Player Tracking Research Articles

Validation Of Player Tracking Sensors For Measuring Sprint Speed With Collegiate Ice Hockey Players

Validation Of Player Tracking Sensors For Measuring Sprint Speed With Collegiate Ice Hockey Players

Going deep: models for continuous-time within-play valuation of game outcomes in American football with tracking data

AbstractContinuous-time assessments of game outcomes in sports have become increasingly common in the last decade. In American football, only discrete-time estimates of play value were possible, since the most advanced public football datasets were recorded at the play-by-play level. While measures such as expected points and win probability are useful for evaluating football plays and game situations, there has been no research into how these values change throughout the course of a play. In this work, we make two main contributions: First, we introduce a general framework for continuous-time within-play valuation in the National Football League using player-tracking data. Our modular framework incorporates several modular sub-models, to easily incorporate recent work involving player tracking data in football. Second, we use a long short-term memory recurrent neural network to construct a ball-carrier model to estimate how many yards the ball-carrier is expected to gain from their current position, conditional on the locations and trajectories of the ball-carrier, their teammates and opponents. Additionally, we demonstrate an extension with conditional density estimation so that the expectation of any measure of play value can be calculated in continuous-time, which was never before possible at such a granular level.

Unsupervised methods for identifying pass coverage among defensive backs with NFL player tracking data

AbstractStatistical analysis of defensive players in football has lagged behind that of offensive players, special teams, and coaching decisions, largely because data on individual defensive players has historically been lacking. With the introduction of player tracking data from the NFL, researchers can now tackle these problems. However, event and strategy annotations in the NFL’s tracking data are limited, especially when it comes to describing what defensive players do on each play. Moreover, methods for creating these annotations typically require extensive human labeling, which is difficult and expensive. Because of the importance of the passing game and the limited prior research on the defensive side of passing, we provide annotations for the pass coverage types of cornerbacks using unsupervised learning techniques, which require no training data. We define a set of features from the tracking data that distinguish between “man” and “zone” coverage. We use mixture models to create clusters corresponding to each group, allowing us to provide probabilistic assignments to each coverage type (or cluster). Additionally, we quantify each feature’s influence in distinguishing defensive pass coverage types. Our work makes possible several potential avenues of future NFL research into defensive backs and pass coverage strategies.

Expected hypothetical completion probability

AbstractUsing high-resolution player tracking data made available by the National Football League (NFL) for their 2019 Big Data Bowl competition, we introduce the Expected Hypothetical Completion Probability (EHCP), a objective framework for evaluating plays. At the heart of EHCP is the question “on a given passing play, did the quarterback throw the pass to the receiver who was most likely to catch it?” To answer this question, we first built a Bayesian non-parametric catch probability model that automatically accounts for complex interactions between inputs like the receiver’s speed and distances to the ball and nearest defender. While building such a model is, in principle, straightforward, using it to reason about a hypothetical pass is challenging because many of the model inputs corresponding to a hypothetical are necessarily unobserved. To wit, it is impossible to observe how close an un-targeted receiver would be to his nearest defender had the pass been thrown to him instead of the receiver who was actually targeted. To overcome this fundamental difficulty, we propose imputing the unobservable inputs and averaging our model predictions across these imputations to derive EHCP. In this way, EHCP can track how the completion probability evolves for each receiver over the course of a play in a way that accounts for the uncertainty about missing inputs.

Bigger data, better questions, and a return to fourth down behavior: an introduction to a special issue on tracking datain the National football League

AbstractMost historical National Football League (NFL) analysis, both mainstream and academic, has relied on public, play-level data to generate team and player comparisons. Given the number of oft omitted variables that impact on-field results, such as play call, game situation, and opponent strength, findings tend to be more anecdotal than actionable. With the release of player tracking data, however, analysts can better ask and answer questions to isolate skill and strategy. In this article, we highlight the limitations of traditional analyses, and use a decades-old punching bag for analysts, fourth-down strategy, as a microcosm for why tracking data is needed. Specifically, we assert that, in absence of using the precise yardage needed for a first down, past findings supporting an aggressive fourth down strategy may have been overstated. Next, we synthesize recent work that comprises this special Journal of Quantitative Analysis in Sports issue into player tracking data in football. Finally, we conclude with some best practices and limitations regarding usage of this data. The release of player tracking data marks a transition for the league and its’ analysts, and we hope this issue helps guide innovation in football analytics for years to come.

Multiple player tracking in basketball court videos

To build a smart basketball court, one basic task is to track players with the aid of the basketball court monitoring. This task can be regarded as a special case of the multiple object tracking (MOT) problem. But different from it, the task puts request to the existing MOT methods with both good accuracy and operational efficiency (toward real time) in the new basketball scenario. To deal with this task, we make the following attempts: (1) Considering the differences between pedestrians and basketball players and the lack of corresponding dataset for basketball players tracking, we construct a new MOT dataset under the basketball court monitoring scene, to better evaluate MOT methods in our task and to help promote future research in the related task, (2) evaluating the performance of the several candidate MOT methods on the new dataset, and (3) proposing the issues to be further addressed in this specific scenario.

Individual Response to External Training Load in Elite Football Players.

To investigate within-player effect, between-player effect, and individual response of external training load from player tracking devices on session rating of perceived exertion training load (sRPE-TL) in elite football players. The authors collected sRPE-TL from 18 outfield players in 21 training sessions. Total distance, high-speed running distance (>14.4m/s), very high-speed running distance (>19.8m/s), PlayerLoad™, PlayerLoad2D™, and high-intensity events (HIE > 1.5, HIE > 2.5, and HIE > 3.5m/s) were extracted from the tracking devices. The authors modeled within-player and between-player effects of single external load variables on sRPE-TL, and multiple levels of variability, using a linear mixed model. The effect of 2 SDs of external load on sRPE-TL was evaluated with magnitude-based inferences. Total distance, PlayerLoad™, PlayerLoad2D™, and HIE > 1.5 had most likely substantial within-player effects on sRPE-TL (100%-106%, very large effect sizes). Moreover, the authors observed likely substantial between-player effects (12%-19%, small to moderate effect sizes) from the majority of the external load variables and likely to very likely substantial individual responses of PlayerLoad™, high-speed running distance, very high-speed running distance, and HIE > 1.5 (19%-30% coefficient of variation, moderate to large effect sizes). Finally, sRPE-TL showed large to very large between-session variability with all external load variables. External load variables with low intensity-thresholds had the strongest relationship with sRPE-TL. Furthermore, the between-player effect of external load and the individual response to external load advocate for monitoring sRPE-TL in addition to external load. Finally, the large between-session variability in sRPE-TL demonstrates that substantial amounts of sRPE-TL in training sessions are not explained by single external load variables.

Detection of Ice Hockey Players and Teams via a Two-Phase Cascaded CNN Model

The accurate detection of ice hockey players and teams during a game is crucial to the tracking of individual players on the rink and team tactical decision making and is therefore becoming an important task for coaches and other analysts. However, hockey is a fluid sport due to its complex situation and the frequent substitutions by both teams, resulting in the players taking various postures during a game. Few player detection models from basketball and soccer take these characteristics into account, especially for team detection without prior annotations. Here, a two-phase cascaded convolutional neural network (CNN) model is designed for the detection of individual ice hockey players, and the jersey color of the detected players is extracted to further identify team affiliations. Our model filters most of the disturbing information, such as the audience and sideline advertising bars, in Phase I and refines the detection of the targeted players in Phase II, resulting in an accurate detection with a precision of 98.75% and a recall of 94.11% for individual players and an average accuracy of 93.05% for team classification with a self-built dataset of collected images from the 2018 Winter Olympics. The results for the regular season games of the 2019-2020 National Hockey League (NHL) covering all 31 teams are also presented to show the robustness of our model. Compared to state-of-the-art approaches, our player detection model achieves the highest accuracy with the self-built dataset.

Alterations in Team Physical Performance and Possession in Elite Gaelic Football Competition

Differences in performance between winning and losing were examined in 1 elite Gaelic football team in 20 games across 2 complete competitive seasons. Possession was codified using Dartfish TeamPro software and distance covered; walking, jogging, running, and running at high and maximum speeds, was evaluated using Catapult OptimEye S5 player tracking devices. Distance covered in low intensity activity (LIA, ˂4.0 m.s-1), high intensity running (HIR, ≥4.0 m•s-1) and very high intensity running (VHIR, ≥5.5 m•s-1) was also examined along with PlayerLoad™, which represented a composite of all accelerations. Data from 53 players (n=405 files) was collated into specific match periods to facilitate a temporal analysis between the first and second halves and from quarter 1 (Q1) to quarter 4 (Q4), with significance accepted at p ≤ 0.05. Total distance and running was higher in games lost, whereas total distance, walking and LIA was higher in halves lost. Only walking was higher in quarters lost. The percentage of possession declined in halves and quarters lost. In games lost, high speed running declined in the second half. From Q1 to Q4; PlayerLoad™, total distance, jogging, high speed running, HIR and VHIR, decreased in all games combined and in games lost. Possession frequency declined in Q4 in all games and in games won. Overall, total distance was higher in games lost and physical performance declines were more pronounced when examined by match quarter compared to half and were only apparent in games lost. Similarly, reductions in possession frequency and percentage were more evident when examined by quarter or period lost, respectively. These findings can inform the prescription of conditioning and field-training strategies to mitigate the reductions in performance observed in losing and towards the end of games.

Use of Machine Learning to Automate the Identification of Basketball Strategies Using Whole Team Player Tracking Data

The use of machine learning to identify and classify offensive and defensive strategies in team sports through spatio-temporal tracking data has received significant interest recently in the literature and the global sport industry. This paper focuses on data-driven defensive strategy learning in basketball. Most research to date on basketball strategy learning has focused on offensive effectiveness and is based on the interaction between the on-ball player and principle on-ball defender, thereby ignoring the contribution of the remaining players. Furthermore, most sports analytical systems that provide play-by-play data is heavily biased towards offensive metrics such as passes, dribbles, and shots. The aim of the current study was to use machine learning to classify the different defensive strategies basketball players adopt when deviating from their initial defensive action. An analytical model was developed to recognise the one-on-one (matched) relationships of the players, which is utilised to automatically identify any change of defensive strategy. A classification model is developed based on a player and ball tracking dataset from National Basketball Association (NBA) game play to classify the adopted defensive strategy against pick-and-roll play. The methodology described is the first to analyse the defensive strategy of all in-game players (both on-ball players and off-ball players). The cross-validation results indicate that the proposed technique for automatic defensive strategy identification can achieve up to 69% accuracy of classification. Machine learning techniques, such as the one adopted here, have the potential to enable a deeper understanding of player decision making and defensive game strategies in basketball and other sports, by leveraging the player and ball tracking data.

Effect of Training Load on Post-Exercise Cardiac Troponin T Elevations in Young Soccer Players.

Training load (TL) metrics are usually assessed to estimate the individual, physiological and psychological, acute, and adaptive responses to training. Cardiac troponins (cTn) reflect myocardial damage and are routinely analyzed for the clinical diagnosis of myocardial injury. The association between TL and post-exercise cTn elevations is scarcely investigated in young athletes, especially after playing common team sports such as soccer. The objective of this study was to assess the relationship between TL measurements during a small-sided soccer game and the subsequent increase in cTn in young players. Twenty male soccer players (age 11.9 ± 2 years, height 151 ± 13 cm, weight 43 ± 13 kg) were monitored during a 5 × 5 small-sided game and had blood samples drawn before, immediately after, and 3 h after exercise for a posterior analysis of high-sensitivity cardiac troponin T (hs-cTnT). Internal, external, and mixed metrics of TL were obtained from the rating of perceived exertion (RPE), heart rate (HR), and GPS player tracking. The results show that the concentration of hs-cTnT peaked at 3 h post-exercise in all participants. The magnitude of hs-cTnT elevation was mainly explained by the exercise duration in the maximal heart rate zone (Maximum Probability of Effect (MPE) = 92.5%), time in the high-speed zone (MPE = 90.4 %), and distance in the high-speed zone (MPE = 90.45%). Our results support the idea that common metrics of TL in soccer, easily obtained using player tracking systems, are strongly associated with the release of hs-cTnT in children and adolescents.

Blood-borne fatigue markers during major international football tournaments – a retrospective analysis of data from the FIFA World Championships and UEFA European Championships 2006 – 2016

ABSTRACTPurpose:Monitoring of recovery needs is considered essential during international football tournaments, with blood-borne markers being an objective, well understood and conveniently measurable option. However, reference values from this level of play are largely lacking. Methods:We therefore report a retrospective analysis of the blood-borne fatigue markers creatine kinase (CK) and urea (U) from the German national football team during the FIFA World Championships and UEFA European Championships from 2006–2016. Values from 68 players could be analysed resulting in 1019 values for CK and 943 values for U. For 2014 and 2016, additional player tracking data were available. Results:Mean values (90% CI) for CK and U were 343.4 (333.3–353.6) U/l and 39.5 (39.0–39.9) mg/dl, respectively. Considerable between-subject variation was present accounting for 44% of random variability for logCK and 45% for logU. A significant decrease over time (p < 0.001) could be observed for both indicators. Total running distance and high-speed running were significantly correlated with CK levels 2 days after the match with a disproportionate CK response for overtime matches. Conclusion:Taken together, monitoring of recovery needs in elite football is likely to benefit from individualized reference values. When group-based reference ranges are used, they should be current and football-specific.

WITHDRAWN: Moving target tracking algorithm and trajectory generation based on Kalman filter in sports video

Abstract This paper takes the video of volleyball game collected by fixed camera as an example to study the moving targets in the volleyball video sequence, complete the prediction and tracking of volleyball and players, and trace the depiction. For moving objects tracking, firstly, the ID and location information on the objects blob is received through image segmentation. Secondly, tracking algorithm is a combination of centroid tracking algorithm and Kalman filter. The algorithm can provide players and volleyball object tracking. In the process of tracking, Kalman filter is imposed to forecast the motion information to improve the tracking precision. On the basis of the moving object tracking, this paper takes path generating the module to describe players tracks and preserve motion information. Based on the principle of double camera measure to reconstruct 3D information by two images taken from different angles at the same time, 3D information is used to match the volleyball trajectory. Experimental results show that the rate of accuracy of moving target tracking algorithm higher than 92%, the algorithm can realize the real-time and stable while ensuring the accuracy, providing the necessary basis for accurately video analysis.

A machine learning approach for automatic detection and classification of changes of direction from player tracking data in professional tennis

ABSTRACTThe purpose of this study was to develop an automated method for identifying and classifying change of direction (COD) movements in professional tennis using tracking data. Three sport science and strength and conditioning experts coded match-play footage of nineteen professional tennis players (9 male and 10 female) from the Australian Open Grand Slam for COD of medium and high intensity. A total of 1,494 changes were identified and aligned with 2D player position sampled at 25 Hz based on camera tracking data. Several machine learning classifiers were trained and tested on a set of 1,128 time-motion features. A random forest algorithm was found to have the best out-of-sample performance, classifying medium and high intensity changes with an F1-score of 0.729. This research offers a novel and applicable way for utilising player tracking data and machine learning techniques to automatically identify and classify COD movements in professional tennis.

Modeling the Quality of Player Passing Decisions in Australian Rules Football Relative to Risk, Reward, and Commitment

The value of player decisions has typically been measured by changes in possession expectations, rather than relative to the value of a player’s alternative options. This study presents a mathematical approach to the measurement of passing decisions of Australian Rules footballers that considers the risk and reward of passing options. A new method for quantifying a player’s spatial influence is demonstrated through a process called commitment modeling, in which the bounds and density of a player’s motion model are fit on empirical commitment to contests, producing a continuous representation of a team’s spatial ownership. This process involves combining the probability density functions of contests that a player committed to, and those they did not. Spatiotemporal player tracking data was collected for AFL matches played at a single stadium in the 2017 and 2018 seasons. It was discovered that the probability of a player committing to a contest decreases as a function of their velocity and of the ball’s time-to-point. Furthermore, the peak density of player commitment probabilities is at a greater distance in front of a player the faster they are moving, while their ability to participate in contests requiring re-orientation diminishes at higher velocities. Analysis of passing decisions revealed that, for passes resulting in a mark, opposition pressure is bimodal, with peaks at spatial dominance equivalent to no pressure and to a one-on-one contest. Density of passing distance peaks at 17.3 m, marginally longer than the minimum distance of a legal mark (15 m). Conversely, the model presented in this study identifies long-range options as have higher associated decision-making values, however a lack of passes in these ranges may be indicative of differing tactical behavior or a difficulty in identifying long-range options.

One Image Segmentation and Discrimination Method of Live Video System Applied in Sport Game

Background: In the field of video image processing, the segmentation and tracking method has become a hot research field. According to the characteristics of soccer video and VAR (Video Assistant Referee) System, in the paper, it proposes a fast algorithm of segmentation and tracking of the players in the video. Methods: First, according to complementary advantages on the color expression in RGB (Red, Green, Blue) space and HSI (Hue, Saturation, Intensity) space, it adopts the method based on combining the main and auxiliary space to make segmentation of the objects in the video. Then, it compares with the normalized color histogram of targets to make identification of which team the players belong to. Results: Finally, it adopts the context features of the players in the soccer video sequence; it combines the method of template matching to realize the player tracking in the soccer video. Conclusion: The experimental results show that the tracking algorithm proposed in this paper can solve the objects occlusion problem of the different players in the soccer match, and can track the different players steadily, and it also can be adopted in the VAR system.

External Load Variables Affect Recovery Markers up to 72 h After Semiprofessional Football Matches

Background: Player tracking devices are commonly used to monitor external load from training and matches in team sports. Yet, how the derived external load variables relate to fatigue and recovery post-training or post-match is scarcely researched. The objective was, therefore, to investigate how external load variables affect recovery markers up to 72 h post-match.Methods: Semiprofessional players from six teams wore tracking devices during three experimental football matches. External load variables including individual playing duration, total distance, PlayerLoad™, high-intensity running, and high-intensity events were derived from the tracking devices, and blood samples and performance tests from 24–59 players were undertaken post-match. The effect of the external load variables on creatine kinase, myoglobin, and countermovement jump at 1, 24, 48, and 72 h, and 30-m sprint and Yo-Yo intermittent recovery tests level 1 at 72 h post-match, were modeled. Effects were gauged as two standard deviations of the external load and interpreted as the difference between a typical high-load and a typical low-load match. The effects were evaluated with 90% confidence intervals and magnitude-based inferences.Results: High-intensity running had very likely substantial effects on creatine kinase and myoglobin (moderate factor increases of 1.5–2.0 and 1.3–1.6 respectively), while duration, total distance, and HIE showed small, likely substantial effects. PlayerLoad™ and total distance had likely substantial effects on 30-m sprint time (small increases of 2.1–2.6%). Effects on countermovement jump performance were generally non-substantial. Despite these relationships, the uncertainty was too large to predict the recovery of individual players from the external load variables.Conclusions: This study provides evidence that external load variables have an effect on recovery markers up to 72 h post-match. Hence, tracking external load in matches may be helpful for practitioners when managing training load and recovery strategies post-match. However, it is recommended that several different external load variables are monitored. Future research should continue to address the problem of predicting recovery from external load variables.

NCAA Preseason Demonstrates Greatest Impact on Heart Rate Variability, Training Load and Sleep In Men's Soccer

PURPOSE: The purpose of this study was to compare training load (TL), sleep, heart rate variability (HRV) and resting (RHR) by season phase over a full NCAA collegiate men’s soccer season. METHODS:Daily TLs (duration [min], distance covered [m], time spent in heart rate >80 % of the individual’s max [min], number of sprints [n], and distance covered while running >14.4 km·h-1 [m]) were monitored in 31 male NCAA Division 1 male soccer athletes (mean ± SD; age, 20 ± 2 y; body mass, 79.92 ± 6.69 kg; height, 181.48±6.35 cm; VO2max, 50.87 ± 4.38 ml·kg-1·min-1) using GPS-enabled player tracking devices during the Fall 2016 soccer preseason (PRE), regular season (REG), and post-season (POST). Sleep duration (min), sleep efficiency (%), and RHR and HRV z-scores (SD) were tracked using a multi-sensor wrist-worn deviceand a validated self-reported sleep questionnaire (Karolinksa Sleep Diary). Linear mixed effects models with a fixed factor of season phase and a random factor of participant was used. Mean differences (MD) were assessed post-hoc with a Tukey HSD, with alpha set at 0.05 for all analysis. RESULTS: During PRE, participants experienced an increased distance (1527 ± 177 m), time (31 ± 3 min) and sprints (8 ± 1) (all p < 0.001) per session vs. REG, and increased distance (1793 ± 312 m), time (39 ± 6 min) and sprints (9 ± 1) per session vs. POST (all p < 0.001). A decrease in sleep efficiency was observed during PRE vs. REG (MD = -2.00 ± 0.05 %; p = 0.004). Self-reported sleep duration during REG was longer vs. PRE (MD = 45 ± 8 min; p < 0.001) and POST (MD = 32 ± 12 min; p = 0.01). No differences were observed for objective sleep duration or RHR at any point throughout the season (p > 0.05). RHR was found to be lower during PRE vs. REG (MD = -0.16 ± 0.06S D; p = 0.02). Finally, HRV during REG was reduced vs. PRE (MD = -0.14 ± 0.06 SD; p = 0.05). CONCLUSION: This study indicates higher HRV, greater physical and physiological loading, decreased self-reported sleep and decreased sleep efficiency during PRE compared to REG. In this men’s soccer team, PRE training was associated with significantly increased physical stress, adverse sleep characteristics and increased HRV. These responses should be considered when designing and implementing optimal training and recovery strategies.

Efficient Large Scale Frequent Itemset Mining with Hybrid Partitioning Approach

In today’s world, voluminous data are available which are generated from various sources in various forms. Mining or analyzing this large scale data in an efficient way so as to make them useful for the mankind is difficult with the existing approaches. Frequent itemset mining is one such technique used for analyzing in many fields like finance, health care system where the main focus is gathering frequent patterns and grouping them to be meaningful inorder to gather useful insights from the data. Some major applications include customer segmentation in marketing, shopping cart analyses, management relationship, web usage mining, player tracking and so on. Many parallel algorithms, like Dist-Eclat Algorithm, Big FIM algorithm are available to perform large scale Frequent itemset mining. In Dist-Eclat algorithm, datasets are partitioned using Round Robin technique which uses a hybrid partitioning approach, which can improve the overall efficiency of the system. The system works as follows: Initially the data collected are distributed by mapreduce. Then the local frequent k-itmesets are computed using FP-Tree and sent to the map phase. Later the mining results are combined to the center node. Finally, global frequent itemsets are gathered by mapreduce. The proposed system is expected to improve in efficiency by using hybrid partitioning approach in the datasets based on the identification of frequent items.

Automatic event detection in basketball using HMM with energy based defensive assignment

Abstract We propose a unsupervised learning framework for automatically labeling events in a basketball game. Our framework uses the the optical player tracking data in the NBA. We first learn the time series of defensive assignments using a novel player and location dependent attraction based model which uses hidden Markov models (HMMs), Gaussian processes, and a “bond breaking” model for changes in defensive assignments. Next, we use the learned defensive assignments as an input to a set of HMMs that automatically detect events such as ball screens, drives and post-ups. We show that our models provide significant improvements over existing benchmarks both on defensive assignments and event detection.