From the Lab to the Field: Controlled and Chaotic Environments!?

Written by Eran Amit, Shai Rosenblit, Amir Zviran, Moran Gad, Steve Barrett

Head of Algorithms, Data Scientist, System Engineer, Chief Technology Officer, Director of Sport Science and Research Innovation, Playermaker

Practitioners and scientists have been investigating the use of different methods of training and analysis in order to improve their athletes performance, or reduce their risk of injury. Within a lab environment, alterations within biomechanical variables (gait analysis) have been associated with the predisposition of increased injury incidence. 

One of these alterations is an increase in an individual’s ground contact time during the gait cycle, at the same velocity. Using Playermaker and the development of such metrics, it becomes possible to take the lab to the field and measure gait metrics in both environments. However, in order to assess changes within these variables, specific activities have been recommended in order to ascertain if players have experienced a change[1][2].

For example, within multiple sports, maximal and sub-maximal exercise tests are performed to mimic the same activity to assess difference. This can also be observed within drills of similar activity over a number of sets. A drill of this type is linear aerobic endurance intervals for distance, in which it can be identified how the player alters his gait between the sets. In Figure 1 below, there was a change in over 13% within the individual player’s contact duration asymmetry (left vs. right), with minimal difference in the distances the player was covering. This can potentially help us identify if and when players become at increased risk of injury, while identifying areas of development for the players to help improve their physical performance. 

Removing the constraint of performing such measurements only in the lab is a big step forward. The next challenge is removing the requirement of a specific, designated drill in order to extract the desired information. Ideally, players should spend all available training time on improving physical/ technical/ tactical elements of their performance rather than on drills or test batteries which are designed for specific data collection. Moreover, if it was possible, the same information should be extracted during training and matches in which the player cannot perform designated drills.

As an example of the type of challenge we are looking to overcome, we look at the finding by Small et. al.[3]. They reported that during soccer specific activities the gait profile of players altered, suggested as a consequence of ‘fatigue’. These changes in the gait profile may put the athletes at higher risk for specific injuries, suggested in the paper as increased risk of hamstring injuries during sprinting activities[3][4]. The experimental setup included multiple cameras and the athletes performed repetitions of a specific drill. In order to use this finding to reduce players’ injury risk the gait pattern parameters should be measured every time a player performs high intensity activity, during either training or matches.  

In order to develop a methodology that extracts gait profile parameters with minimal or no constraints on the athlete activity,  we chose to look for indicators of fatigue during a training session. More specifically, whether or not the player needs a longer ground contact time in order to run at the same speed at the beginning of the session versus the latter part of the session.

To start investigating the ground contact time behaviour outside the lab data was taken from 31 matches and opposed sessions. The contact time versus speed was plotted for all ~2.5M steps of 131 players (Figure 2). Using a logarithmic scale, brighter colours (yellow) indicates a higher frequency of the contact time and speed combination. 

Two distinctive clusters can be seen: The running cluster at speeds higher than ~2m/s and the walking cluster at speeds lower than ~2m/s and contact time longer than 0.4s. While there are a lot of data points that can be analysed, we need to consider the potential mechanisms associated with why these clusters may occur. Since different motion types require different contact times (accelerations, decelerations, change of directions, jumps etc) in this study we focused only on forward running at approximately constant speed (up to 20% variation in speed between consecutive leg strides).

For this ‘fatigue’ study we investigated a single training session. It started with various drills (opposed and unopposed) followed by a 50 minutes match. It was divided into two periods: The first period contains the training drills and first half of the match while the second period is the second half of the match. In total 18 player records were analyzed. Results of one of the players were removed due to a very high uncertainty level caused by insufficient number of high velocity sprints. 

The relevant steps of a specific player are shown in Figure 3. Contact time vs. speed is plotted for both early and later session periods as red and black marks, respectively. The essential characteristics of each session period can be described using a simple formula which is plotted in solid lines.

The players’ contact times are longer during the second part of the game. At speeds higher than 5m/s the differences start to become greater, similar to the differences observed for changes in sprint mechanics[3]. But how can this effect be quantified?

As expected the data looks very noisy- because this is a real match and not a controlled clean lab experiment. As a result looking at a specific speed will give a very narrow perspective with a high error. We therefore fitted the entire data range to a single representative function. The first function coefficient represents the contact duration at 6.5m/s (sprint; the threshold used by the team) while the second coefficient represents how the contact time decays as a function of speed. Since these coefficients are extracted from the entire speed range the player does not have to actually run at a speed of 6.5m/s.

In Figure 4 the predicted player contact times at a speed of 6.5m/s are presented. Each dot shows the extracted contact time at the latter vs. early session period. The errors are estimated using statistical analysis. The dashed line shows where all points will fall if there is no change in the contact time between periods. Points above it have longer contact time at the latter period and vice versa. For 6 out of the 17 players there is significant change in the contact time between the periods, for 5 players the contact time becomes bigger at the later session period. This finding is consistent with the fatigue effect reported in the literature.


Gait parameters traditionally measured only in lab environments can now be reported within multiple environments using Playermaker. Extracting contact time based metrics from uncontrolled environments such as matches may provide insights into players’ response to soccer specific fatigue, when velocity is accounted for. This information may help practitioners and scientists understand their players’ gait alterations, allowing them to intervene and support their performance while potentially reducing the risk of injury. 


[1] A Turner et. al., “A Testing Battery for the Assessment of Fitness in Soccer Players“,  Strength and Conditioning Journal (2011), doi: 10.1519/SSC.0b013e31822fc80a

[2] S Barrett et. al., “Elite-youth and university-level versions of SAFT90 simulate the internal and external loads of competitive soccer”, In: Science and Football VII. H. Nunome, B. Drust, and B. Dawson, eds. London, United Kingdom: Routledge, 2013. 

[3] K Small et. al., “Soccer Fatigue, Sprinting and Hamstring Injury Risk”, Int J Sports Med (2009), DOI: 10.1055/s-0029-1202822 

[4] Joke Schuermans, “Deviating Running Kinematics and Hamstring Injury Susceptibility in Male Soccer Players: Cause or Consequence?”, Gait Posture (2017),