Application of motion capture technology for sport performance analysis

In sport performance, motion capture aims at tracking and recording athletes’ human motion in real time to analyze physical condition, athletic performance, technical expertise and injury mechanism, prevention and rehabilitation. The aim of this paper is to systematically review the latest developments of motion capture systems for the analysis of sport performance. To that end, selected keywords were searched on studies published in the last four years in the electronic databases ISI Web of Knowledge, Scopus, PubMed and SPORTDiscus, which resulted in 892 potential records. After duplicate removal and screening of the remaining records, 81 journal papers were retained for inclusion in this review, distributed as 53 records for optical systems, 15 records for non-optical systems and 13 records for markerless systems. Resultant records were screened to distribute them according to the following analysis categories: biomechanical motion analysis, validation of new systems and performance enhancement. Although optical systems are regarded as golden standard with accurate results, the cost of equipment and time needed to capture and postprocess data have led researchers to test other technologies. First, non-optical systems rely on attaching sensors to body parts to send their spatial information to computer wirelessly by means of different technologies, such as electromagnetic and inertial (accelerometry). Finally, markerless systems are adequate for free, unobstructive motion analysis since no attachment is carried by athletes. However, more sensors and sophisticated signal processing must be used to increase the expected level of accuracy.


Introduction
Performance analysis of sport aims at evaluating athletes' improvements as a consequence of training over an extended period of time, the degree of physical condition or technical expertise.Coaches looking for optimal performance and technique develop a tailored training plan and monitor outcomes by means of different tools (Southgate, Prinold, & Weinert-Aplin, 2016).
Analysis of human motion can be regarded as one of the most precise technique to address elite sports' outcomes.Topics as competitive performance, for example, achieving greater joint extension and acceleration to boost jump height (Chiu, Bryanton, & Moolyk, 2014), technique evaluation, such as changes in hip flexion-abduction to enhance technique in bars (Huchez, Haering, & Holvoët, 2013), prevention of injuries, for example by stating than females are more prone to knee injuries during the fencing lunge (Jonathan Sinclair & Bottoms, 2013), equipment optimization, for example, optimal bat weighting (Laughlin, Fleisig, Aune, & Diffendaffer, 2016), or team interaction, for example, in rowing (Ruffaldi, Peppoloni, & Filippeschi, 2015).All these systems aims at tracking the motion of body segments with which to compute appropriate joint and segment kinematics.Additional information can also be extracted with further processing, either from the device itself or by means of complex muscular and/or skeleton models.
There are two main methods of analyzing human movement for sport motion.On the one hand, in the video-based approach, cameras are used to record athletes during training or competition and provide them with extrinsic visual feedback (Pueo, 2016).This feedback can be delivered as soon the task is completed or after manual digitizing of the body landmarks of interest by mean of dedicated software.On the other hand, automatic tracking motion analysis systems, also known as Power law could be implemented in an agility drill thus opening the way for establishment of a more representative measure of agility performance instead of drill duration.
a (Bisesti et al., 2015) Qualisys, 8, 240 Hz 15: 8m, 7f Running Running barefoot alters foot strike patterns and decreases knee extension joint moments.Cutting barefoot exposes the anterior cruciate ligament to strain no more than when shod, and may decrease undesirable mechanics related to anterior cruciate ligament injury.a (Cockcroft & Van Den Heever, 2016) Vicon, 8, 200 Hz 15 Rugby Even though the participants had variability in their run-up to the tee, final foot position next to the tee was very similar and consistent.
a (Crotin & Ramsey, 2015) Vicon, 8, 240 Hz 19 Baseball Tracking ball velocity throughout a game provides instantaneous feedback to the pitcher, information that is required to maintain ball velocity irrespective of pitch accumulation.
a (Crotin et al.,  The interaction moment of the kicking was the main factor causing the pelvis counter-clockwise rotation within the horizontal plane from the overhead view that precedes a proximal-to-distal sequence of segmental action of the swing leg. a (Sinclair & Bottoms, 2013) Qualisys, 8, 250 Hz 16: 8m, 8f Fencing Females produced greater knee abduction at the end of the lunge as well as greater hip adduction, so they could be at a greater risk of knee injuries due to the greater knee abduction and hip adduction produced during the fencing lunge.
a (Joyce et al., 2013) Vicon, 10, 250 Hz 15 Golf Three of the six variables used to explain the variance in clubhead speed were from the lower trunk, which shows, in part, the importance of including segments such as the lower trunk when examining the golf swing.

Results
An initial search identified 892 records.After duplicate removal and screening of the remaining records, 81 full-text papers were retained for inclusion in this review, distributed as 53 records for optical systems (Table 1), 15 records for non-optical systems (Table 2) and 13 records for markerless systems (Table 3).Resultant records were screened to distribute them according to the following analysis categories: (a) biomechanical motion analysis, (b) validation of new systems and (c) performance enhancement.

Optical systems
These systems track either passive (reflective spheres) or active (LEDs) markers placed on body landmarks to triangulate their 3D position.Generally, eight or more cameras are necessary for full-body capture through different marker sets comprising around 40 markers at specific anatomic landmarks (Cappozzo, Catani, Della Croce, & Leardini, 1995).Actual systems can capture from 100 frames per second (fps) to 500 fps, although higher frame rates can be achieved by decreasing image resolution.Alternatively, fps number is also displayed as image capture frequency in hertz (Hz), where 1 fps equals 1 Hz.
Optical motion capture has been used in sports with implements for assessing ball velocity, as a function of several biomechanical conditions, such as stride length (Crotin, Kozlowski, Horvath, & Ramsey, 2014) or sequence-dependent rotation axes (Hansen C, Rezzoug N, Gorce P, 2015) in baseball, lower trunk (Joyce, Burnett, Cochrane, & Ball, 2013) and torso rotation (Sinclair, Currigan, Fewtrell, & Taylor, 2014) in golf and upper limb joint (Martin et al., 2014;Martin, Kulpa, Delamarche, & Bideau, 2013;Martin, Kulpa, Ropars, Delamarche, & Bideau, 2013).In running, studies focus mainly on foot strike patterns The Kinect agreed well with the marker-based system in terms of the pattern of the motion, the reliability, and peak angles calculated, so it may be a feasible markerless motion capture tool that can be used in the clinic.
b (Bonnechère et al., 2014) Kinect, 30 Hz (SDK 1.5) [Vicon,8,60 Hz] 48 Laboratory Kinect reproducibility was found to be statistically similar to marker-based system results for the four exercises.However, measured ROM however were found different between the systems.The Kinetic could provide clinicians with significant advantages for assessing some spatiotemporal gait parameters.However, some commonly reported variables cannot be accurately measured.b (J.Kim et al., 2015) Xtion, 30 Hz [Vicon, 120 Hz] 10; 5m, 5f Walking The proposed treadmill achieves similar performance to a typical, costly, interactive treadmill that contains a motion capture system and an instrumented treadmill.
Optical motion capture is a time-consuming system that requires long preparation at the laboratory (each athlete must wear a full body capture suit with marker sets) and postprocessing to clean all corrupted data, such as occlusions or marker swaps.Therefore, the sample is rather low, being 20 a mean value across records.In validation studies, where absolute differences from various methods may arise under same circumstances, standard calibration frames are also used.These tools consist of geometrical structures carrying two or more markers at known distances, generally used for system calibration.
Regarding optical systems' manufacturers, there is a preference for two systems: Vicon, Oxford Metrics, UK (26 records) and Qualysis AB, Sweden (17 records), with the 9 remainder records distributed between 3 manufacturers.The number of cameras ranges between 24 for a structural lesions study in tennis players (Lädermann, Chagué, Kolo, & Charbonnier, 2016) and only two for an underwater motion capture proposal made with action cameras (Bernardina, Cerveri, Barros, Marins, & Silvatti, 2016) by means of Direct Linear Transformation (DLT), first introduced by Abdel- Aziz & Karara (1971).Finally, except for two outliers, image capture frequency goes from 100 Hz to 500 Hz, depending on the system.From these figures, a typical optical system would consist of 12 cameras working at 200 Hz from the two major manufacturers.Such an experimental setup would fulfill a number of sport analyses.

Non-optical systems
Unlike the above technology, non-optical systems use sensors attached to body parts to send data about their position in space.Electromagnetic systems are composed of small electric coils (three per sensor, orthogonally) within a well-defined electromagnetic field created by a powerful source.Athlete's movements are tracked when coils generate small voltage or current inside a constant magnetic flux.As demonstrated in the four records with EM tag in Table 2, each sensor is able to read location and orientation simultaneously, but any metal object in the vicinity of sensors may introduce distortions in the magnetic field, ruining measurement accuracy.For example, in (Y.-K.Kim, 2013), the weighting supplement on arms and bats (Valeo, USA), as well as the bat itself, is non-metal to avoid interferences with the 4" by 4" global reference frame transmitter.Only two manufacturers are used in these records: on the one hand, Polhemus, USA, in a single-sensor setup, ranging from 60 Hz (Kasmer, Ketchum, & Liu, 2014) to 120 Hz (Norcross et al., 2013a(Norcross et al., , 2013b) ) and on the other hand, Motion Star, USA, in a 12-sensor setup working at 240 frames per second.
Inertial system technology is another kind of non-optical systems based on inertial measurement units (IMU), which are small micromechanical sensors containing accelerometers (Leirós-Rodríguez, Arce, García-Soidán, & Naveira-Barbeito, 2017), usually tri-axial to measure the three coordinates, and often combined with gyroscopes and magnetometers.For static subjects, accelerometers are able to measure angular rotations, however, for dynamic motions, gyroscopes and magnetometers are combined with accelerometry through sensor fusion algorithms.Inertial system data is often transmitted wirelessly to a computer in real-time with 6 degrees of freedom per sensor.
Since IMUs can be regarded as an emerging technology, studies on sport science are devoted to well-known sport biomechanics, such as gait/walking (5 records) and running (2 records), out of 10 records.Generally, such systems are compared to optical systems as reference criterion or used in conjunction with them to complement data, with the characteristics described above.However, in (Zorko, Nemec, Babiè, Lešnik, & Supej, 2015), a Real Time Kinematic Global Navigation Satellite Systems (RTK GNSS) is used to merge inertial data with highdefinition and high-frequency real-time geolocalization to track skis in alpine skiing.Similarly, Gageler, Thiel, Neville, & James (2013), used an IMU with a ±8g tri-axial accelerometer placed in the middle to upper thoracic vertebrae and an additional custom-build IMU containing a 100g accelerometer attached to the distal Fibulas to measure vertical accelerations in running, where ground impacts were expected to exceed the range of 8g provided by the first IMU.The number of IMUs in each study is very low to date, with one or two IMUs at specific locations.However, Ruffaldi, Peppoloni, & Filippeschi (2015) proposed a novel approach to track outdoor rowing performance from different instruments, including one IMU located on the trunk in the midpoint between the scapula, and two IMUs for each upper limb, on the arm and the forearm.This study shows that one of the IMUs potential is their suitability for indoor and outdoor experiments due to their low mass and minimum invasive behavior.
New non-optical technologies have also been tested in the sport science field.For example, Telfer, Woodburn, & Turner (2014) used an intraoperative 13 MHz linear array ultrasound probe (IOE 323, Italy) at 47 Hz to demonstrate that, together with an optical motion capture, intrinsic foot kinematics may be measured accurately.Similarly, Wang, Skubic, Rantz, & Cuddihy (2014) proposed to use a shelf pulse doppler range control radar (PrecisionLine, USA), originally designed as motion detectors in home security systems, to track the walk segment suitable for gait analysis as different human body parts generate different radar signatures.
Finally, there are other motion capture systems, like the mechanical motion (Menache, 2011), which consist of rigid structures attached to joints of the body that articulate as subject's body is in motion.Although being a low-cost and real-time via wireless system, such technology is not used in sport sciences since exoskeletons prevent athletes from performing unobstructed natural motions.

Markerless systems
Finally, sophisticated computer image processing is emerging to reconstruct 3D motion from video and other sensors without subjects needing to wear special equipment for tracking.Generally, commercial videogame and entertainment 3D cameras, comprising one or two regular video cameras, together with infrared and ultrasonic sensors, are used to identify human forms from which to extract body segments and joints.
As with non-optical systems, this relatively newly introduced technology has been used in studies with tested biomechanics, such as gait/walking (6 records) and laboratory tests in which controlled motions are assessed (4 records).The remaining 3 records cover different activities, such as jumping, stability or tennis.The fact that no marker is needed for capturing allows for the test of a number of subjects: excluding a study with one subject, the mean value is 17 subjects per experiment, with several studies comprising 48 and 20 subjects.
The Kinect camera is the technology mostly used in markerless systems (8 out of 9 records).As part of Xbox video console, it consists of a 30 Hz video camera, a depth sensor (infrared laser projector) and multi-array microphone.The original aim was to provide full body motion capture, facial and voice recognition for gaming, but these features have been used by sport scientists with third-party developments and custom-build software by means of software development kit (SDK), provided by Microsoft.Alternative depth cameras are found in (J.Kim, Gravunder, & Park, 2015), where a Asus Xtion was used with similar features as Kinect: video camera and depth sensor.The main disadvantage of both depth cameras is the level of spatial and temporal accuracy.Video cameras in such systems usually work at 640x480 pixels resolution, which would be enough for body recognition in gaming, but for applications where the aim is to track the motion of a body segment, some issues may appear.Across studies presented in this paper, conclusions pointed out that it is a promising technology, but more data processing is necessary, together with fusion with other sensors, such as additional video cameras, to reach a proper accuracy.Regarding temporal resolution, a fixed 30 fps capability may limit its use for slow to moderate velocity motions.
Another technique to retrieve information on body segments is the use of a number of video cameras and a later image processing (Abdel-Aziz & Karara, 1971).Abrams, Harris, Andriacchi, Safran, & Safran (2014) scanned seven tennis players with laser and then used eight cameras (resolution 640x480 pixels, 200 fps) to track tennis serves.Sandau et al. (2014) used eight cameras with higher resolution but lower framerate (2048x2048 pixels, 75 fps) to track gait motion.O' Keefe et al. (2014) tracked the travelled distance of the body center of mass, forearm, and foot and motion frequency of the latter of fragile X syndrome patients with fourteen cameras distributed in a canvas of 3.5 m x 4.9 m.
Finally, non-conventional methods have also been explored to validate them as alternative to optical and non-optical systems.Krigslund et al. (2013) proposed the use of Radio Frequency Identification (RFID) tags, generally used for object identification and gross localization, to track thigh and shank motions in the sagittal plane.

Conclusion
In this paper, a systematic review of the latest developments of motion capture systems for the analysis of sport performance has been addressed.Although restricted to journals in the last four years, the number of records has been very high, indicating the interest of this topic.
Optical systems have been tested in sport sciences and other disciplines for years, so they can be regarded as reference criterion in most studies with accurate results.However, being a time-consuming process and the cost of laboratory equipment makes it difficult for individuals and small organizations to use them.
In order to overcome these drawbacks, non-optical systems have emerged over the years.They all rely on attaching sensors to body parts to send their space coordinates and rotations in three perpendicular axes to computer wirelessly.Two major technologies are used to this end, electromagnetic and inertial (accelerometry) with accurate results.Less conventional methods have also been tested, such as ultrasonics and radar, but their suitability for sport sciences is rather questionable.
Finally, markerless systems are a step forward in motion capture, since no attachment is carried by subjects, allowing for a free, unobstructive motion.Kinect technology is the key technology used in these studies since it offers several advantages: low cost equipment, easy setup, both subject and researcher and real-time results.However, the level of accuracy expected for clinical and experimental research is still far from proper standards.By means of integrating more sensors and signal processing, accurate results may be obtained in a near future.

Table 1
Ball speed was mainly influenced by the flexion of the supporting leg (ankle, knee and hip), by the rotation of the hip and of the trunk, and by the rotation and abduction-adduction movements of the wrist of the hand on the top of the stick.
(Goss et al., 2015)laar, 2015)rategy, which is associated with greater net joint moments and segment accelerations, all together contributing to better vertical jump performance.c(Estevanetal., 2015)Qualisys, 8, 240 Hz 10 Taekwondo During the roundhouse kick in taekwondo inter-segment motion seems to be based on a proximo-distal pattern.c(Estevanetal., 2016)Qualisys, 8, 247 Hz 8 Taekwondo Including exercises that require in-phase movement could not only help athletes to acquire coordination stability but also efficiency.c(Estevanetal., 2013)Qualisys, 8, 247 Hz 9 Taekwondo Athletes should not adopt the 90°stance position because it will not enable them to achieve the best performance in the roundhouse kick.c(Gomo&vandenTillaar,2015)Qualisys, 6, 500Hz 12m Powerlifting Due to the differences in moment arm of the barbell about the elbow joint in the sticking region, there still might be a poor mechanical region for total force production that is joint angle-specific.c(Gossetal., 2015)Vicon, 8, 240 Hz 89 Running Runners often cannot report their foot-strike patterns accurately and may not automatically adopt an anteriorfoot-strike pattern after transitioning to minimalist running shoes.c (Hart, Cochrane, Spiteri, Nimphius, & Newton, 2016) Vicon, 10, 250 Hz 31 Australian Football Resistance training and skill acquisition programs may play a role on the development of kicking accuracy since relative lean mass was positively correlated with kicking accuracy.c (Huchez, Haering, & Holvoët, 2013) Vicon, 10, 250 Hz 8f Gymnastics The study provided directions for training: to increase hip flexion-abduction, to transfer leg and arm angular momentum to the trunk and to straighten hand path to the bar.c (Sinclair, Currigan, et al., 2014) Qualisys, 8, 500 Hz 50 Golf Sagittal plane wrist velocity and peak transverse plane torso rotation are the strongest contributors to ball velocity and potentially overall driving performance.synchronized metronome training on the underlying brain-based motor control strategies lead to a more coordinated movement pattern of the golf-swing performance.c (Z. Zhang et al., 2016) Vicon, 10, 200 Hz 24 Badminton Trunk rotation was shown to be vital for maximizing the release speed of the shuttlecock -an important measure of the quality of the forehand smash that skilled players master.c (Zago et al., 2015) BTS, 9, 60 Hz 10m Soccer Faster players are able to run with the ball through a shorter path in a more economical way so coaches are encouraged to design specific practices where high stride frequency and narrow run trajectories are required.First column.a: Motion Analysis, b: System validation, c: Performance enhancement.Third column.System manufacturer, number of cameras, frames per second.Criterion reference system between brackets where available.Fourth column.m: Male, f: Female, SCP: Standard Calibration Frame.

Table 2
Greater sagittal-plane initial impact phase energy absorption likely indicates greater anterior cruciate ligament loading, whereas in the frontal-plane, it was associated with less favorable frontal-plane biomechanics.recordingsystems aims to assess the temporal and spatial descriptors, such as that heel-off moment was reliably at 51% of the step cycle, irrespective of gait velocity.This study introduced an easy-to-use system to measure the spatio-temporal parameters in diagonal stride XC skiing by integrating the IMUs in the poles and skis.Motion Analysis, b: System validation, c: Performance enhancement.Third column.System manufacturer, number of sensors, adquisition frequency.IMU: Inertial Measurement Unit, EM: Electromagnetic, US: Ultrasonics, RTK GNSS: Real Time Kinematic Global Navigation Satellite Systems.Criterion reference system between brackets where available.Fourth column.m: Male, f: Female, SCP: Standard Calibration Frame