ASSESSING THE ACCURACY OF THE

PHOTO FINISH: AUTOMATIC TIMING

ANDROID APP

Johann-Lukas Voigt, Arthur Voigt, Paulo Leite, Andreas Plewnia

Photo Finish GbR

support@photofinish-app.com
3.2.2024

Abstract
This study presents an in-depth evaluation of Photo Finish: Automatic Timing [1], an innovative sports
timing app developed for Android smartphones. The primary focus was on assessing the accuracy of
this system for various starting modes and a wide array of different phone models. We also included a
comparison with an ordinary light barrier system, where Photo Finish demonstrated significantly higher
accuracy due to its chest detection algorithm. In all tested configurations, three-quarters of all
measurements of Photo Finish were accurate within 10 milliseconds or less. Furthermore, 95% of the
measurements fell within a 15-millisecond margin, and all recorded times were consistently within 20
milliseconds of the times obtained through manual video evaluation. Thus, we conclude that the Photo
Finish system offers a highly accurate, cost-effective, and accessible solution for timing in sports,
demonstrating an accuracy level of less than 0.02 seconds under the test conditions.
Table of Contents
1 Introduction 3

1.1 Setups 3

1.2 Phone Pairings 4

2 Analysis Setup 5

2.1 Flying Start Measurement Setup 7

2.2 Touch Start Measurement Setup 8

2.3 Beep Start Measurement Setup 9

2.4 Sound Detection Start Measurement Setup 10

2.5 Audio and Video Synchronization 11

3 Results and Discussion 11

4 Technical Details and Potential Error Sources 14

4.1 Bluetooth Synchronization Accuracy 14

4.2 Smartphone Clock Drift 14

4.3 Camera Timestamps 15

4.4 Exposure Duration 15

5.5 Chest Detection Algorithm 15

4.6 Touch Screen and Microphone Latencies 15

References 17

Appendix 1: Flying Start Measurements 18

Appendix 2: Touch Start Measurements 19

Appendix 3: Beep Start Measurements 20

Appendix 4: Sound Start Measurements 21

Appendix 5: Additional Measurements with an Older Smartphone 22

1 Introduction
Photo Finish: Automatic Timing is a smartphone application designed to meet the demands of accurate
timing in various sports. The app leverages the capabilities of modern smartphones to offer a practical
yet sophisticated solution for measuring sports performance.

Photo Finish performs automatic detection of the exact position of the athlete's chest, using the
phone's camera. This approach aims to address the timing inaccuracies commonly found in traditional
systems such as light barriers, which may be inadvertently triggered by non-target parts of an athlete's
body like the knee or hands.

The focus of this study is to rigorously evaluate the accuracy of the Photo Finish app. Through a series
of multiple experiments, we aim to establish not only its effectiveness in accurate timing but also to
compare its precision against a commercial light barrier system, the SpeedTech S-003 Wireless Laser
Timing System [2].

1.1 Setups
Four distinct experiments have been selected for their direct relevance to real-world applications.
These scenarios were chosen due to their common occurrence in training for competitive sports and
their ability to provide a comprehensive evaluation of the timing app’s accuracy across a range of setup
conditions. The tests were conducted on a track and field course during daylight to ensure optimal
lighting conditions for Photo Finish.

1. 30-meter flying sprint

A common test in athletics is to measure an athlete’s top-end speed. The corresponding test
setup is explained in 2.1.

2. 40-yard dash without reaction time

The 40-yard dash is a staple in evaluating an athlete's acceleration and speed, particularly in
sports like American football. This test, incorporating reaction time, is pivotal in assessing an
athlete's quickness off the mark. The corresponding test setup is explained in 2.3.

3. 40-yard dash with touchpad trigger.

In this variation of the 40-yard dash, the start is triggered the moment the athlete lifts his
thumb from a starting touchpad. The corresponding test setup is Touch Start with two connected
phones.

4. 100m sprint with a start gun

The 100-meter sprint, often initiated with a start gun, is a quintessential track event testing an
athlete's explosive speed and endurance. The corresponding test setup is Sound Detection Start
with two connected phones.
1.2 Phone Pairings
Additionally, we intend to demonstrate that Photo Finish operates with precision across a diverse range
of smartphones, including cheaper ones. Therefore, for each of the four experimental setups, seven
different phone pairings were chosen from the following list of 12 different phone models:

Manufacturer Model

Samsung Galaxy S20 FE

Samsung Galaxy S8

Samsung Galaxy A52

Huawei P30 Pro

Huawei P30 Lite

Huawei P8 Lite

Google Pixel 3 XL

One Plus Nord CE2

Xiaomi Poco

Xiaomi Redmi Note 9s

Cubot Note 20 Pro

Realme C3
2 Analysis Setup
In each of the setups discussed in this chapter, we tested seven phone pairings. For every pairing, 8
measurements were conducted, totaling 4x7x8 = 224 individual measurements. In Flying Start Mode,
we also compared Photo Finish against a commercial light barrier timing system, the SpeedTech S-003
Wireless Laser Timing System [2], for which we also determined the measurement error.
The study was performed on two consecutive days. Two male test subjects aged 30 and 32 alternated
in performing the runs for the required measurements. We used Version 3.3.1 of Photo Finish:
Automatic Timing which was installed on all phones before the experiments. The tests were conducted
from around 11 am to 4 pm with overcast weather intermixed with occasional sunshine.

To analyze the experiments, we used a high-speed smartphone camera capable of continuous audio
and video capture at 240 frames per second to analyze the Photo Finish app's performance. On day
two of the study, we also adjusted the font size on the phones to be extra large, ensuring that the times
recorded by Photo Finish were visible in the high-speed camera's video feed. This enables interested
readers to independently verify the measurements using the original video material which can be
requested via email. The video footage was then analyzed using the Da Vinci Resolve Video editing
software [3].

Note that the Start and Finish Phones are close by in the setups described below. In practical scenarios,
they would typically be placed further apart, such as in a 30m flying sprint test. However, for this
experiment, we positioned the Start and Finish Phones on the same line to allow for the use of a single
high-speed camera that captures both the start and the finish of each run. This arrangement does not
impact the app's precision, as the times are transmitted over the internet and the phones synchronize
their clocks upon initial connection. Further technical details are discussed in Section 3.

The alignment of the whole setup for Photo Finish, the laser timing system, and the camera evaluation
was thoroughly checked to allow for a measurement accuracy of one millisecond. The alignment was
rechecked after every phone swap.
This means, that at an expected speed of 5m/s or higher of the passing test subjects, the laser beam
must be aimed precisely perpendicular to the track and the end of the laser beam and may deviate at
most 5m/s * 1/1000 s = 5mm from the exact middle at both ends. To be able to precisely determine
the vertical plane of crossing on the high-speed camera image, a rectangular wooden triangle was
constructed and placed in the background at the measurement line (Figure 1).

The camera has a resolution of 1280x720 pixels at a 35° horizontal field of view. The total distance
between the laser sensors is kept at 4m. This results in a theoretical timing resolution of the camera at
a running pace of 5m/s or higher of

π 4𝑚
35/720 × 180 5𝑚/𝑠
≈ 0. 7𝑚𝑠

The almost perfectly linear movement of an athlete in 2 consecutive frames of 240fps video footage
allows for frame interpolation from 240 to 1000fps using the built-in frame interpolation algorithm of
The Da Vinci Resolve video editing software to allow for easy and precise analysis, see Figure 2:
Figure 1: Flying Start Measurement Setup. A wooden triangle is placed at the starting line to enable precise
video analysis. Two timing gates are placed opposing each other on the same measurement line. An
Additional Smartphone films the complete setup at 240fps for later video analysis.

Figure 2: 1000fps video analysis. The timestamp of the frame where the chest is exactly at the height of the
wooden beam is evaluated for every run. The right 3 images show detail views 1ms before, at, and 1ms after
the crossing, showing both the need for precise alignment of the equipment as well as the large influence of
triggering on the legs or the arms when assessing runs with millisecond precision.
2.1 Flying Start Measurement Setup
The Flying Start mode in Photo Finish starts the clock when an athlete passes a starting phone after a
runup and finishes when it passes the last measurement line. With additional phones, as many
measurement points as desired can be added. This mode therefore allows for direct comparison with a
laser timing system.
The experimental setup is depicted in Figure 3. For a single measurement, the test subject, acting as an
athlete, runs past the Start Phone and the Start Light Barrier, triggering the start of the timer. They
then turn around and sprint past the Finish Phone and the Finish Light Barrier, which stops the timer
and completes the measurement.
We used the following seven phone pairings, with eight measurements per pairing (See Appendix 1):

Start Phone Finish Phone

Samsung Galaxy S20 FE Huawei P30 Pro

Samsung Galaxy S8 Huawei P30 Pro

Realme C3 Huawei P30 Pro

Xiaomi Redmi Note 9S Huawei P30 Pro

Huawei P30 Lite Google Pixel 3XL

Samsung Galaxy A52 Google Pixel 3XL

Samsung Galaxy S20 FE Huawei P30 Pro

Figure 3: Two Smartphones are connected in Flying Start mode and placed back to back on a single line. The
light barrier timing system is placed on the same line. The high-speed camera is also placed on the same line
and aligned such that it can accurately monitor the passing of an athlete on both sides.
2.2 Touch Start Measurement Setup
Photo Finish's touch start mode allows for low start sprint measurements without reaction time, for
example, the 40-yard dash. The athlete places his thumb on the phone and the timer is started as soon
as the thumb is lifted from the touchscreen.
The setup for the Touch Start experiment is shown in Figure 4. In this test, first, a finger is placed on the
touch screen of the Start Phone. The measurement then begins once the touch screen is released by
lifting the finger. The test subject then starts his run and the measurement once again concludes as he
crosses the measurement line of the Finish Phone.

We used the following seven phone pairings, with eight measurements per pairing (See Appendix 2):

Start Phone Finish Phone

Samsung Galaxy S20 FE Huawei P30 Pro

Samsung Galaxy S8 Huawei P30 Pro

Realme C3 Huawei P30 Pro

Xiaomi Redmi Note 9S Huawei P30 Pro

Google Pixel 3 XL Huawei P30 Pro

Samsung Galaxy A52 Huawei P30 Pro

One Plus Nord CE 2 Huawei P30 Pro

Figure 4: Two smartphones are connected in Touch Start mode. The Start Phone is mounted on a tripod
within the field of view of the high-speed camera. This setup allows for precise determination of the moment
when the finger is released from the screen. The Finish Phone is aligned along the same line as the high-speed
camera.
2.3 Beep Start Measurement Setup
Photo Finish has several starting sequences that end with a 3 kHz beep sound initiating the start. This
can either happen after a random delay ("Ready-Set-Go!" mode) to test the athlete's reaction or after a
predictable fixed delay to eliminate reaction time ("3-2-1-Go!" mode). Photo Finish uses a beep signal
instead of imitating a gunshot sound for two reasons: The small smartphone speakers are able to play
a beep sound at a much higher volume, and the beep sound allows for frequency analysis of the
microphone input to robustly determine the exact moment the sound was played independent of
possible delays (i.e. through Bluetooth speakers) and loud wind noises.
The setup for the Beep Start experiment is shown in Figure 5. In this test, a single measurement begins
with the press of the start button on the Start Phone, which then emits a beep sound signaling the start
of the run. The athlete runs past the Finish Phone, at a distance of approximately 1.5 meters. The
measurement concludes as the runner crosses the measurement line of the Finish Phone.

We used the following seven phone pairings, with eight measurements per pairing (See Appendix 3):

Start Phone Finish Phone

Samsung Galaxy S20 FE Huawei P30 Pro

Samsung Galaxy S8 Huawei P30 Pro

Huawei P8 Lite Huawei P30 Pro

Xiaomi Redmi Note 9S Huawei P30 Pro

Google Pixel 3 XL Huawei P30 Pro

OnePlus Nord CE 2 Huawei P30 Pro

Xiaomi Poco Huawei P30 Pro

Please note that the last measurement is missing for the Pixel 3 XL / P30 Pro combination due to an
oversight.

Figure 5: In Beep Start mode, two smartphones are connected. The Start Phone is placed within 0.5 meters of
the high-speed camera, its specific location being less critical as long as it is close to the high-speed camera
to make sound travel time irrelevant. The Finish Phone is aligned on the same line as the high-speed camera.
2.4 Sound Detection Start Measurement Setup
Photo Finish can also trigger on a loud external sound, such as a starting clapper, a starting gun, or
simply a clap with the hands.
The setup for the Sound Detection Start experiment is shown in Figure 6. In this test, a single
measurement begins when two metal rods are struck against each other. This emits a loud sound
which is detected by the Start Phone, which then starts the timer. The athlete then runs past the Finish
Phone, again at a distance of approximately 1.5 meters. The measurement concludes as the runner
crosses the measurement line of the Finish Phone, upon which it takes the final time.

We used the following seven phone pairings, with eight measurements per pairing (See Appendix 4):

Start Phone (Sound Detection) Finish Phone

Huawei P30 Lite Google Pixel 3XL

Samsung Galaxy S20 FE Google Pixel 3XL

Xiaomi POCO Samsung Galaxy S20 FE

Xiaomi Redmi Note 9S Huawei P30 Pro

Huawei P30 Lite Huawei P30 Pro

Huawei P30 Pro Google Pixel 3 XL

Cubot Note 20 Pro Google Pixel 3XL

Figure 6: In Sound Detection Start mode, two smartphones are connected. The Start Phone is placed within
0.5 meters of the high-speed camera, its specific location being less critical as long as it is close to the
high-speed camera to make sound travel time irrelevant. The Finish Phone is aligned on the same line as
the high-speed camera. To start a measurement, two metal bars are struck against each other. The emitted
sound is detected by the Start Phone, which then starts the timer. The two metal rods also allow for a precise
alignment of the audio and video tracks of the high-speed camera.
2.5 Audio and Video Synchronization
For the analysis of the Beep and Sound Start modes, accurate audio and video synchronization is
paramount. The high-speed camera already captures an audio signal at a 48khz sampling rate during
video recording. However, the audio and video signals are not accurately synchronized by default. A
starter clapper was intended to be used for accurate synchronization: During the video analysis, the
audio track is aligned such that the peak of the clap sound happens precisely at the frame when both
halves of the clapper touch. Preliminary tests, however, showed that a typical start clapper is unsuitable
for the desired precision, as it bends during the clapping movement and emits several sound waves
over the span of roughly three to five milliseconds (See Figure 7). Therefore, two iron rods were used
instead. At the end of each run, the iron rods were triggered a second time to verify that the tracks
were still synchronized and apply a potential offset correction.
The speed of sound at ground level is roughly 34cm per millisecond [6]. Since the phone and the
clapper are positioned at roughly the same distance from the camera, the synchronization offset
cancels out and the synchronized audio and video tracks of the high-speed camera can be directly used
for measuring the time between the Photo Finish start sound and the athlete finish crossing for an
overall measurement accuracy of ±1 millisecond.

Figure 7: A typical start clapper is unsuitable for precise audio and video alignment as it bends during
clapping and emits sound over several milliseconds. We therefore used two metal rods.
3 Results and Discussion
The video analysis shows very high accuracy of Photo Finish across all tested modes of less than 0.02
seconds and notable performance benefits in comparison to the laser timing system. In Flying Start
mode, 75% of Photo Finish measurements were within 10 milliseconds or less of the manual video
evaluation time, 95% were within 14 milliseconds, and all measurements were within 19ms of the video
evaluation time. In contrast, the laser timing system exhibited 75, 95, and 100 Percentile errors of 26,
43, and 47 milliseconds respectively, which are further discussed below. In total, Photo Finish is about
two and a half times more accurate than the laser timing system (See Appendix 1).

In the Touch Start mode, the results were similar with 75% of Photo Finish's measurements differing by
less than 7 milliseconds, 95% by less than 10 milliseconds, and the maximum error being 19
milliseconds (See Appendix 2). The laser timing system does not have a comparable mode which we
were able to evaluate. It's important to note that our evaluation accuracy was slightly lower in Touch
Start mode, as even with quick movement of the finger the exact moment of the finger lifting from the
touchscreen is hard to determine in the video footage because the finger accelerates from rest instead
of a linear movement like the chest on the finish line. During the first 2milliseconds, the finger typically
still touches the screen somewhat to its elasticity but with less pressure. Therefore, in this mode, our
video analysis measurements themselves are subject to up to 3 millisecond error.

In the Beep- and Sound Start modes, the results are similar with 75% being within 10 milliseconds and
all measurements equal or less than 20 milliseconds off from the video evaluation time. (See Appendix
3 and Appendix 4). The Laser timing system does also have an additional Loudspeaker which can be
triggered by a button and start the clock. We tested this mode and saw a very large timing offset of
around 450 milliseconds (See first phone pairing in Appendix 4) which was even noticeable during the
measurements without any video evaluation tools. We did contact the manufacturer about this and
were informed that the timing system automatically subtracts 150ms from the measurements in this
mode to eliminate the reaction time. This, however, was nowhere mentioned in the manual and also
does only explain less than half of the offset, so we assume that there is some calibration error
internally. We therefore excluded the speaker mode of the laser timing system from further
measurements.

In the Flying Start measurements, the comparatively lower accuracy of the laser timing system can be
primarily attributed to its mechanism of triggering: Any part of the athlete's body rather than just their
torsos can trigger the laser- typically a runner's knee or hand. This detection method introduces
variability, as the exact moment an athlete's torso crosses the finish line—the relevant measure in
sprinting events—is not directly measured. This is also clearly visible in the video evaluation footage,
see Figure 8.

These resulting measurement errors are also relevant for training environments. As shown by the video
analysis, these errors can be up to 0.05s, which are significant for short sprint measurements such as
30m flying, where Chu (1996) for example divides athletes into sprinting ability categories and where
the width of categories is as low as 0.20s in total.
 Figure 8: Early triggering of the Laser timing on the thigh instead of the torso. The left image and middle
image show the moment the laser was broken and the clock was started, visible as the laser light dims
significantly because it is not being reflected by the mirror anymore. When the chest of the passing runner
has reached the measurement line/wooden beam, the timer on the Laser timing system already shows 34
milliseconds even though there is an additional screen latency.

Despite all the possible error sources that a smartphone-based timing system using wireless
connection faces discussed in Section 4, Photo Finish manages to surpass the accuracy of light barrier
timing due to its chest detection algorithm, offering a more precise, accessible, and cost-effective
solution for amateur and professional sports timing as well as performance evaluation.

At this point, we also want to emphasize that despite its primary intended use for running sports, Photo
Finish can also be used in a variety of other athletic disciplines such as mountain biking and skiing. The
chest detection algorithm has been calibrated specifically for running, such that for example it also
detects the chest of a cyclist instead of triggering on the frontmost part of the front wheel. This leads to
slight, but mostly constant differences in timing, which depending on the application can either be
ignored or just added or subtracted from the measurements to still get automatic timing results that
are vastly more accurate than manual timing in a variety of sports.

Gender Excellent Above Average Below Poor

Average Average

Male <2.6 secs 2.6 - 2.9 secs 2.9 - 3.1 secs 3.1 - 3.3 secs >3.3 secs

Female <3.0 secs 3.0 - 3.3 secs 3.3 - 3.5 secs 3.5 - 3.7 secs >3.7 secs

Figure 9: 30m flying sprint normative data from Chu (1996)

4 Technical Details and Potential Error
Sources
In the previous chapters, the overall accuracy of the Photo Finish Timing App in the various modes has
been established. In this chapter, we want to discuss how we managed to achieve this level of precision
despite the limitations of the Android operating system and smartphone technology in general and
what the sources of the remaining errors are, while also stating the preliminary conditions that ensure
that these are kept as low as possible.

4.1 Bluetooth Synchronization Accuracy

Photo Finish uses the Bluetooth protocol to synchronize the smartphones. Photo Finish initiates
multiple connection attempts and uses the one with the lowest measured ping. The resulting typical
synchronization latency ranges between 20-40 milliseconds, potentially introducing an error margin of
up to half the ping. However, our preliminary analysis, which involved multiple trials with a high-speed
camera setup like in Section 2.1 and alternating start and finish positions, indicated that the actual
impact on accuracy is minimal. We observed that Bluetooth ping is typically highly symmetrical,
resulting in a synchronization accuracy that is generally under 5 milliseconds. However, due to the large
range of possible phone conditions and vendors, a higher ping with other phone combinations can't be
ruled out.

4.2 Smartphone Clock Drift

The internal quartz oscillators of smartphones are subject to clock drift post-synchronization. According
to Chu Luo et al (2017), the worst-case drift scenario is approximately 2 parts per million (ppm),
equating to a potential error of about 7 milliseconds per hour. To counteract this, Photo Finish
implements a re-synchronization protocol every 5 minutes, therefore negating the impact of the clock
drift on short-time measurements such as sprints and interval training. The resynchronization is
however only successful if the phones are within Bluetooth range, which is up to 40m in an open area.
If the phones are placed further apart, such as in a 100m sprint, or for long-range timings such as
marathon runs, Photo Finish accuracy is expected to decrease by up to 0.01s/hour since the session
was started.

Since the quartz oscillator of the smartphone camera used for the high-speed analysis is equally
subject to these fluctuations, all individual runs were kept to a length of fewer than 20 seconds so that
the clock drift was negligible for the analysis.
4.3 Camera Timestamps
Every time the smartphone camera captures an image, the time at the beginning of the exposure is
noted down. This timestamp can however be off by a significant margin due to different delays and
interruptions in the Android operating system. Our investigation into the timestamp accuracy of
smartphone cameras encompassed over 15 different models spanning various release years. We found
that models released within the last 8 years generally maintained timestamp accuracy within less than
10 milliseconds. An outlier, the Huawei P8 Lite from 2015, demonstrated a video timestamp deviation
of approximately 20 to 30 milliseconds which did significantly impact this phone's accuracy. The table in
the appendix contains the measurements with this phone, showing an overall error of up to 30ms. We,
therefore, excluded this phone from the analysis and recommend using more recent phone models.
Note that the achieved accuracy still excels the light barrier system even on this outdated smartphone.

4.4 Exposure Duration

Photo Finish image capture rate of 30 frames per second dictates that camera exposure duration can
vary based on lighting conditions, ranging from approximately 0.5 to a maximum of 33 milliseconds.
Our empirical research showed that in lower light conditions (such as dusk or indoor environments),
Photo Finish's chest detection algorithm triggers between the middle and the end of a blurred image
section. Photo Finish therefore automatically optimizes measurement accuracy in low light conditions
by adding three-quarters of the exposure duration to the video frame timestamp. For this study, we
confined our accuracy tests to bright daylight conditions which keep the exposure length to around one
millisecond to ensure optimal accuracy.

5.5 Chest Detection Algorithm

The chest detection algorithm is a critical component for accurate measurements. Challenges arise
when there is low color contrast between the athlete and the background, potentially leading to missed
detections or slight misplacements of the chest marker. In our study, we mitigated this risk by using
brightly colored shirts (See Figure 10). Additionally, Photo Finish's feature of saving and displaying
captured images with chest markers facilitates easy verification of accurate athlete capture.

4.6 Touch Screen and Microphone Latencies

Android smartphones, not inherently designed for high-speed, real-time applications, exhibit notable
latencies in sound playback, microphone buffering, and touchscreen responsiveness. Sound playback
latency is additionally unpredictable and can, depending on the phone model, be occasionally as high
as 400ms or more. Luckily, touchscreen and microphone latencies are constant, as otherwise, this
would lead to corrupted microphone data and/or missed touchscreen events.
Photo Finish therefore uses the microphone to determine the exact moment in time when the start
sound is emitted. This still requires knowing the exact microphone latency, which also varies from
phone model to phone model. We successfully developed a calibration routine to calculate these
latencies and subtract them from the measurements, which the user is requested to perform before
using Photo Finish for the first time.

Figure 10: Typical recording of Photo Finish. The white markers show where the chest detection algorithm
triggered. The final time on the top right is calculated by linear interpolation of the frame timestamps and the
distance of the chest marker from the measurement line (wooden beam in the background).
References
1. Photo Finish: Automatic Timing, https://photofinish-app.com

2. Chu Luo et al. "Rapid clock synchronisation for ubiquitous sensing services involving multiple
smartphones." Proceedings of the 2017 ACM International Joint Conference on Pervasive and
Ubiquitous Computing and Proceedings of the 2017 ACM International Symposium on Wearable
Computers (2017). https://doi.org/10.1145/3123024.3124432.

3. S-003 Wireless Laser Timing System,

https://66start.com/en-de/products/s-003-wireless-laser-timing-system

4. Da Vincy Resolve Video Editing Software,

https://www.blackmagicdesign.com/de/products/davinciresolve

5. CHU, D.A. (1996) Explosive Power and Strength. Champaign: Human Kinetics

6. http://hyperphysics.phy-astr.gsu.edu/hbase/Tables/Soundv.html
 Appendix 1: Flying Start Measurements
Phone Photo Finish Laser Timing High-Speed Error Error
Pairing Run Measurement System Measurement Camera Manual Photo Finish Laser Timing
(seconds) (seconds) Evaluation (ms) System (ms)
1 7.623 7.613 7.612 11 1
2 7.410 7.411 7.398 12 13
3 9.301 9.333 9.296 5 37
Samsung Galaxy S20 4 6.634 6.650 6.630 4 20
FE /
Huawei P30 Pro 5 8.489 8.476 8.484 5 -8
6 9.648 9.653 9.648 0 5
7 9.946 9.930 9.944 2 -14
8 11.081 11.114 11.089 -8 25
1 7.753 7.741 7.748 5 -7
2 8.704 8.721 8.699 5 22
3 9.412 9.385 9.402 10 -17

Galaxy S8 / 4 8.200 8.207 8.188 12 19

Huawei P30 Pro 5 9.849 9.881 9.849 0 32
6 11.564 11.545 11.551 13 -6
7 15.239 15.199 15.220 19 -21
8 11.109 11.090 11.104 5 -14
1 10.401 10.389 10.392 9 -3
2 9.943 9.927 9.933 10 -6
3 11.518 11.523 11.511 7 12

Realme C3 / 4 10.111 10.096 10.103 8 -7

Huawei P30 Pro 5 11.311 11.306 11.311 0 -5
6 7.970 7.961 7.962 8 -1
7 9.432 9.443 9.427 5 16
8 10.802 10.770 10.796 6 -26
1 7.061 7.107 7.060 1 47
2 7.401 7.372 7.399 2 -27
3 9.736 9.745 9.733 3 12
Xiaomi Redmi Note 4 8.744 8.714 8.740 4 -26
9S /
Huawei P30 Pro 5 9.451 9.461 9.448 3 13
6 10.306 10.286 10.305 1 -19
7 13.340 13.307 13.338 2 -31
8 5.626 5.655 5.622 4 33
1 8.510 8.531 8.513 -3 18
2 7.302 7.286 7.297 5 -11
3 11.278 11.292 11.277 1 15

Huawei P30 Lite / 4 13.269 13.308 13.273 -4 35

Google Pixel 3XL 5 10.934 10.960 10.935 -1 25
6 12.222 12.223 12.227 -5 -4
7 10.936 10.917 10.934 2 -17
8 12.219 12.262 12.220 -1 42
1 8.177 8.167 8.168 9 -1
2 9.092 9.094 9.081 11 13
3 13.402 13.392 13.399 3 -7
Samsung Galaxy A52 4 12.084 12.061 12.068 16 -7
/
Google Pixel 3XL 5 9.977 9.951 9.966 11 -15
6 9.647 9.636 9.644 3 -8
7 8.686 8.633 8.676 10 -43
8 7.195 7.180 7.181 14 -1
1 5.773 5.765 5.771 2 -6
2 6.863 6.838 6.853 10 -15
3 6.061 6.009 6.050 11 -41
Samsung Galaxy S20 4 6.411 6.388 6.397 14 -9
FE /
Huawei P30 Pro 5 6.031 5.982 6.027 4 -45
6 6.852 6.797 6.842 10 -45
7 8.043 8.015 8.028 15 -13
8 6.187 6.148 6.178 9 -30

75th Percentile Error (ms): 10.0 26.0

95th Percentile Error (ms): 14.3 43.5
Maximum Error (ms): 19.0 47.0
Appendix 2: Touch Start Measurements
Phone Photo Finish High-Speed Camera Error
Pairing Run Measurement Manual Evaluation Photo Finish
(seconds) (ms)
1 3.879 3.872 7
2 4.151 4.152 -1
3 4.145 4.145 0

Samsung Galaxy S20 FE / 4 4.368 4.371 -3

Huawei P30Pro 5 4.528 4.532 -4
6 3.023 3.020 3
7 3.623 3.619 4
8 3.194 3.193 1
1 4.182 4.175 7
2 3.336 3.335 1
3 3.446 3.438 8

Galaxy S8 / 4 3.409 3.408 1

Huawei P30 Pro 5 3.614 3.605 9
6 3.316 3.308 8
7 4.286 4.279 7
8 4.024 4.015 9
1 4.626 4.625 1
2 3.470 3.467 3
3 3.840 3.840 0

Realme C3 / 4 3.682 3.663 19

Huawei P30Pro 5 3.145 3.145 0
6 2.930 2.925 5
7 4.153 4.148 5
8 3.622 3.623 -1
1 10.767 10.760 7
2 3.628 3.633 -5
3 4.370 4.363 7

Xiaomi Redmi Note 9S / 4 4.129 4.125 4

Huawei P30Pro 5 4.004 4.008 -4
6 4.061 4.053 8
7 4.648 4.646 2
8 3.713 3.694 19
1 6.071 6.068 3
2 3.241 3.231 10
3 3.402 3.392 10

Google Pixel 3 XL / 4 3.255 3.253 2

Huawei P30 Pro 5 4.337 4.331 6
6 4.230 4.232 -2
7 3.979 3.971 8
8 3.692 3.694 -2
1 6.674 6.675 -1
2 4.705 4.712 -7
3 4.254 4.258 -4

Samsung Galaxy A52 / 4 4.335 4.345 -10

Huawei P30 Pro 5 4.181 4.189 -8
6 4.452 4.460 -8
7 4.304 4.308 -4
8 3.495 3.502 -7
1 3.495 3.494 1
2 3.099 3.101 -2
3 4.198 4.194 4

OnePlus Nord CE 2 / 4 4.061 4.062 -1

Huawei P30 Pro 5 3.439 3.435 4
6 3.662 3.667 -5
7 3.689 3.687 2
8 3.442 3.443 -1

75th Percentile Error (ms): 7.0

95th Percentile Error (ms): 10.0
Maximum Error (ms): 19.0
Appendix 3: Beep Start Measurements
Phone Photo Finish High-Speed Camera Error
Pairing Run Measurement Manual Evaluation Photo Finish
(seconds) (ms)
1 5.111 5.111 0
2 5.112 5.107 5
3 3.998 3.996 2
Samsung Galaxy S20 4 4.760 4.755 5
FE /
Huawei P30 Pro 5 4.734 4.735 -1
6 6.102 6.091 11
7 6.027 6.028 -1
8 5.144 5.143 1
1 4.321 4.321 0
2 4.239 4.240 -1
3 3.875 3.867 9

Samsung Galaxy S8 / 4 3.045 3.047 -2

P30 Pro 5 4.899 4.898 1
6 5.005 5.000 5
7 4.316 4.302 14
8 4.039 4.038 1
1 3.317 3.318 -1
2 3.619 3.618 1
3 3.765 3.773 -8

P8 Lite / 4 3.596 3.590 6

Huawei P30 Pro 5 4.005 4.014 -9
6 3.995 4.005 -10
7 3.983 3.995 -12
8 3.756 3.767 -11
1 4.367 4.366 1
2 4.445 4.454 -9
3 4.214 4.219 -5
Xiaomi Redmi Note 9S 4 4.300 4.308 -8
/
Huawei P30 Pro 5 4.862 4.864 -2
6 3.189 3.193 -4
7 4.220 4.221 -1
8 3.830 3.837 -7
1 4.158 4.157 1
2 5.392 5.392 0
3 4.537 4.537 0

Pixel 3 XL / 4 4.604 4.603 1

Huawei P30 Pro 5 6.525 6.522 3
6 4.063 4.066 -3
7 5.143 5.143 0
8 Measurement missing due to an oversight
1 4.503 4.502 1
2 4.298 4.295 3
3 4.384 4.385 -1

OnePlus Nord CE 2 / 4 4.473 4.469 4

Huawei P30 Pro 5 4.246 4.233 13
6 4.505 4.505 0
7 5.268 5.264 4
8 4.348 4.348 0
1 4.677 4.676 1
2 4.140 4.145 -5
3 5.075 5.074 1

Xiaomi POCO / 4 4.259 4.260 -1

Huawei P30 Pro 5 4.669 4.670 -1
6 4.121 4.119 2
7 4.239 4.239 0
8 4.090 4.093 -3

75th Percentile Error (ms): 5.3

95th Percentile Error (ms): 11.3
Maximum Error (ms): 13.7
 Appendix 4: Sound Start Measurements
Phone Photo Finish Laser Timing Error Error
Pairing Run Measurement System Measurement High-Speed Camera
Manual Evaluation Photo Finish Laser Timing
(seconds) (seconds) (ms) System (ms)
1 4.283 3.816 4.284 -1 -468
2 4.369 3.913 4.369 0 -456
3 7.858 7.394 7.858 0 -464
Huawei P30 Lite / 4 8.506 8.032 8.504 2 -472
Google Pixel 3XL 5 9.068 8.619 9.075 -7 -456
6 11.573 11.119 11.581 -8 -462
7 10.129 9.721 10.127 2 -406
8 8.748 8.321 8.755 -7 -434
1 8.175 Laser Timing System 8.189 -14 Laser Timing
2 4.308 was excluded from 4.312 -4 System was
3 3.173 further 3.182 -9 excluded from
Samsung Galaxy S20 4 3.217 measurements 3.221 -4 further
FE / because of obvious measurements
Google Pixel 3XL 5 4.682 4.694 -12
large systematic because of
6 3.373 3.376 -3
error (See Chapter obvious large
7 3.809 3.826 -17
3) systematic error
8 4.345 4.348 -3 (See Chapter 3)
1 23.219 23.230 -11
2 3.359 3.371 -12
3 3.054 3.068 -14
Xiaomi POCO / 4 4.881 4.890 -9
Samsung Galaxy S20
FE 5 3.622 3.639 -17
6 4.357 4.366 -9
7 2.926 2.936 -10
8 4.103 4.114 -11
1 4.419 4.422 -3
2 3.974 3.980 -6
3 3.715 3.718 -3
Xiaomi Redmi Note 4 4.103 4.108 -5
9S /
Huawei P30 Pro 5 4.200 4.194 6
6 4.257 4.264 -7
7 4.265 4.268 -3
8 4.060 4.071 -11
1 4.764 4.770 -6
2 4.717 4.725 -8
3 4.079 4.086 -7

Huawei P30 Lite / 4 4.064 4.071 -7

Huawei P30 Pro 5 3.616 3.618 -2
6 5.024 5.021 3
7 4.174 4.176 -2
8 4.333 4.330 3
1 7.203 7.214 -11
2 4.605 4.612 -7
3 4.233 4.240 -7

Huawei P30 Pro / 4 4.801 4.810 -8

Google Pixel 3 XL 5 3.895 3.901 -6
6 3.986 3.993 -7
7 4.588 4.592 -4
8 3.367 3.387 -20
1 5.543 5.550 -7
2 3.427 3.432 -5
3 5.605 5.606 -1

Cubot Note 20 Pro / 4 4.936 4.946 -9

Google Pixel 3XL 5 5.253 5.258 -5
6 5.210 5.220 -9
7 3.657 3.671 -14
8 4.643 4.657 -14
75th Percentile Error (ms): 9.6 -
95th Percentile Error (ms): 14.7 -
Maximum Error (ms): 20.0 -
Appendix 5: Additional Measurements
with an Older Smartphone
Photo Finish Laser Timing High-Speed Error Error
Phone Run Measurement System Camera Photo Finish Laser Timing
Pairing (seconds) Measurement Manual (ms) System (ms)
(seconds) Evaluation

1 7.328 7.346 7.349 -21 -3

2 8.632 8.677 8.653 -21 24
3 8.307 8.304 8.333 -26 -29

Huawei P8 Lite / 4 8.383 8.405 8.401 -18 4

Huawei P30 Pro 5 8.506 8.478 8.527 -21 -49
6 11.933 11.975 11.956 -23 19
7 8.097 8.110 8.127 -30 -17

8 6.868 6.898 6.888 -20 10

75th Percentile Error (ms): 23.8 25.3

Maximum Error (ms): 30.0 49.0