets hit by the game engine. As a concept the

condition of the ball can be divided into three primary states

[22]:

(a) The ball gets closer to the player

(b) The ball is within the controllable range of the player

(c) The ball moves away from the player

The controllable range means that the player can change the

speed and/or direction of the ball. The actions that happen with

the ball during the time it is in the player-controllable range are

essential to game elements that should be forwarded to the game

engine. The game engine then keeps and updates the scores of

each player according to this information. For example, the

system could know who throws the ball from the combination of

(b) and (c). For the first stage of developing the thrower detection

systems, we focus on detecting (b) and (c) states.

3.1 Helmet system

The first-generation augmented dodgeball system consisted

of a helmet and a ball system for detecting the player who is

throwing the ball. When a player wants to throw the ball, they

must place it on their helmet just before throwing to register who

holds the ball. When the ball is correctly registered, a short alarm

sound is generated to let the player know that it registered

correctly. Adding this extra action, not present in the regular

―130―

Rebane ‧ Hoernmark ‧ Shijo ‧ Sakurai ‧ Hirota ‧ Nojima: Developing the Thrower Detection System for Seamless Player-Ball Interaction in Augmented Dodgeball

dodgeball game, slows down the game progress. This helps

players gain composure, especially making physically weak

players feel more relaxed during gameplay. At the same time, the

ball registration is an extra, unwanted action that interferes with

the smoothness of the game. Unfortunately, it is necessary due

to the system’s structure. Thus, to reduce the feeling of bothering,

we described this extra action as “giving special attack power to

the ball to damage the opposing team’s player”. The average

delay caused by the extra move was about 4 seconds for novice

players. The players were instructed to move the ball near the

helmet until they heard an audio signal indicating registration, so

timing the registration or passing the ball without registration

also became a tactic for the game.

Figure 1: Augmented dodgeball player holding the ball

The ball used in the game was a sponge ball (ɸ 160 mm). The

developed system was covered with 14 Radio-frequency

identification (RFID) tags (Figure 1, FeliCa by Sony [23]).

These tags enable contactless data transfer within a few

centimeters. The helmet worn by the player was equipped with

an RFID tag reader to compose a thrower registering system

(Figure 2). When the ball is placed near the tag reader for around

a second, the tag reader detects the tag’s existence, which means

the ball is near the helmet. This action of reading is then

processed by a microcontroller (Arduino UNO board [24]). Each

microcontroller is assigned a unique player identification

number (Player ID). This player ID is then sent to the game

engine to notify who holds the ball via the wireless module

(XBee [25]) on the helmet. The helmet is also equipped with a

small speaker unit that signals the player after the ball’s RFID

tag is registered. The whole system is powered with a 9V battery

and mounted into the box on top of the helmet. The system’s

game engine side was equipped with an XBee module to receive

the player ID. In the system, the last ID received is marking the

player who holds the ball. As a concept, this is the state we think

that the ball is in the controllable range of the player. In

dodgeball, the ball is thrown between players. So, we assume

that after registering the ball, the player throws the ball, making

it move away from it and there is a chance some other player gets

hit. To create an automatic augmented dodgeball system, this

information needs to be recorded and passed to the game engine.

The information flow of the thrower detection system is shown

in Figure 3.

Figure 2: Thrower registering system in the helmet

When using this system, the game engine can recognize who

holds the ball, identical to status (b). However, this system can

only detect action with intention: holding a ball in front of the

head. Therefore, sudden and instant action such as hitting the ball

or catching the ball cannot be detected.

Figure 3: Thrower detection device information flow

3.1.1 Discussions

This system consisted of a relatively simple circuit and the

development and building times were short. However, there

were some challenges with the system:

• Registering the ball correctly while playing the game was

difficult as it had to be placed very precisely near the reader.

The players could not see where the reader was, so sometimes

a fellow player would help them. Although this required more

collaboration than the traditional dodgeball game, sometimes

it slowed down the game too much.

• The box on top of the helmet was not secure enough when a

ball was thrown at it and could easily come loose. While the

equipment remained working in general, it needed to be

assembled, so the game had to be stopped for that time.

―131―


• The system took lots of time and effort to set up, such as

checking and changing batteries and setting player IDs when

the player needed the equipment changed.

Based on those findings, the second generation of the

augmented dodgeball system is developed.

3.2 Hall sensor and magnet system

The second prototype system is proposed based on the

findings through the first prototype. This system aims to detect

the ball being in the player’s controllable range without the

player having to make an additional movement for registering it.

The controllable range means that the player can interact with

the ball, change its movement direction and/or speed. The

information flow remains similar, but the detection is done via a

Hall sensor (Figure 4).

Figure 4: Information flow in Halls sensor and magnet system

More natural motion is deployed in this system to detect who

interacts with the ball by using a glove with a Hall sensor (Figure

5) and a magnet-embedded ball (Figure 6). In this system, the

Hall sensor monitors the magnetic field nearby. The sensor’s

output voltage is proportional to the applied magnetic field.

When the magnetic field value is over the set threshold value,

the system sends out a message with a plyer ID. When receiving

the message, the game engine assumes that the player ID

received represents the player who is holding the ball. When a

hit is registered in the system, the game engine assumes that the

last player holding the ball was the one who performed the hit.

This glove system makes reporting who interacts with the ball

seamless. The game engine can get that information without a

player making any extra movements, so it does not interfere with

the natural game flow. Each player wears the developed glove

only on their dominant hand.

Figure 5: Prototype of the Hall sensor and magnet system for

augmented dodgeball

In the magnetic ball, cylindrical neodymium magnets with a

diameter of 10 mm, a thickness of 2 mm, and a grade of N50 are

arranged on the surface dispersedly of a sponge ball having a

diameter of 170 mm. The sponge ball was chosen because it is

softer than regular dodgeball, even if magnets are on it. This

contributes to achieving a safer game [22].

Figure 6: Augmented dodgeball ball with magnets

The magnetic detection device is composed of a Hall element

and an Arduino Fio board [26]. The Hall element is Allegro

MicroSystems Phil Inc.’s A1324 LUA - T sensor. The resolution

is 1 G in the use environment, and the range can be measured up

to about ± 500 G. The threshold value for detecting the ball was

set to 10 Gauss.

The sensor position is set at the tip of a little finger. This is

because the little finger usually gets the smallest impact when

catching the ball, and therefore, it was thought to be the most

comfortable place for the player.

3.2.1 Catch and strike detection accuracy

We carried out a catch and strike detection experiment with 4

participants to prove the proposed system’s concept (Figure 7).

Striking the ball means that the player would only contact the

ball momentarily until it is bounced back from their hand.

Catching means that the player holds the ball and a throwing

motion occurs before the ball leaves the controllable range of the

player holding it. Each of the participants had to catch and strike

―132―


the ball 20 times wearing the system. For catching the ball, the

system could recognize it 100% of the time for all participants.

Although a small number of catch events were investigated, the

system shows high potential in detecting catching in this

experimental condition. This was achieved by creating a tense

enough magnetic field on the ball and assuring that the sensor is

in the necessary proximity of the magnetic field when the ball is

caught. For striking, the average detection rate was decreased to

90%. The event of striking the ball by hand happens rarely in

dodgeball. However, this strike-detecting attribute is essential to

play augmented sports. Therefore, the detection rate should be

improved.

Figure 7: Measuring the accuracy of the catch and strike

detection

3.3 The second generation of the Hall sensor and

magnet system

Figure 8: The second-generation Hall sensor system

Based on the findings of the first prototype system, we

primarily focus on shrinking the device’s size and improving the

detection rate’s accuracy. Figure 8 shows our second prototype

based on the concept. In the updated system, Arduino Fio was

replaced with an Arduino-compatible Feather Huzzah32 board

[27]. Due to this replacement and rearranging of other circuit

elements, the circuit board size became smaller from 72x47 mm

to 47x36 mm. The system worn by the player can be seen in

Figure 9. The wooden box is designed to mount the circuit of the

system to the player's wrist. The top hole is to access a DIP

switch to easily change the player ID associated with the device.

There is a hole on the front of the box to let the Hall sensor wires

be mounted to the glove. In the back, there is another hole for

charging the battery.

Figure 9: Updated Hall sensor system worn on the wrist

The glove used was changed to the stretchable knitted fabric

glove as seen in Figure 10. This was because it can fit well for a

wide size of different hands. The sensor was placed outside of

the glove to make putting on the glove more comfortable. As the

wrist device connected with the sensor is placed on the top side

of the wrist, the sensor wire is also guided there. The wire used

flexible cables and was stitched to the glove so that it would

allow some slack to fit players with different hand sizes. The

connection between the sensor and wire was covered with a hot-

melt adhesive to protect the soldering point from breaking.

Figure 10: Player wearing the glove with a sensor placed in the

palm.

―133―


To improve catch and strike detection accuracy, we

investigated causes that worsen the detection rates. When

considering the situation of detecting striking, it could have been

a temporal problem: a strike is a momentary event, and the

device detection speed might not have been fast enough to

register the event. Alternatively, the problem could also have

been positional: the participants were only instructed to strike

the ball. When striking with a palm, the little finger that had the

sensor attached may not have reached a value over the threshold

necessary to register the event. Our playtests suggested the Hall

sensor to be very sturdy for the hit caused by the ball we used.

We decided to change the sensor’s location from the tip of the

little finger to inside the palm, as indicated in Figure 10. This is

because the palm is where players usually contact the ball both

when holding and striking the ball.

3.3.1 Sensor evaluation

Figure 11: Ball divided into sections for testing

Then, we investigate the amplitude of the magnetic field of

the ball’s surface by using the hall sensor. This test is done to

choose the appropriate threshold value to detect catch and strike

events. We divided the ball into six segments to get the sensor

readings data and ran the sensor on the ball’s surface according

to the paths marked in red (Figure 11). The sensor was moved by

a human hand on the trajectories and was in contact with the ball.

For each path, 2000 data points were obtained. We also defined

the sensor faces during the test, as seen in Figure 12. The

experiments were conducted on two independent catch detection

devices referred to as Specimen 1 and Specimen 2 in our

experiment. The only difference between the devices’ setup was

that Specimen 2 had a bypass capacitor of 0.1 µF placed between

the sensor’s input power supply and ground. Specimen 1 did not

have the bypass capacitor installed. We found no significant

differences between the readings of the two different setups

during our experiments. The significance was tested with a two-

tailed paired t-test. The significant result is measured at the level

of p ≤ 0.05.

Figure 12: P-side of the sensor (left) and PF-side (right)

When moving the sensor on the ball’s surface and taking

readings of the Hall sensor without the Wi-Fi module enabled,

we found significant differences in detecting the magnetic field.

When the P-side of the sensor was facing the ball, the magnetic

field was recognized more often than when the PF side of the

sensor was facing the ball. Both Specimen 1 and Specimen 2

were tested with both setups of facing the p-side towards the ball

and facing the PF-side towards the surface of the ball.

The results of the experiment are seen in Figure 13. Green

columns represent the results obtained with Specimen 1 and

purple columns represent the results obtained with Specimen 2.

The horizontal axis shows which side of the sensor was facing

the ball and which path was tested. (P-12 means the P-side of the

sensor was facing the ball and the result is for path 1-2). We can

see that the side on which the sensor is placed on the ball matters,

and when using this specific sensor, directing the P-side towards

the object that we are trying to measure against gives a more

reliable reading. On average, 91% of the readings detected a

magnetic field over the threshold value when the p-side was

facing the ball. With the PF-side facing the ball, only 19% of the

readings were over the threshold value. The significance of the

sensor positioning was also confirmed with a two-tailed paired

t-test, where p ≤ 0.05.

Figure 13: Data for testing different sides of the sensor facing

the surface with magnets

―134―


3.3.2 Calibration

The second experiment to increase the system reliability was

made on the initial sensor calibration when turning on the device.

The first algorithm used for the sensor calibration was to read

the sensor 50 times with a 100 µs delay between the reads. The

average value of these readings was set as the baseline. The

threshold value was set to 10 Gauss above the baseline. The

catch detection worked well with this method, but sometimes

false positive catches were detected when playing augmented

dodgeball.

To track the false positive calls, we investigated the

following variables to see how they affect the calibration:

• Number of calibration iterations: 50, 250, 1000.

• Calibration delay (µs): 100, 10000, 50000.

• Number of loop iterations: 5000.

• Loop delay (µs): 100, 10000, 50000.

Calibration of the device is done once when the device is

turned on. The number of calibration iterations means how many

times is the sensor value measured during the initial calibration

step. Calibration delay is the amount of time between the

readings when calibrating the sensor. Loop iteration is the

number of times the sensor was read after calibrating to evaluate

the calibration effectiveness. Loop delay is the delay between

sensor readings after calibration to evaluate the sensor

calibration effectiveness. We focused on the minimum and

maximum measured magnetic flux density and the delta between

these two values during the calibration on idle settings without

any magnets present (Equation (1)). The lower the delta, the

better the result.

𝛿 = 𝐵𝑚𝑎𝑥 − 𝐵𝑚𝑖𝑛 (𝐺) , where (1)

δ – delta

Bmax – maximum recorded magnetic flux density

Bmin- Minimum magnetic flux density

G – Gauss

We found that the increase in the calibration iterations and an

increased delay positively affect the calibration. With this, the

minimum Gauss value becoming closer to zero. The maximum

Gauss-value increases with larger sample size. We can say that

the calibration is shifting to a more positive baseline. The raw

sensor data (vertical axis) readings can be seen in Figure 14. The

blue line represents the raw sensor data reading. The green circle

represents the area that was initially used to take the readings for

setting the baseline.

Figure 14: Raw data readings with 100 µs delay. The green

circle marks the place where calibration values were measured

using the initial calibration algorithm.

Another interesting thing to notice is that the readings at the

beginning of turning on the devices are lower than the ones

towards the end. We could determine that turning on the Wi-Fi

module produces noise and can influence the sensor’s readings

if the calibration is done without considering that noise. The

initial algorithm used only 50 readings with a 100 µs delay. This

means that the baseline was decided before the shift in the

baseline occurred. To account for the shift, the calibration

algorithm was changed to have a waiting period after starting the

device and take the readings after that. The results of applying

new parameters to calibration can be seen in Figure 15. The old

algorithm refers to the initial algorithm that was used. The new

algorithm refers to the proposed algorithm with a calibration

start delay, increased reading times, and increased delay between

readings.

Figure 15: Comparison between old and new calibration

algorithm

The specific changes that were made to the initial calibration

algorithm:

• calibration delay (time between reading the sensor when

calibrating) was changed (100 µs to 5000 µs),

• calibration iterations were increased (number of times the

sensor was read; from 50 to 1000)

• the readings during the first 200 ms were disregarded from

calculating the baseline value.

―135―


• loop delay (time between sensor readings after the calibration

is done) was changed from 100 µs to 5000 µs.

As can be seen in Figure 15, with the initial algorithm, the

minimum value measured by the sensor was -9.5 Gauss and the

maximum 0 Gauss with an average of -8.5 Gauss. The new

algorithm found a minimum value of -0.3 Gauss and a maximum

of 1.6 Gauss with an average reading of 0.5 Gauss. The delta of

minimum and maximum readings decreased from 9.5 Gauss (old

algorithm) to 2 Gauss (new algorithm). This means that the

calibration with the new algorithm is giving a steadier baseline

value.

These changes resulted in a calibration time of about 5

seconds after starting up the device. Additionally, it is essential

to keep the distance from any magnets during the calibration

time, affecting the baseline value. We still preserved the initial

10 Gauss threshold value.

3.3.3 Hall sensor system during gameplay

The developed system with the Hall sensor is an important

part of enabling the augmented dodgeball game. The system can

record automatically which player is holding the ball based on

the magnetic field. This information is then used to identify the

player who threw the ball that made the hit. We verified in our

first concept test that the Hall sensor can accurately detect the

occurrence of catching the ball with 100% accuracy in our test.

After improving the calibration algorithm, a steadier baseline

could be set for the system. In the gameplay, it meant that with

the old algorithm that had no steady base, sometimes constant

false-positive calls were made by some device. Figure 16

illustrates the case. This would mean that when entering a hit

occurrence to the system, the system would think that the player

wearing the device would be considered as the player who threw

the ball. When this episode happened, the device had to be reset.

As after resetting the calibration would take place again. As the

old algorithm used fewer readings to set the base, the set baseline

was sometimes steady enough, but sometimes not.

Figure 16: Device where the baseline had drifted, making

repeated false-positive calls to the database

With the new algorithm, we could not identify any false

positive calls to the database in an idle testing situation because

of having a steadier baseline. Figure 17 illustrates the readings

of a device in idle settings, calibrated with the new algorithm.

This would also mean that playing augmented dodgeball was

smoother and the game did not need to be stopped because of a

malfunctioning device.

Figure 17: Device calibrated with the new algorithm did not

result in false-positive calls to the database.

Not being able to register the ball when holding it was another

problem we sometimes faced when playing. After discovering

the difference between the p-side and pf-side facing the ball’s

surface, the detection rate also increased. This means that the

game engine had more precise data about who was the last

person holding the ball when a hit took place. This contributed

to an accurate point deduction from the person who got hit.

System reliability plays an important part in providing a good

playing experience for the players.

The devices with the modified sensor facing (P-side facing

outside) and the new algorithm were used in three public

demonstration sessions, each approximately three hours long.

Additionally, the setup was used in several private experiments

and demonstration sessions over approximately 300 minutes in

total. During that time, we could not identify any false positive

calls made to the database and the players did not show any

frustration towards the system for not being able to recognize

them holding the ball.

4 Conclusion

We have explained the process of developing a catch detection

device for the use of an augmented dodgeball game. We

introduced two different systems and how we calibrated the

system to ensure a more robust and reliable catch detection

during the game. Augmenting sports is a new and exciting field

with lots of potentials to make the exercise more engaging and

fun. However, these kinds of sports also rely on technology, and

the reliability of the technology determines how well the players

can experience the game design. That is why it is essential to

ensure the devices’ quality and reliability in such games. We

hope that our findings can provide interesting insights for further

developing equipment to enable augmented sports and provide

―136―


ideas on making the system both player-friendly and reliable.

Our work is limited by the device used for detecting which

player is holding the ball. However, the design principles for

player comfort and the functions to ease the setup can be used to

design other systems. Also, the way we troubleshoot to make the

system more reliable can be used in similar systems that use

some sensor(s) and/or different modules that produce noise that

can interfere with the intended device’s accuracy.

Acknowledgments

This work was supported by the Japan Society for the

Promotion of Science grants JP16H01741 and 19H01129.

References

[1] F. J. Penado and J. R. Dahn, “Exercise and well-being: a

review of mental and physical health benefits associated

with physical activity,” Curr. Opin. Psychiatry, vol. 18,

no. 2, pp. 189–193, 2005, doi: 10/cfcxb2.

[2] D. E. R. Warburton, C. W. Nicol, and S. S. D. Bredin,

“Health benefits of physical activity: the evidence

Review,” CMAJ, vol. 174, no. 6, p. 801, 2006, doi:

10.1503/cmaj.051351.

[3] “Physical activity.” https://www.who.int/news-room/fact-

sheets/detail/physical-activity (accessed Jun. 01, 2020).

[4] T. S. Church et al., “Trends over 5 Decades in U.S.

Occupation-Related Physical Activity and Their

Associations with Obesity,” PLoS One, vol. 6, no. 5, p.

e19657, May 2011, doi: 10.1371/journal.pone.0019657.

[5] L. Straker, P. Coenen, D. Dunstan, N. Gilson, and H.

Genevieve, “Sedentary work. Evidence on an emergent

work health and safety issue,” 2016.

[6] M. Kondrič, J. Sindik, G. Furjan-Mandić, and B.

Schiefler, “Participation motivation and student’s physical

activity among sport students in three countries,” J. Sport.

Sci. Med., vol. 12, no. 1, pp. 10–18, Mar. 2013.

[7] S. Allender, G. Cowburn, and C. Foster, “Understanding

participation in sport and physical activity among children

and adults: a review of qualitative studies,” Health Educ.

Res., vol. 21, no. 6, pp. 826–835, Dec. 2006, doi:

10.1093/her/cyl063.

[8] K. Gerling and R. Mandryk, “Custom-designed motion-

based games for older adults: A review of literature in

human-computer interaction,” Gerontechnology, vol. 12,

no. 2, pp. 68–80, 2014, doi: 10.4017/gt.2013.12.2.001.00.

[9] K. M. Gerling, M. Miller, R. L. Mandryk, M. V. Birk, and

J. D. Smeddinck, “Effects of balancing for physical

abilities on player performance, experience and self-

esteem in exergames,” 32nd Annu. ACM Conf., pp. 2201–

2210, 2014, doi: 10.1145/2556288.2556963.

[10] K. Rebane, T. Kai, N. Endo, T. Imai, T. Nojima, and Y.

Yanase, “Insights of the augmented dodgeball game

design and play test,” in ACM International Conference

Proceeding Series, Mar. 2017, pp. 1–10, doi:

10.1145/3041164.3041181.

[11] T. Ohta, S. Yamakawa, T. Ichikawa, and T. Nojima,

“TAMA: Development of Trajectory Changeable Ball for

Future Entertainment,” 2014, doi:

10.1145/2582051.2582101.

[12] K. Nitta, K. Higuchi, Y. Tadokoro, and J. Rekimoto,

“Shepherd Pass: Ability Tuning for Augmented Sports

using Ball-Shaped Quadcopter,” doi:

10.1145/2832932.2832950.

[13] K. Masai, Y. Sugiura, and M. Sugimoto, “ACTUATE

Racket: Designing Intervention of User’s Performance

through Controlling Angle of Racket Surface,” doi:

10.1145/3041164.3041200.

[14] H. Dogai, M. Oki, and K. Tsukada, “SomaticBall: Ball-

Type Device Providing ‘Sticking Feeling,’” doi:

10.1145/3001773.3001810.

[15] S. Sakai, Y. Yanase, Y. Matayoshi, and M. Inami, “D-Ball:

Virtualized sports in diminished reality,” in Proceedings

of the First Superhuman Sports Design Challenge: First

International Symposium on Amplifying Capabilities and

Competing in Mixed Realities, 2018, pp. 1–6, doi:

10.1145/3210299.3210305.

[16] P. Baudisch et al., “Imaginary reality gaming: Ball games

without a ball,” in UIST 2013 - Proceedings of the 26th

Annual ACM Symposium on User Interface Software and

Technology, 2013, pp. 405–410, doi:

10.1145/2501988.2502012.

[17] C. Eichhorn, A. Jadid, D. A. Plecher, S. Weber, G.

Klinker, and Y. Itoh, “Catching the Drone-A Tangible

Augmented Reality Game in Superhuman Sports,” in

2020 IEEE International Symposium on Mixed and

Augmented Reality Adjunct (ISMAR-Adjunct), Mar. 2020,

pp. 24–29, doi: 10.1109/VR.2019.8798056.

[18] Y. Sano, K. Sato, R. Shiraishi, and M. Otsuki, “Sports

Support System: Augmented Ball Game for Filling Gap

between Player Skill Levels,” doi:

10.1145/2992154.2996781.

[19] P. Blank, T. Kautz, and B. M. Eskofier, “Ball Impact

Localization on Table Tennis Rackets using Piezo-electric

Sensors,” doi: 10.1145/2971763.2971778.

[20] L. P. Cuspinera, S. Uetsuji, F. J. O. Morales, and D.

Roggen, “Beach Volleyball serve type recognition,” doi:

10.1145/2971763.2971781.

[21] B. Zhou et al., “Smart Soccer Shoe: Monitoring Foot-Ball

Interaction with Shoe Integrated Textile Pressure Sensor

―137―


Matrix,” no. 16, doi: 10.1145/2971763.2971784.

[22] 四條亮太, Rebane K., and 野嶋琢也, “球技拡張のため

のプレイヤ・ボールインタラクション検出デバイス,” in the

22nd Annual Conference of the Virtual Reality Society of

Japan, September 2017, 2017.

[23] “Sony Japan | FeliCaウェブサイト.”

https://www.sony.co.jp/Products/felica/ (accessed Oct. 08,

2020).

[24] “Arduino Uno Rev3 | Arduino Official Store.”

https://store.arduino.cc/usa/arduino-uno-rev3 (accessed

Oct. 08, 2020).

[25] “Digi XBee Ecosystem - Everything you need to explore

and create wireless connectivity | Digi International.”

https://www.digi.com/xbee (accessed Oct. 08, 2020).

[26] “Arduino - ArduinoBoardFio.”

https://www.arduino.cc/en/pmwiki.php?n=Main/Arduino

BoardFio (accessed Dec. 14, 2020).

[27] “Adafruit HUZZAH32 – ESP32 Feather Board ID: 3405 -

$19.95 : Adafruit Industries, Unique & fun DIY

electronics and kits.”

https://www.adafruit.com/product/3405 (accessed Oct. 08,

2020).

(Received on December 26th, 2020)

[著者紹介]

Kadri Rebane （非会員）

Kadri Rebane is a Ph.D. student at the

University of Electro-Communications. Her

research field is augmented sports. Previously

she has obtained a has a BSC degree in

mechanical engineering from Tallinn University of Technology

and a M.Eng degree in informatics from the University of

Electro-Communications in Tokyo, Japan.

David Hoernmark （非会員）

David Hoernmark received a Bachelor’s

Degree in Software Engineering from

Blekinge Institute of Technology (BTH) in

2020. In 2018, during his education, he

attended one semester at the University of Electro-

Communications (UEC) in Chofu, Tokyo, Japan. At UEC he

joined the Nojima Laboratory, where he conducted his research

in the field of augmented sports. His current occupation is as a

software developer consultant in the field of telecommunications.

Ryota Shijo （学生会員）

Ryota Shijo is a Ph.D. student at the

University of Electro-Communications.

Previously he received the BE and M.Eng

degrees from the University of Electro-

Communications in Tokyo, Japan in 2015 and 2017. His research

interests are human interface, human-computer interaction and

human-robot interaction.

Sho Sakurai （正会員）

Sho Sakurai received a BE degree in social

and information studies from Gunma

University in 2007, a MAE degree in inter-

disciplinary information studies in 2010, and

a Ph.D. degree in Engineering in 2014 from the University of

Tokyo. She is currently a project assistant professor in the

Graduate School of Information Systems, the University of

Electro-Communications. Her research interests are multi-

modal/cross-modal interfaces, human-computer interaction, and

Perceptual/Cognitive psychology. She is also active as a manga

artist to introduce the latest research on virtual reality, human

interface, and artificial Intelligence.

Koichi Hirota （正会員）

Koichi Hirota received the BS and Ph.D.

degrees from the University of Tokyo, Japan, in

1988 and 1994, respectively. He was then an

Assistant Professor with the Toyohashi

University of Technology in 1995. In 2000, he was an Associate

Professor at the University of Tokyo. He is currently a Professor

with the Department of Informatics, Graduate School of

Informatics and Engineering, The University of Electro-

Communications. His research interests include haptic rendering

and human interfaces.

Takuya Nojima （正会員）

Takuya Nojima received a Ph.D. degree in

Engineering from The University of Tokyo,

Japan, in 2003. He joined the Japan Aerospace

Exploration Agency as a pilot interface and VR

simulation researcher in 2003. He is currently an Associate

Professor with the Department of Informatics University of

Electro-Communications, Tokyo, Japan from 2008. His research

interests include haptic interaction, superhuman sports, human

interface, and virtual reality.

―138―

Documents

Developing the Thrower Detection System for Seamless