Normative In-Game Data for Collegiate Baseball Pitchers Using Markerless Tracking Technology

Abstract

Background:

The current standard for motion capture data collection in baseball biomechanics is marker-based optical motion capture. Recent advancement in markerless motion capture capabilities has greatly improved accessibility to in-game, high-precision motion capture data, but specific values may differ from markered systems, necessitating separate normative values. For future data comparison, reference data are needed.

Purpose:

To describe common kinematic variables in baseball pitching using an in-game markerless motion capture system.

Study Design:

Descriptive laboratory study.

Methods:

Kinematic data were collected in 2 collegiate stadiums in the National Collegiate Athletic Association Division I Southeastern Conference using 8 synchronized 300-Hz KinaTrax cameras placed around the playing field. The in-game biomechanics data of 51 pitchers from 5 different teams during the 2023 season were collected; only pitchers with >3 outings during the season were included. After averaging all available fastballs thrown in the first inning of a game, we analyzed triaxial trunk (rotation, flexion, and lean) and pelvic rotation angles in the global reference frame, as well as shoulder rotation, shoulder horizontal abduction, shoulder abduction, elbow flexion, stride knee flexion, and hip-shoulder separation.

Results:

A total of 509 fastballs were analyzed. Mean fastball velocity was 40.98 m/s (91.5 mph), with a vertical break of 42.0 ± 10.6 cm (16.5 ± 4.2 inches) and a horizontal break of 28.0 ± 11.1 cm (11.0 ± 4.4 inches). Mean stride length was 1.41 ± 0.08 m. The mean arm slot was 59.4°± 9.3°, and the mean upper arm slot was 72.9°± 9.3° at ball release. Additional normative time-series data are presented for commonly analyzed variables, and discrete metrics are provided at distinct time points in the pitching cycle.

Conclusion:

These data fill a need for separate norms for in-game, markerless motion capture performance metrics, and biomechanics for collegiate baseball pitchers to be used as reference values for researchers, coaches, and clinicians.

Clinical Relevance:

Clinicians should use these results as reference values to contextualize injury mechanisms and the injury-performance trade-off. These data will also educate clinicians on what athletes’ bodies must do to perform when constructing plans of care and return-to-play timelines.

Keywords

descriptive data baseball pitching KinaTrax markerless motion capture

As research popularity has increased in baseball biomechanics, technology has advanced to meet the demands of researchers and provide greater volumes of data.^9,12,16 The current standard for motion capture data collection is marker-based optical motion capture.²¹ While great progress is attributable to marker-based motion capture, the process is not without pitfalls.¹⁶ Donning reflective markers is burdensome and requires participants to be largely undressed so cameras have lines of sight to the markers and to minimize motion artifact.^4,13 Markered systems are restricted to controlled settings and cannot be used for in-competition analysis. Further, during motions that produce high velocity, longitudinal segment rotation, such as in baseball pitching, marker-based systems are highly susceptible to skin artifact.² Therefore, recent technological advancements have sparked interest and increased the popularity of markerless systems in sport assessment.^6,16,20

Three-dimensional markerless motion capture systems use high-resolution images from synchronized cameras to produce models of human movement. Because markerless technology is newer, it is logically compared with current gold standard marker-based systems when being validated.^3,6,7,11,17 However, due to inherent differences and the recognition that marker-based systems are not without error, solely comparing markerless systems with marker-based systems may be unfair.^4,13 Only recently have reliability metrics for markerless systems entered the literature.^8,15 Because of these differences, Fleisig et al⁶ recommend that markerless systems may be used for baseball pitching motion capture but that values should not be compared across systems. Therefore, it is necessary to establish normative data using data from a variety of markerless motion capture systems to compare for future analyses.

The purpose of this study was to provide the scientific community with reference data for commonly used kinematic variables in baseball pitching using in-game, markerless motion capture data from National Collegiate Athletic Association (NCAA) baseball pitchers.

Methods

This research study received the universities’ institutional review board exemption approval. In-game, markerless motion capture was collected at 2 NCAA Division I, Southeastern Conference (SEC) baseball stadiums on 51 pitchers (12 freshmen, 3 redshirt freshmen, 12 sophomores, 1 redshirt sophomore, 12 juniors, 3 redshirt juniors, 4 seniors, and 4 graduates [“redshirt” indicates players who did not compete during their first year in favor of development, meaning they are typically 1 year older than their grade year would indicate]; height of 1.89 ± 0.66 m and mass of 95.3 ± 8.5 kg 32 right-handed pitchers and 19 left-handed pitchers) from 5 different SEC teams during the 2023 baseball season. Pitchers were included in the analysis if they played for a SEC team, had >3 outings of the season, and threw ≥4 four-seam fastballs in the respective outing analyzed. If multiple innings were thrown in an outing, data were analyzed from the first inning of the game in which the pitcher appeared. Within that inning, all fastball pitches were averaged for analysis. We chose to include pitchers with ≥3 outings to ensure that they were regularly participating in competition and to avoid potential early-season performance differences (due to psychological factors of first competition, for example). Pitchers were excluded if they threw in an underhand/submarine manner, as data from these pitchers would not be generalizable to the majority of the applicable population.¹

Each stadium has an 8-camera markerless KinaTrax motion capture system setup to collect pitching data. The cameras are permanently mounted around the playing field. Kinematic and temporal data were processed through KinaTrax software and synchronized with TrackMan pitch performance data. All cameras captured movement at 300 Hz for full resolution. Computed musculoskeletal metrics, which were consistent with International Society of Biomechanics recommendations,^22,23 were filtered using a second-order, low-pass variable filter with the trunk and pelvis at 10 Hz, legs at 6 Hz, and arms at 20 Hz, as set by KinaTrax. Custom MATLAB (MathWorks) algorithms were used to extract discrete variables from KinaTrax data, and time-series data were normalized to 100% of the pitch cycle. The pitch cycle was considered from stride-foot contact to maximum shoulder internal rotation.

Variables of interest included triaxial trunk (rotation, flexion, and lean) and pelvic rotation angles, in the global reference frame, shoulder rotation, shoulder horizontal abduction, and shoulder abduction, elbow flexion, stride knee flexion, and hip-shoulder separation, defined as the difference in global axial rotation between the thorax and pelvis. All variables were consistently defined with accepted recommendations.^22,23 Each of these variables was extracted from the following time points: (1) peak knee height, (2) stride-foot contact, (3) maximum shoulder external rotation, (4) ball release, and (5) maximum shoulder internal rotation.⁵ All events were defined by KinaTrax using proprietary event detection codes, where foot contact was defined as the first frame in which any part of the lead foot contacted the ground; this was determined by a deep learning neural network trained on each stadium environment that took into account foot position and the slope of the mound. Additionally, stride length was measured, defined as the distance between the 2 ankles at stride-foot contact. Arm slot was defined as the inclination between the throwing hand and throwing shoulder at ball release from a vertical axis, and upper arm slot was defined as the upper arm inclination from the vertical axis at ball release. The variables above were extracted from each individual pitch and then averaged within each pitcher.

Pitch velocity, vertical break, and horizontal break were measured using TrackMan. Vertical break was defined as the vertical distance between the ball and a trajectory of the released ball at home plate had gravity been the only force acting on it, with positive values indicating increased vertical height from the trajectory. Horizontal break was defined as the horizontal distance between the ball at home plate and a trajectory of the released ball with no horizontal deviation, with positive values indicating movement toward the throwing-arm side.

Results

A total of 509 fastballs were analyzed from the 51 pitchers included in the analysis. The mean ball velocity was 40.98 ± 1.17 m/s (91.52 ± 2.61 mph), with a mean vertical break of 42.0 ± 10.6 cm (16.5 ± 4.2 inches) and a horizontal break of 28.0 ± 11.1 cm (11.0 ± 4.4 inches). The mean stride length was 1.41 ± 0.08 m. The mean arm slot was 59.4°± 9.3° and the mean upper arm slot was 72.9°± 9.3° at ball release. Descriptive statistics for each kinematic parameter at each event are displayed in Table 1. Figures 1 through 10 show grand mean and individual mean time-series plots.

Table 1

Variables of Interest at Each Time Point^a

Variable^b	Time Point
Variable^b	Peak Knee Height	Stride-Foot Contact	Maximum Shoulder External Rotation	Ball Release	Maximum Shoulder Internal Rotation
Trunk rotation	−110.84 ± 9.57	−110.32 ± 12.29	1.12 ± 5.99	18.69 ± 6.82	28.49 ± 8.44
Trunk flexion	−16.03 ± 6.53	−14.09 ± 9.03	−20.37 ± 6.89	−36.79 ± 7.17	−51.99 ± 9.08
Trunk lean	2.09 ± 5	3.1 ± 6.14	−23.01 ± 9.86	−19.02 ± 9.73	−16.03 ± 9.48
Pelvic rotation	−118.15 ± 10.1	−52.98 ± 14.66	9.08 ± 8.88	12.46 ± 9.25	12.4 ± 8.96
Shoulder rotation	21.64 ± 20.4	37.84 ± 22.74	179.07 ± 9.75	109.2 ± 13.17	3.97 ± 15.42
Shoulder horizontal abduction	54.81 ± 18.62	−23.85 ± 9.81	1.06 ± 7.61	5.71 ± 6.89	49.42 ± 13.95
Shoulder abduction	46.2 ± 17.16	82.99 ± 12.79	89.64 ± 7.45	92.55 ± 7.71	105.76 ± 8.27
Elbow flexion	116.02 ± 11.34	102.49 ± 16.56	83.05 ± 9.75	26.48 ± 4.49	26.83 ± 9.2
Stride knee flexion	115.99 ± 10.32	52.2 ± 6.65	48.83 ± 11.54	39.78 ± 14.72	27.58 ± 17.31
Hip-shoulder separation	−4.74 ± 8.53	57.3 ± 9.8	21.58 ± 8.51	12.18 ± 9.38	5.51 ± 10.7

Data are presented in degrees as mean ± SD.

Trunk rotation: – throwing-arm side, + glove side; trunk flexion: – flexion, + extension; trunk lean: – throwing-arm side, + glove side; pelvic rotation: – throwing-arm side, + glove side; shoulder rotation: – internal rotation, + external rotation; shoulder horizontal abduction: – abduction, + adduction; shoulder abduction: – adduction, + abduction; elbow flexion: – extension, + flexion; stride knee flexion: – extension, + flexion; hip-shoulder separation: – torso oriented more toward glove side, + torso oriented more toward throwing-arm side.

Figure 1.

Trunk rotation from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher. Positive values indicate rotation toward glove side, and negative values indicate rotation toward throwing-arm side.

Figure 2.

Trunk flexion from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher. Positive values indicate away from the batter, and negative values indicate toward the batter.

Figure 3.

Trunk lean from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher. Positive values indicate lean toward throwing-arm side, and negative values indicate lean toward glove side.

Figure 4.

Pelvic rotation from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher. Positive values indicate rotation toward glove side, and negative values indicate rotation toward throwing-arm side.

Figure 5.

Shoulder rotation from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher. Positive values indicate external rotation, and negative values indicate internal rotation.

Figure 6.

Shoulder horizontal abduction from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher. Positive values indicate horizontal adduction, and negative values indicate horizontal abduction.

Figure 7.

Shoulder abduction from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher.

Figure 8.

Elbow flexion from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher.

Figure 9.

Knee flexion from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher.

Figure 10.

Hip-shoulder separation from stride-foot contact (0% pitch cycle) to 0.1 second after maximal internal rotation of the throwing shoulder (100% pitch cycle). (A) Mean values with SD (shaded area) and (B) values for each individual pitcher. Positive values indicate pelvis is oriented more toward the glove side than the torso.

Discussion

In this study, we described ball performance metrics and pitching kinematics for collegiate pitchers in a game setting. We recommend these values not be compared with studies using data from marker-based motion capture systems. This is illustrated through the mean fastball velocity in our study, recorded at 40.98 ± 1.17 m/s (91.52 ± 2.61 mph). Other studies using similar populations in laboratory settings typically report lower fastball velocities, in the range of mid-30 m/s.^1,10,14,18 Additionally, fastball velocities are often artificially inflated by analyzing only high-velocity pitchers from a larger cohort.¹⁴ While we do not disagree with this practice to answer specific research questions, our inclusion criteria allowed any pitcher pitching in the SEC to be included for analysis and still obtain a mean velocity higher than typically reported. We believe this is due to the in-game setting of our data, which likely affected kinematic parameters in addition to ball velocity, necessitating context for in-game descriptive data.

Interestingly, in 3 of our variables—shoulder abduction, elbow flexion, and trunk lean—there were distinct, qualitatively different movement strategies, which can be seen in Figures 7, 8, and 3, respectively. While we do not know if these are simply outliers, given the performance of the pitchers in the SEC, they are pitching at a high level despite outlier behavior. Generally, shoulder abduction is relatively flat for the first 40% of the pitch cycle, around 80° to 85°, before progressively becoming more abducted until right before ball release (Figure 7A). Two pitchers had substantial humps in their shoulder abduction time series before ball release, with one coming soon after foot contact and the other coming before ball release, with a negative shoulder abduction velocity through ball release (Figure 7B). This could represent a movement strategy where the pitcher attempts to “pull the ball down” as he throws to home plate. Another pitcher had a clear dip in shoulder abduction in the middle of his pitching motion, which is sometimes seen in pitchers that were catchers in their youth and adopted motor patterns from regularly throwing the ball back to the pitcher. Elbow flexion stayed right around 100° for the first half of the pitching cycle before rapidly extending through roughly 80% of the pitch cycle before a small oscillation back into flexion (Figure 8A). One pitcher had a distinctive oscillation into elbow extension during arm cocking, followed by a return to increased elbow flexion, before a steeper elbow extension velocity before ball release (Figure 8B). We speculate this could be due to lingering effects of learning to pitch from a youth coach who taught pitchers to “show the ball to second base as you load,” a common cue at the youth level. Last, trunk lean typically started just past neutral to the throwing-arm side and steadily progressed to 20° to 25° toward the glove side at roughly 60% of the pitch cycle before progressing about 5° back toward neutral for the rest of the pitch cycle (Figure 3A). Four pitchers kept more neutral trunk lean throughout the pitching motion (Figure 3B). This could be indicative of an altered arm slot or a balance strategy. Whether these altered movement strategies are beneficial or harmful should be addressed through future research. Further, they should be confirmed in other populations, but given that our entire population were high-achieving pitchers, there may be multiple movement strategies that achieve similarly optimal levels of performance.

Limitations

Implementing markerless motion capture will advance sports biomechanics by making in-game data available to researchers and coaches. However, concisely describing these data in a study comes with several limitations. First, not all in-game settings are created equal. There are higher stress and lower stress game situations resulting in different psychological or hormonal states, which may affect mechanics. However, we believe including this variability is beneficial in describing in-game data. Second, we chose the threshold of at least the third pitching outing of the year. Mechanics may differ from the beginning to end of a season, and different pitchers may have their third appearance at different times of the year based on their role on the team (starter, reliever, lower ranking, etc). We defend this decision as long enough to allow pitchers to settle into the season, avoiding potential early-season psychological factors or movement pattern refinement while not making the number of appearances so high that we exclude too many pitchers from our sample. We could not use the same numeric appearance for each pitcher due to data availability (systems being brought online, away games, etc). Third, we provide grand means of within-pitcher means to describe pitching kinematics. To provide a baseline reference, we felt it necessary to provide simple kinematic data first, so analyses based on situation and statistical analyses relating to other variables may be performed later. However, the trade-off is that we did not account for other confounders or situations that may alter mechanics (weather, arm soreness, game situation, etc). Last, it is impossible to validate in-game markerless motion capture data to marker-based data. To date, the validity of the KinaTrax system has only been established for gait analysis.¹⁹ Only 1 published study has validated a different markerless system in baseball pitching; this was performed in a laboratory setting to compare with a laboratory-based marker system.⁶ However, there is no reason to believe the data derived from this markerless motion capture system are inferior to marker-based data, given the issues of marker-based data with fast longitudinal rotation movements.^4,13 Future research should build on this study to provide descriptive data for kinetic measures and confirm other findings in the literature to see if similar relationships hold true in game.

Conclusion

This study described various kinematic parameters among collegiate pitchers with which an individual athlete can be compared. We chose variables processed by and represented in the motion capture platform, increasing accessibility, interpretability, and usability for coaches and researchers using this system. These data can be used as reference values to understand injury mechanisms and provide clinicians with context for the injury-performance trade-off. Clinicians can also be educated to better design return-to-play timelines based on understanding true, in-game demands of pitching.

Footnotes

Final revision submitted January 16, 2024; accepted February 26, 2024.

The authors declared that there are no conflicts of interest in the authorship and publication of this contribution. AOSSM checks author disclosures against the Open Payments Database (OPD). AOSSM has not conducted an independent investigation on the OPD and disclaims any liability or responsibility relating thereto.

Ethical approval for this study was waived by Auburn University.

ORCID iDs

Abigail Schmitt

Gretchen D. Oliver

References

Aguinaldo

Chambers

Correlation of throwing mechanics with elbow valgus load in adult baseball pitchers. Am J Sports Med. 2009;37(10):2043-2048.

Cerveri

Pedotti

Ferrigno

Kinematical models to reduce the effect of skin artifacts on marker-based human motion estimation. J Biomech. 2005;38(11):2228-2236.

Colyer

Evans

Cosker

Salo

AI.

A review of the evolution of vision-based motion analysis and the integration of advanced computer vision methods towards developing a markerless system. Sports Med Open. 2018;4(1):1-15.

Federolf

PA.

A novel approach to solve the “missing marker problem” in marker-based motion analysis that exploits the segment coordination patterns in multi-limb motion data. PLoS One. 2013;8(10):e78689.

Fleisig

Barrentine

Escamilla

Andrews

JR.

Biomechanics of overhand throwing with implications for injuries. Sports Med. 1996;21(6):421-437.

Fleisig

Slowik

Wassom

, et al. Comparison of marker-less and marker-based motion capture for baseball pitching kinematics. Sports Biomech. 2022:1-10.

Harsted

Holsgaard-Larsen

Hestbæk

Boyle

Lauridsen

HH.

Concurrent validity of lower extremity kinematics and jump characteristics captured in pre-school children by a markerless 3D motion capture system. Chiropr Man Therap. 2019;27(1):1-16.

Hauenstein

Huebner

Wagle

, et al. Reliability of markerless motion capture systems for assessing movement screenings. Orthop J Sports Med. 2024;12(3):23259671241234339.

Huang

J-H

Hsu

Y-C.

A multidisciplinary perspective on publicly available sports data in the era of big data: a scoping review of the literature on Major League Baseball. Sage Open. 2021;11(4):21582440211061566.

10.

Kageyama

Sugiyama

Kanehisa

Maeda

Difference between adolescent and collegiate baseball pitchers in the kinematics and kinetics of the lower limbs and trunk during pitching motion. J Sports Sci Med. 2015;14(2):246-255.

11.

Kanko

Laende

Davis

Selbie

Deluzio

KJ.

Concurrent assessment of gait kinematics using marker-based and markerless motion capture. J Biomech. 2021;127:110665.

12.

Knudson

Future trends in the kinesiology sciences. Quest. 2016;68(3):348-360.

13.

Liu

McMillan

Estimation of missing markers in human motion capture. Vis Comput. 2006;22:721-728.

14.

Matsuo

Escamilla

Fleisig

Barrentine

Andrews

JR.

Comparison of kinematic and temporal parameters between different pitch velocity groups. J Appl Biomech. 2001;17(1):1-13.

15.

Moran

Rogler

Wager

JC.

Inter-session repeatability of marker-less motion capture of treadmill running gait. Appl Sci. 2023;13(3):1702.

16.

Mündermann

Corazza

Andriacchi

TP.

The evolution of methods for the capture of human movement leading to markerless motion capture for biomechanical applications. J Neuroeng Rehabil. 2006;3(1):6.

17.

Perrott

Pizzari

Cook

McClelland

JA.

Comparison of lower limb and trunk kinematics between markerless and marker-based motion capture systems. Gait Posture. 2017;52:57-61.

18.

Post

Laudner

McLoda

Wong

Meister

Correlation of shoulder and elbow kinetics with ball velocity in collegiate baseball pitchers. J Athl Train. 2015;50(6):629-633.

19.

Ripic

Signorile

Kuenze

Eltoukhy

Concurrent validity of artificial intelligence-based markerless motion capture for over-ground gait analysis: a study of spatiotemporal parameters. J Biomech. 2022;143:111278.

20.

Sonnenfeld

Crutchfield

Swindell

, et al. An analysis of in vivo hip kinematics in elite baseball batters using a markerless motion-capture system. Arthrosc Sports Med Rehabil. 2021;3(3):e909-e917.

21.

Van der Kruk

Reijne

. Accuracy of human motion capture systems for sport applications; state-of-the-art review. Eur J Sport Sci. 2018;18(6):806-819.

22.

Siegler

Allard

, et al. ISB recommendation on definitions of joint coordinate system of various joints for the reporting of human joint motion—part I: ankle, hip, and spine. J Biomech. 2002;35(4):543-548.

23.

Van der Helm

Veeger

, et al. ISB recommendation on definitions of joint coordinate systems of various joints for the reporting of human joint motion—part II: shoulder, elbow, wrist and hand. J Biomech. 2005;38(5):981-992.