Abstracts Track 2023

Area 1 - Computer Systems in Sports
Area 2 - Health and Support Technology
Area 3 - Sport Performance and Support Technology

Area 1 - Computer Systems in Sports

Nr: 19
Title:

FenceNet: Fine-grained Footwork Recognition in Fencing
Authors:

Kevin Zhu, Alexander Wong and John McPhee
Abstract: Introduction There is a current need from national-level fencing teams for the development of analytical research to enhance performance and training. The first step is to achieve a deeper understanding of the physical, tactical, and technical demands of fencing. Once these demands are better understood, performance benchmarks can be created for different skill levels to identify gaps and evaluate athletes. This contributes to athlete selection, skills progression, and training interventions. The main bottleneck is the lack of a reproducible means to collect the high quality, high resolution, objective data required to create these benchmarks. Recognizing this deficiency in data quality, Malawski and Kwolek were the first to automate data collection in fencing from video. They proposed the JLJA method to classify fencing footwork from visual and inertial signals. JLJA was trained and evaluated on the Fencing Footwork Dataset (FFD). JLJA is currently the best performing method on FFD. However, the requirement of wearable sensors and depth video prevents JLJA from analyzing athletes during competition, from other teams, and from the past. To overcome this limitation, we introduce a novel architecture, FenceNet, that takes only 2D skeleton data as input and show that a variant of FenceNet outperforms JLJA on the same classification task. Methods FenceNet takes 2D pose data as input and classifies fencing footwork using a skeleton-based human action recognition approach that incorporates temporal convolutional networks to capture temporal information. This way, coaches and analysts could extract information directly from videos, by training FenceNet on 2D pose data extracted from a 2D pose estimator. Results FenceNet is trained and evaluated on FFD using 10-fold cross-validation. In each fold, data from one fencer is taken out as the test set. FenceNet achieved a classification accuracy of 85.4%, within 1% of JLJA (86.3%). A variant, BiFenceNet, that incorporates bidirectional temporal information achieved a classification accuracy of 87.6%, outperforming JLJA. Conclusion Current video analysis for the Canadian Olympic fencing team is primarily done manually by coaches and analysts. Due to the highly repetitive, yet dynamic and subtle movements in fencing, manual data analysis can be inefficient and inaccurate. We propose FenceNet to automate the classification of fine-grained footwork techniques in fencing. We show that our method outperforms the current state-of-the-art algorithm, JLJA, on the Fencing Footwork Dataset. Additionally, FenceNet improves on JLJA in the following: - Transferability to competition videos: Omitting the need for wearable sensors and depth videos al- lows FenceNet to be adapted and trained on competition videos. This allows coaches to extract in- formation from fencers from other teams and from the past. - Transferability to other techniques: Actions in fencing are highly composite. For example, an at- tack usually consists of a long sequence of varying movements used to counteract and react to the opponent’s movements. JLJA splits feature vectors into windows of 16 frames, which limits memory retention. In contrast, TCNs have access to substantially longer memory due to dilated convolutions, allowing FenceNet to be trained to classify other techniques in fencing. - Simplicity and automation: Unlike JLJA, FenceNet does not require manual feature extraction, feature selection, or feature fusion.
Nr: 54
Title:

Physiological Indices Describing Running Cost and Performance in Recreational Runners
Authors:

Hsiang Yu Hung and Amy P. Chiu
Abstract: Objective The purpose of this study is to assess the cardiorespiratory performance and running efficiency through cardiopulmonary exercise test (CPET) between male and female recreational runners. While most of researches were assessing performace with athletic level of runners, this study, however, focused on general publics for related features inspection. Determine the indicators that best describe the physiological differences based on the oxidative cost durning running may promotively benefit recreational runners with skill improvement and training strategies. Methods Forty-six recreational runners, male (n=23) and female (n=23), were recruited in this study for physiological assessments. Cardiopulmonary exercise tests were performed on the treadmill to determine the maximal oxygen uptake (VO2 max). A staged protocol was employed, 3 minutes of continuous running with 1 minute rest to achieve a steady state at each stage along with 1kph increment per stage. Male runners started at 10kph while females started at 8kph. The treadmill inclination was set to 1% for the whole test. The full completion stage prior to their exhaustion was analyzed for all subjects which further named as submaximal effort. Cardiorespiratory indexes, VO2 and vVO2 at maximal and submaximal effort and the cost of running (CR) were calculated through the respiratory measurements. (Medbø et al, 1988) (Maldonado-Martin et al, 2004) Results The test results showed that male runners have significantly higher cardiorespiratory profile in VO2 max (51±7, 42±7 ml/kg-min, p<0.01), VO2 submax (47±9, 40±8 ml/kg-min, p<0.01) and the running speed vVO2 submax (4.12±0.46, 3.46±0.43 m/s, p<0.01), while female runners arrived higher heart rate HRsubmax (178±11, 186±13 bpm, p=0.028) and %HRmax (95.5±3.7%, 97.3±1.6%, p=0.037) at sub-maximal effort, expressed in male and female respectively. Additionally, the respiratory exchange ratio (RER) for males and females reached (1.003 ± 0.037, 1.004 ± 0.058) and (1.028 ± 0.059, 1.017 ± 0.053) at sub-maximal and maximal effort (p=0.950, 615). On the other hand, the cost of running (CR) presented a significant correlation with VO2 submax (R2=0.632, p<0.01) and VO2 max (R2=0.459, p<0.01) in both males and females. However, both vVO2 submax (R2=0.039, p<0.01) and vVO2 max (R2=0.043, p<0.01) were significant but less correlated to the cost of running (CR). There is no significant difference found between gender comparison in CR (p=0.805). Conclusion Running efficiency can be represented in terms of the oxygen consumption during a steady state running. While male individuals are able to produce more energy due to higher oxgen consumption (VO2 max) as well as performance profile (vVO2 max), the cost of running showed no difference between male and female, indicating the running cost is not essentially related to VO2 max. As CR has no difference between genders and lower correlation to VO2 max, higher anaerobic limitation in male recreational runners may be assumed. Unlike well-trained athletes, it is suggested that recreational male runners focus on aerobic threshold improvement, while females concentrate on enhancing muscle strength as well as explosive power for not only performance advancement but physiological balance.

Area 2 - Health and Support Technology

Nr: 9
Title:

Digital Postural Analysis Using a Machine Learning Model: Applicability in Healthy Adults
Authors:

Federico Roggio, Sarah Di Grande, Salvatore Cavalieri and Giuseppe Musumeci
Abstract: Background: Postural alterations among the healthy population are steadily increasing. According to the Global Burden of Disease, musculoskeletal changes rank highly among the primary causes of disability in young adults. Postural analysis is an important approach for detecting musculoskeletal deviations. If not addressed promptly, these could lead to acute or chronic pain. However, traditional analysis methods can be prone to operator bias or have high costs due to sophisticated approaches. Recently, digital alternatives have emerged as more accessible methods for human motion analysis. MediaPipe Pose, a machine learning (ML) approach provided by Google, is one such algorithm. It estimates 3D human poses using a deep learning approach, enabling accurate measurements with standard digital cameras without special equipment or environments. This study proposes a machine learning-based approach for postural analysis using MediaPipe Pose. Methods: We analyzed the posture of 100 healthy adults, 50 males and 50 females, with an average age of 27.4 (SD ± 3.2) years. We excluded those with past or current musculoskeletal, spine, and neurological pathologies. After positioning a camera on a tripod 2 meters away from the subject, we collected three photos (front, back, and sagittal). We then analyzed the anatomical landmark three-dimensionally with an algorithm able to match the 3D position of the same landmark in both the front and back photo. We evaluated the joint angles, horizontal angles, vertical angles, and the lateral inclination of the neck and trunk. Results: The postural parameters obtained matched between the 3D parameters of the front and back images. They provided significant differences between males (m) and females (f) with a medium to large effect size for almost all the parameters. Concerning the joint angles, we observed a mean value of the shoulder angle of m= 16.78 ° ± 2.20 vs f= 13.58 ° ±1.55 (p < 0.001, d= 1.67), hip angle m=11.45 ° ± 2.05 vs f= 8.62 ° ± 1.73 (p < 0.001, d= 1.48), knee angle m= 2.75 ° ± 1.25 vs f=2.29 ° ± 1.01 (p= 0.064, d= 0.40). For the horizontal angles we evaluated the head line m=2.13 ° ± 1.25 vs f= 1.93 ° (0.73 ± p= 0.351, d= 0.20), shoulders line m= 1.41 ° (0.69 vs f= 1.55 ° ± 0.63 ± p= 0.316, d= -0.22), hips line m= 1.19 ° (0.53 vs f=1.13 ° ± 0.53 (p= 0.577, d= 0.12). For the vertical angles the body balance is m= 1.02 ° ± 0.48 vs f= 1.04 ° (0.45 ± p= 0.870, d= -0.03). Finally we evaluated also the neck inclination and the trunk inclination, with m= 13.80 ° ± 3.95 vs f=15.42 ° ± 3.61 (p= 0.052, d= -0.42) and m=2.76 ° ± 1.99 vs f=2.22 ° ± 1.83 (p= 0.204, d= 0.27) respectively. Discussion: The machine learning approach proved valuable in postural analysis, revealing gender differences in most parameters, but not in horizontal angles and body imbalance. The method requires no anatomical expertise, as the algorithm accurately identifies and measures key points. Future research will test its validity with pathological populations. Due to its simplicity, it could be used routinely by health professionals and as a tool to monitor rehabilitation or training progress.
Nr: 15
Title:

A Novel Approach for Optimal Exercise Intensity Assessment by Finger Pulse Wave During Cycle Ergometer Test in Healthy Young Men
Authors:

Makoto Ayabe
Abstract: The finger pulse wave threshold (FPT) is a new concept for detecting individual optimal exercise intensity. The products of the a-wave and pulse rate rapidly increased at the intensity corresponding to the VT during a graded exercise test (GXT), and named the the threshold intensity as the finger pulse wave threshold (FPT). The present study aimed to reveal the validity and reliability of the FPT in estimating the ventilatory threshold (VT) during GXT in young healthy men. METHODS. Ten healthy young men, aged from 20 to 26, participated in the present investigation. Two GXTs were conducted within 7 days to evaluate the reliability of the FPT. The GXT, using a bicycle ergometer, was cosistented by a 5-min resting period, a 2-min warm-up period at 10 watts, and the main exercise period which intensity was increased by 15 W/min in the ramp protocol. During the GXTs, finger pulse waves, heart rate (HR), systolic blood pressure (SBP), and expired gas were continuously obtained. The FPT was defined as the exercise intensity (watts) which the products of the a-wave height and pulse rate, thereafter HR cooresonding to the FPT was also calcurated. The VT was determined by the V-slope methods. The double product breakpoint (DPBP) was also determined as the criterion measurement for estimating VT. RESULTS. The watts and HR at FPT did not differ significantly from those at VT, and strong correlations were observed between the two parameters (r = 0.846 in WR, r = 0.888 in HR, p < 0.05). The deviation of FPT with VT was -0.8 ± 8.8 watts and -0.9 ± 5.4 bpm. Regarding the test-retest reliability of the FPT, a strong correlation was obtained (r = 0.878 in WR, r = 0.939 in HR, p < 0.05), and the deviation within two GXT was -0.01 ± 0.12 watts/kg and -1.0±4.6 bpm. The deviation from VT and the deviation within the two GXTs did not differ significantly between FPT and DPBP. CONCLUSIONS. The FPT has high reliability and good agreement with VT and DPBP during cycle-ergometer GXT in healthy young men. These findings suggest that the FPT may be a useful and simple method to determine VT. Thus, the FPT is a valid, simple assessment of the optimal exercise intensity for exercise prescription.
Nr: 45
Title:

Simulation of Head Impact Events During Rugby and the Evaluation of Protection Afforded by a Foam Headguard
Authors:

D. MacManus, L P. Del Olmo, A. Elliott, T. Kechadi, B. Caulfield, A. NíAnnaidh and M. D. Gilchrist
Abstract: This preliminary study involves the simulation of a small pilot dataset of five real-world head impact events in rugby to assess the level of protection provided by a novel energy absorbing foam headguard, i.e., the N-Pro headguard [1]. The hypothesis behind this investigation is that a suitably designed headguard will attenuate accelerations and indicators of mTBI sustained by a rugby player if their head should impact the ground. A set of professional rugby head impact events that involved various unprotected head-ground impact scenarios was established. The impact kinematics were obtained from two sources: broadcast video footage of match-related impacts and real-time data obtained through players using instrumented mouthguards. These were reconstructed using three-dimensional finite element models, with and without a headguard and accounting for friction at the ground-head and head-helmet interfaces. The UCD Brain Trauma Models [2,3] were used to reconstruct these actual impacts, and hypothetical equivalent protected impacts that involved the use of headguards. Linear and angular accelerations, and stress/strain levels within the brain were quantified while wearing or not wearing a headguard in each situation. All simulations were performed against a rigid, non-compliant surface to represent impact against either the ground. In all cases, the level of acceleration reduced when the headguard was worn. The reduction varied between 60-90%, with one specific case corresponding to when the greatest level of protection was afforded by wearing a headguard, i.e., a back-of-head Vs ground impact scenario. Von Mises stress and maximum principal strain for this case were also reduced by over 60%. This set of 5 reconstructed rugby tackle impact events confirm that a headguard can indeed provide a clear and quantifiable level of head protection against injury if worn while playing a contact sport such as rugby. Depending on the particular ground impact situation sustained by a person, and the ground compliance, the level of attenuation that was predicted to be associated with wearing a headguard can be as much as 90% less than what would be sustained if a headguard had not been worn. The results obtained from this preliminary study demonstrate the significant potential of the N-Pro headguard in reducing peak head kinematics and brain tissue responses compared to unprotected heads. This highlights the N-Pro's potential in reducing concussion incidence and injury severity in contact sports such as rugby. Additionally, the study supports the recommendation in current literature that kinematic data collected from wearable sensors should be supplemented by video analysis to improve accident reconstruction. Future planned work will investigate a large prospective set of such impacts and complement mouthguard and video data with comprehensive clinical data for each rugby player over a full playing season. References: [1] Ganly and McMahon (2018) New generation of headgear for rugby: Impact reduction of linear and rotational forces by a viscoelastic material-based rugby head guard. BMJ Open Sport & Exercise Medicine, 4(1),000464. [2] Horgan and Gilchrist (2003) The creation of three-dimensional finite element models for simulating head impact biomechanics, Int J Crashworthiness, 8(4),353–366. [3] Trotta et al. (2020) Biofidelic finite element modelling of brain trauma: Importance of the scalp in simulating head impact, Int J Mech Sci, 173,105448.
Nr: 56
Title:

The Opportunities of Using the Inertial Measurement Unit System on Older Adults’ Lower Extremity Functional Performance Evaluation
Authors:

Linda L. Lin, Chih-Yi Li and Tsai Hsuan Ho
Abstract: Purpose: To explore the trend and opportunities of using the Inertial Measurement Units (IMUs) to objectively assess the older adults' lower extremity functional fitness and risk of Sarcopenia in community-dwelling older adults. Methods: After the warm-up exercise, the 30-s chair standing test was conducted on 16 older adults (age 70.7±6.03 years) collecting the signals and angular position of the right thigh's vastus lateralis muscle, vastus medialis muscle, and rectus femoris muscle using a device incorporating triaxial accelerometers, gyroscopes, and magnetometers by the 3 wearable nine-axis inertial measurement units. There were 10 Kinematic parameters were collected, including BMI, Skeletal muscle mass index (SMI), variance of center of gravity change in preparation (PRE CG)(°), hip joint angle in preparation (PRE hip)(°), knee joint angle in preparation (PRE knee)(°), hip joints angle in standing (TO hip)(°), knee joints angle in standing (TO knee)(°), hip and knee joint overall average angular velocity from preparation to standing (PRE-TO OAAV hip and knee)(deg/sec), hip joint average angular velocity from preparation to standing (PRE-TO AAV hip)(deg/sec), knee joint average angular velocity from preparation to standing (PRE-TO AAV knee)(deg/sec). Pearson's correlation analysis was used to analyze the correlation between the IMU signals, SMI and the achievements and percentage of chair stand metrics. Furthermore, the multiple stepwise regression analysis predicted the key influencing factors of muscle fitness by the chair standing motion test. Results: The skeletal muscle index (kg/m^2), PRE CG(.012), PRE-TO OAAV hip and knee(.045), PRE-TO AAV hip(<.001), PRE-TO AAV knee(<.001) are significantly correlated with the achievements of the chair standing test. The major impact factors in predicting 30-s chair standing performance came from mean PRE knee and PRE-TO AAV knee(R2 = 0.884). However, weight and TO knee are key factors that affect muscle mass (R2=0.585). Conclusion: Using the inertial measurement unit system to test chair standing can be an objective assessment method for evaluating muscle strength of lower limbs for older adults, it is also considered as one of the factors for assessing the risk of Sarcopenia. Based on this study, the IMU sensors, data fusion, and data analysis techniques can record each muscle's status during training and convert it into data to help the trainer design the best way to improve older adults' lower extremity power performance and muscle mass.
Nr: 5
Title:

Differences in Online Cycling Performance Due to the Presence of Spectators and Personality Traits
Authors:

Muhammet Ali Metoglu and Utku Isik
Abstract: The aim of this study is to understand the effect of online spectators on online cycling performance.In addition, another aim of this research is to investigate how effective the personality traits of the participants are on the performance. Twenty-four students (Meanage=23.87+/-2.77) participated in this study. The participants performed two cycling, 4 weeks apart, on the Zwift (a multiplayer online cycle programme) The participants performed their first cycling alone, and their second cycling were performed with the online spectators connected to support the participant through a Google Meeting. During both cycling, some markers of exertion (heart rate,cadance,power,driving time,calories) were recorded. In addition, in order to understand whether the measure changes obtained by the participants during the two cycling vary according to their personality characteristics, the participants were given the "Five Factor Personality Inventory Short Form”. Two-factor ANOVA analysis was used for mixed measures of whether or not having the mentioned characteristics makes a difference. As a result of the analysis, it was seen that the presence of online spectators during online cycling of the subjects had a positive effect on the driving time, calories consumed, power applied to the bike, cadence, maximum cadence, heart rate, maximum heart rate, and the person's perception of effort. Significant differences were found both the power and cycling times of the participants who had high rates ​​in the conscientiousness sub-dimension.
Nr: 52
Title:

Estimation of Human Anthropometric Parameters Using Kinect v2 Depth Camera
Authors:

Krzysztof Przednowek, Élvio R. Gouveia and Tomasz Krzeszowski
Abstract: Anthropometric measurements of the human body are an important problem that affects many aspects of human life. However, anthropometric measurement often requires the application of an appropriate measurement procedure and the use of specialized, sometimes expensive measurement tools. Sometimes the measurement procedure is complicated, time-consuming, and requires properly trained personnel. This study aimed to develop a system for estimating human anthropometric parameters based on a three-dimensional scan of the complete body made with an inexpensive depth camera in the form of the Kinect v2 sensor. The research included 129 men aged 18 to 28. The men featured a weight at a level of 79.4±11.7 kg and a body height of 180.2±6.5 cm. All participants of the research gave their written consent to the anthropometric examination and consent to perform the 3D body scan. The developed system consists of a rotating platform, a depth sensor (Kinect v2), and a PC computer that was used to record 3D data, and to estimate individual anthropometric parameters. The 3D scan of the measured person is recorded using a rotating platform (the platform rotates by 360°with a constant speed), which allows a full 3D scan of the human body. The scanned person should stance in a T-pose and his clothing should be limited to a minimum (e.g. tight-fitting underwear). During scanning the sensor records multiple 3D scans that present the human body from different sides. To determine somatic parameters from a 3D scan (point cloud), was perform segmentation in order to separate individual body segments. With a segmented 3D scan, somatic features can be determined. To evaluate the method, a statistical analysis was carried out in the form of a U Mann-Whitney test and Bland-Altman charts. Experimental studies have shown that the accuracy of the proposed system for a significant part of the parameters is satisfactory (<7%). The largest error was in the waist circumference parameter. The results obtained confirm that the method can find application in anthropometric measurements. All the results of the conducted experiment were published in the paper: Krzeszowski, T., Dziadek, B., França, C., Martins, F., Gouveia, É. R., & Przednowek, K. (2023). System for Estimation of Human Anthropometric Parameters Based on Data from Kinect v2 Depth Camera. Sensors, 23(7), 3459. https://doi.org/10.3390/s23073459.

Area 3 - Sport Performance and Support Technology

Nr: 58
Title:

ML-Based Scene Classification Using Basketball Players’ Tracking Data for Team Performance Analysis
Authors:

Takeshi Tanaka, Takuya Magome and Norio Gouda
Abstract: Introduction Popular professional sports teams, such as football and basketball, have introduced advanced measurement technology to strengthen their teams. There is a need to develop data analysis methods to evaluate organizational performance instead of individual performance to innovate team training [1]. To create new organizational evaluation methods and hypotheses, it is necessary to collect and analyze a large number of scenes that lead to team scores and wins. However, the challenge has been to reduce the human cost of tagging subjective evaluations in the collection of data sorted as scenes of high team performance. In this study, we propose introducing a deep learning model that estimates the likelihood of scoring opportunities from the movement trajectory data of players and the ball, aiming to automate scene evaluation for tracking data. Methods Using a method for generating tagged data by extracting scenes that seem to be scoring opportunities in a deep learning model, we evaluated an analysis flow that accelerates the development of network indices and visualization methods using transfer entropy with the tagged data. In this evaluation, we considered using existing official tracking data of professional sports as the large amount of data required for training deep learning models. We applied the NBA's professional basketball tracking data, which is publicly available. The deep learning model used a CNN (Convolutional Neural Network) based on previous studies [2] to predict whether scoring or not by inputting a composite image of two groups of five players each from the attacking/defensive side and the ball movement trajectory as one data unit while the ball was being held. In evaluation, we evaluated the accuracy of a model for predicting scoring opportunities trained on basketball tracking data. Next, we applied our proposed method of analyzing team performance using transfer entropy to the scenes in which we predicted the likelihood of scoring opportunities and evaluated whether the results showed similar trends to the existing knowledge. Results The prediction accuracy of the deep learning model was AUC=0.927, exceeding the initially set goal. The transfer entropy between players was calculated and compared in scenes predicted as potential scoring opportunities and other scenes. The results showed that the transfer entropy of the attacking team from the opposing team was significantly higher in the predicted scoring scene than in the other scene. This result was similar to that obtained in previous studies using data subjectively tagged as a high organizational performance by experts. Conclusion The results suggest that deep learning models trained on existing tracking data can predict the likelihood of scoring opportunities and that the data can be used to accelerate the development of network analysis of transfer entropy. [1] Tanaka, Takeshi, et al., IPSJ Journal (2021) [2] Harmon, Mark, et al., International Journal of Sport and Health Sciences (2021)
Nr: 33
Title:

Reducing Cockpit Workload for Sporting Plane Pilots by an Automated Flight Protocol Smartphone App
Authors:

Hans Weghorn
Abstract: While steering sporting aeroplanes, pilots are required to use simultaneously both hands and even feet, since control of such machines requires multi-dimensional input. The first hand directs the stick for ailerons and elevator, both feet set the rudder, and the second hand circulates between primary instrument adjustments, power-level setting, flaps control, and handling secondary equipment like communication devices and other. Already when starting the aero engine, pilots are fully occupied with such manual actions, and on top of that they are furthermore required to protocol the relevant flight operation stages with time accuracy as it is prescribed by air law [1]. Commonly in sports flying, records are noted by hand writing, which theoretically has to be performed in especially critical phases of the flight operation, e.g., while the plane is just taking off from the runway. Another demanding manoeuvre in training represents a “touch and go”, where a landing approach is completed until the gear wheels touch the ground, and then the plane is powered fully for an immediate re-start. In this action sequence, pilots rarely do have good opportunity for noting the touch down time, usually there is not even time for looking at any clock display. Modern consumer technology offers with smartphones appropriate devices, that feasibly can trace flight movements with sufficient accuracy, since they are equipped with a broad range of suitable sensor features. In the research project described here, a smartphone app is under on-going development, which automatically records the required flight protocol information, so that pilots are relieved from this disturbing duty and can concentrate fully on steering their aerial vehicle. Several technologies and processing concepts had to be combined in this smartphone app, so that the protocol system works fully automated with minimal or even no input actions of the plane pilot. Flight Activity and its stages are traced and detected by GPS and audio signal sensing, specifically developed filter mechanisms are applied as pre-stage of signal analysis. Final decision logic is based on pattern recognition and Geo fencing. The control logic frame of the app is realised as finite state machine, which is needs, e.g., seven states, for the operational part till first take-off. A set of different, complementing key techniques is used for detection of transitions. At the moment, the plane type and the mission end are recognized by Geo fencing on hangar parking positions. Plane identification is essential for knowing the proper take-off and landing speeds. Only when using such parameters correctly, accurate starting and landing times can be differentiated as well as complicate actions like "touch and go". Although the protocol app is in use since longer time and has proven in hundreds of starts and landings its reliability - there were only few, single deviations from protocols taken manually, when air traffic controllers were available as ground service -, the functionality concepts shall be further enhanced. In future, pattern recognition on the engine and propeller sound shall support identifying the particular plane, and gravitational and acceleration smartphone sensors shall be exploited in addition to GPS input for improving the recognition reliability in more complicate flight manoeuvres. [1] EASA, "FCL.050 Recording of flight time" in: Annex I - Part FCL, V1, p29-39, June 2016.