Cricket fast bowling detection in a training setting using an inertial measurement unit and machine learning

Fast bowlers are at a high risk of overuse injuries. There are specific bowling frequency ranges known to have negative or protective effects on fast bowlers. Inertial measurement units (IMUs) can classify movements in sports, however, some commercial products can be too expensive for the amateur athlete. As a large number of the world's population has access to an IMU (e.g. smartphones), a system that works on a range of different IMUs may increase the accessibility of automated workload monitoring in sport. Seventeen elite fast bowlers in a training setting were used to train and/or validate five machine learning models by bowling and performing fielding drills. The accuracy of machine learning models trained using data from all three bowling phases (pre-delivery, delivery and post-delivery) were compared to those trained using only the delivery phase at a sampling rate of 250 Hz. Next, models were trained using data down-sampled to 125 Hz, 50 Hz, and 25 Hz to mimic results from lower specification sensors. Models trained using only the delivery phase showed similar accuracy (> 95%) to those trained using all three bowling phases. When delivery-phase data were down-sampled, the accuracy was maintained across all models and sampling frequencies (>96%).     

影像测量数据在击剑弓步技术诊断中的应用

采用影像测量法、数理统计法,对佩剑运动员弓步技术特征点数据进行研究,分析认为:利用人体重心在垂直方向的速度与位移曲线相结合的方法,可以更清楚地反映并较准确地划分弓步技术动作阶段,获得相关数据,掌握运动员各阶段的人体重心变化情况;利用前脚重心点位移与速度特征指标,可以分析"摆"、"伸"步法技术各个环节的问题;通过膝关节角度指标,能对运动员站位高低,人体重心偏移进行诊断,并利用膝关节角度值变化,分析脚步摆动腿幅度,支撑腿用力等情况,判断弓步结束姿势的稳定性.应用影像测量数据简单易行,有助于对击剑运动技术进行分析研究,解决技术诊断中的实际问题.     

The validation and application of Inertial Measurement Units to springboard diving

Inertial Measurement Units (IMUs) may offer an ecologically valid, reliable, and practical method for biomechanical performance analysis. With such potential in mind, Part 1 of this study examined the accuracy of IMUs gyroscopes with an optical system (Cortex 3.3). A calibration formula standardised the IMUs angular velocity output with the optical system. The percentage differences between the two measures = 0.5% (p < 0.05), suggest IMU’s efficacy for application. In Part 2, the aim was to examine and understand how dive flight angular velocity time series plots change and increase according to dive degree of difficulty. With IMUs attached to three competitive divers performing forward somersault dives, dive flight kinematics were assessed. Biomechanically, a 4½ tuck somersault dive differed to lower degree of difficulty dives in terms of: (1) a rotational delay immediately after takeoff (to gain greater vertical translation); (2) increased total time of flight; (3) greater muscle effort to resist increased centrifugal forces produced by the increased angular velocity (1,090 °/s); and (4) greater eccentric control during deceleration allow a safe and vertical entry into the water. IMUs can be effectively utilised and integrated into contexts such as springboard diving for performance analysis and optimisation purposes.     

Physical Activity Classification in Youth Using Raw Accelerometer Data from the Hip

ABSTRACT

This study developed and evaluated machine learning algorithms to predict children’s physical activity category from raw accelerometer data collected at the hip. Fifty participants (mean age = 13.9 ± 3.0 y) completed 12 activity trials that were categorized into 5 categories: sedentary (SED), light household activities and games (LHHAG), moderate-vigorous games and sports (MVGS), walking (WALK), and running (RUN). Random Forest (RF) and Logistic Regression (LR) classifiers were trained with features extracted from the vector magnitude using 10?s non-overlapping windows. Classification accuracy was evaluated using leave-one-subject-out cross validation. Overall accuracy for the RF and LR classifiers was 95.7% and 94.3%, respectively. Classification accuracy was excellent for SED (96.3% – 98.1%), LHHAG (92.3% – 95.2%), WALK (94.5% – 97.1%), RUN (99.5% – 99.6%); and MVGS (87.5% – 92.7%). The results indicate that classifiers trained on features in the raw acceleration from the hip can be used for activity recognition in young people.

Abbreviations: VM: Vector Magnitude; RF: Random Forest; LR: Logistic Regression; LOSO: Leave-One-Subject-Out     

重剑运动员弓步刺动作协调性的非线性动力学特征

采用视频动作解析法,研究重剑运动员在不同速度、不同情境下弓步刺动作肘膝关节角度变化的运动学特征.结果显示:不同速度要求下,重剑运动员弓步刺动作协调性存在非线性动力学特征;不同情境下,重剑运动员弓步刺动作协调性可能存在非线性动力学特征;高、低水平重剑运动员在平时训练时的弓步刺动作协调模式相似;比赛情境时,高水平重剑运动员的弓步刺动作协调性显著好于低水平重剑运动员.     

Classification of playing position in elite junior Australian football using technical skill indicators

?In team sport, classifying playing position based on a players’ expressed skill sets can provide a guide to talent identification by enabling the recognition of performance attributes relative to playing position. Here, elite junior Australian football players were a priori classified into 1 of 4 common playing positions; forward, midfield, defence, and ruck. Three analysis approaches were used to assess the extent to which 12 in-game skill performance indicators could classify playing position. These were a linear discriminant analysis (LDA), random forest, and a PART decision list. The LDA produced classification accuracy of 56.8%, with class errors ranging from 19.6% (midfielders) to 75.0% (ruck). The random forest model performed at a slightly worse level (51.62%), with class errors ranging from 27.8% (midfielders) to 100% (ruck). The decision list revealed 6 rules capable of classifying playing position at accuracy of 70.1%, with class errors ranging from 14.4% (midfielders) to 100% (ruck). Although the PART decision list produced the greatest relative classification accuracy, the technical skill indicators reported were generally unable to accurately classify players according to their position using the 3 analysis approaches. This player homogeneity may complicate recruitment by constraining talent recruiter’s ability to objectively recognise distinctive positional attributes.     

A biomechanical analysis of common lunge tasks in badminton

Abstract

The lunge is regularly used in badminton and is recognized for the high physical demands it places on the lower limbs. Despite its common occurrence, little information is available on the biomechanics of lunging in the singles game. A video-based pilot study confirmed the relatively high frequency of lunging, ～15% of all movements, in competitive singles games. The biomechanics and performance characteristics of three badminton-specific lunge tasks (kick, step-in, and hop lunge) were investigated in the laboratory with nine experienced male badminton players. Ground reaction forces and kinematic data were collected and lower limb joint kinetics calculated using an inverse dynamics approach. The step-in lunge was characterized by significantly lower mean horizontal reaction force at drive-off and lower mean peak hip joint power than the kick lunge. The hop lunge resulted in significantly larger mean reaction forces during loading and drive-off phases, as well as significantly larger mean peak ankle joint moments and knee and ankle joint powers than the kick or step-in lunges. These findings indicate that, within the setting of this investigation, the step-in lunge may be beneficial for reducing the muscular demands of lunge recovery and that the hop lunge allows for higher positive power output, thereby presenting an efficient lunging method.     

Calibration of two objective measures of physical activity for children

A calibration study was conducted to determine the threshold counts for two commonly used accelerometers, the ActiGraph and the Actical, to classify activities by intensity in children 5 to 8 years of age. Thirty-three children wore both accelerometers and a COSMED portable metabolic system during 15 min of rest and then performed up to nine different activities for 7 min each, on two separate days in the laboratory. Oxygen consumption was measured on a breath-by-breath basis, and accelerometer data were collected in 15-s epochs. Using receiver operating characteristic curve (ROC) analysis, cutpoints that maximised both sensitivity and specificity were determined for sedentary, moderate and vigorous activities. For both accelerometers, discrimination of sedentary behaviour was almost perfect, with the area under the ROC curve at or exceeding 0.98. For both the ActiGraph and Actical, the discrimination of moderate (0.85 and 0.86, respectively) and vigorous activity (0.83 and 0.86, respectively) was acceptable, but not as precise as for sedentary behaviour. This calibration study, using indirect calorimetry, suggests that the two accelerometers can be used to distinguish differing levels of physical activity intensity as well as inactivity among children 5 to 8 years of age.     

Foot loading characteristics during three fencing-specific movements

Abstract

Plantar pressure characteristics during fencing movements may provide more specific information about the influence of foot loading on overload injury patterns. Twenty-nine experienced fencers participated in the study. Three fencing-specific movements (lunge, advance, retreat) and normal running were performed with three different shoe models: Ballestra (Nike, USA), Adistar Fencing Lo (Adidas, Germany), and the fencers' own shoes. The Pedar system (Novel, Munich, Germany) was used to collect plantar pressures at 50 Hz. Peak pressures, force–time integrals and contact times for five foot regions were compared between four athletic tasks in the lunge leg and supporting leg. Plantar pressure analysis revealed characteristic pressure distribution patterns for the fencing movements. For the lunge leg, during the lunge and advance movements the heel is predominantly loaded; during retreat, it is the hallux. For the supporting leg, during the lunge and advance movements the forefoot is predominantly loaded; during retreat, it is the hallux. Fencing-specific movements load the plantar surface in a distinct way compared with running. An effective cushioning in the heel and hallux region would help to minimize foot loading during fencing-specific movements.     

Lunge performance and its determinants

For activities such as squash, badminton and fencing, the ability to quickly complete a lunge and return to the start or move off in another direction is critical for success. Determining which strength qualities are important predictors of lunge performance was the focus of this study. Thirty-one male athletes performed: (1) a unilateral maximal squat (one-repetition maximum, 1-RM) and unilateral jump squat (50% 1-RM) on an instrumented supine squat machine, and (2) a forward lunge while attached to a linear transducer. We performed stepwise multiple regression analysis with lunge performance as the dependent variable and various strength, flexibility and anthropometric measures as the independent variables. From the many strength and power measures calculated, time to peak force was the best single predictor of lunge performance, which accounted for 55% of the explained variance. The best three-variable model for predicting lunge performance accounted for 76-85% of the explained variance. The models differed, however, according to whether lunge performance was expressed relative to body mass (time to peak force, mean power and relative strength = 76%) or taken as an absolute value (time to peak force, leg length and flexibility = 85%). We conclude that one to two trials were reliable for strength diagnosis and that one strength measure cannot accurately explain functional performance because other factors, such as body mass, flexibility and leg length, have diverse effects on the statistical models.     

Foot loading characteristics during three fencing-specific movements

Plantar pressure characteristics during fencing movements may provide more specific information about the influence of foot loading on overload injury patterns. Twenty-nine experienced fencers participated in the study. Three fencing-specific movements (lunge, advance, retreat) and normal running were performed with three different shoe models: Ballestra (Nike, USA), Adistar Fencing Lo (Adidas, Germany), and the fencers' own shoes. The Pedar system (Novel, Munich, Germany) was used to collect plantar pressures at 50 Hz. Peak pressures, force-time integrals and contact times for five foot regions were compared between four athletic tasks in the lunge leg and supporting leg. Plantar pressure analysis revealed characteristic pressure distribution patterns for the fencing movements. For the lunge leg, during the lunge and advance movements the heel is predominantly loaded; during retreat, it is the hallux. For the supporting leg, during the lunge and advance movements the forefoot is predominantly loaded; during retreat, it is the hallux. Fencing-specific movements load the plantar surface in a distinct way compared with running. An effective cushioning in the heel and hallux region would help to minimize foot loading during fencing-specific movements.     

The relationship between inertial measurement unit-derived ‘force signatures’ and ground reaction forces during cricket pace bowling

Abstract

This study assessed the reliability and validity of segment measured accelerations in comparison to front foot contact (FFC) ground reaction force (GRF) during the delivery stride for cricket pace bowlers. Eleven recreational bowlers completed a 30-delivery bowling spell. Trunk- and tibia-mounted inertial measurement units (IMUs) were used to measure accelerations, converted to force, for comparisons to force plate GRF discrete measures. These measures included peak force, impulse and the continuous force–time curve in the vertical and braking (horizontal) planes. Reliability and validity was determined by intra-class correlation coefficients (ICC), coefficient of variation (CV), Bland–Altman plots, paired sample t-tests, Pearson’s correlation and one-dimensional (1D) statistical parametrical mapping (SPM). All ICC (0.90–0.98) and CV (4.23–7.41%) were acceptable, except for tibia-mounted IMU braking peak force (CV = 12.44%) and impulse (CV = 18.17%) and trunk vertical impulse (CV = 17.93%). Bland–Altman plots revealed wide limits of agreement between discrete IMU force signatures and force plate GRF. The 1D SPM outlined numerous significant (p < 0.01) differences between trunk- and tibia-located IMU-derived measures and force plate GRF traces in vertical and braking (horizontal) planes. The trunk- and tibia-mounted IMUs appeared to not represent the GRF experienced during pace bowling FFC when compared to a gold-standard force plate.     

Response timing in the lunge and target change in elite versus medium-level fencers

Abstract

The aim of the present work is to examine the differences between two groups of fencers with different levels of competition, elite and medium level. The timing parameters of the response reaction have been compared together with the kinetic variables which determine the sequence of segmented participation used during the lunge with a change in target during movement. A total of 30 male sword fencers participated, 13 elite and 17 medium level. Two force platforms recorded the horizontal component of the force and the start of the movement. One system filmed the movement in 3D, recording the spatial positions of 11 markers, while another system projected a mobile target over a screen. For synchronisation, an electronic signal enabled all the systems to be started simultaneously. Among the timing parameters of the reaction response, the choice reaction time (CRT) to the target change during the lunge was measured. The results revealed differences between the groups regarding the flight time, horizontal velocity at the end of the acceleration phase, and the length of the lunge, these being higher for the elite group, as well as other variables related to the temporal sequence of movement. No significant differences have been found in the simple reaction time or in CRT. According to the literature, the CRT appears to improve with sports practice, although this factor did not differentiate the elite from medium-level fencers. The coordination of fencing movements, that is, the right technique, constitutes a factor that differentiates elite fencers from medium-level ones.     

A multi-sensor system for automatic analysis of classical cross-country skiing techniques

Commercial systems utilizing data from inertial measurement units (IMUs) to analyse movement patterns have not yet been adapted to monitor daily training in cross-country (XC) skiing. The main purposes of this study are to investigate: (1) the feasibility and potential of a multi-sensor system consisting of a heart rate sensor, global navigation satellite systems (GNSSs) data and seven IMUs placed at multiple locations on the body for outdoor XC skiing, and (2) the validity of employing hard decision rules based on the correlation between arms and legs for detecting sub-techniques in classical XC skiing. All sensor data were synchronously sampled and synchronized with GNSS data from a commercially available sports watch while XC skiing on varying tracks, from amateur skiers and world-class athletes. An algorithm based on the correlation of the angular velocity of arms and legs was developed to detect the three main classic sub-techniques, diagonal, double poling with a kick and double poling. Other sub-techniques were classified as miscellaneous (0–20%). The system is shown to work well outdoors on snow during different conditions, and the implemented algorithm was validated by video analyses to detect the three sub-techniques with a sensitivity of 99–100%. This study is the first to detect and link sub-techniques in XC skiing to GNSS data, thereby associating the detection and distribution of sub-techniques to different terrains. Such information gives insight into the technical and tactical aspects of skiers’ daily training and competitions, thereby providing a tool for coaches and athletes.     

Spatial characteristics of professional tennis serves with implications for serving aces: A machine learning approach

This study sought to determine the features of an ideal serve in men’s professional tennis. A total of 25,680 first serves executed by 151 male tennis players during Australian Open competition were classified as either aces or returned into play. Spatiotemporal (impact location, speed, projection angles, landing location and relative player locations) and contextual (score) features of each serve were extracted from Hawk-Eye data and used to construct a classification tree model (with decision rules) that predicted serve outcome. k-means clustering was applied to the landing locations to quantify optimal landing locations for aces. The classification tree revealed that (1) serve directionality, relative to the returner; (2) the ball’s landing proximity to the nearest service box line and (3) serve speed classified aces with an accuracy of 87.02%. Hitting aces appeared more contingent on accuracy than speed, with serves directed >5.88° from the returner and landing <15.27 cm from a service box line most indicative of an ace. k-means clustering revealed four distinct locations (≈0.73 m wide × 2.35 m deep) in the corners of the service box that corresponded to aces. These landing locations provide empirically derived target locations for players to adhere to during practice and competition.     

The role of ultrasonography in the diagnosis of anterior cruciate ligament injury: A systematic review and meta-analysis

To evaluate the value of ultrasonography in the diagnosis of anterior cruciate ligament injury (ACL injury) by conducting a systematic review and meta-analysis. A literature search was carried out in the Cochrane Library, Embase, Pubmed databases and included studies prior to April 2017. Based on inclusion and exclusion criteria, studies evaluating ultrasound to diagnose ACL injury were selected. MRI, arthroscopy and clinical-follow were considered the reference standards. The diagnostic accuracy of ultrasound was assessed using a combination of sensitivity, specificity, likelihood ratio (LR), post-test probability, diagnostic odds ratio (DOR) and by summarizing the area under the receiver operating characteristic (SROC) curve. A total of 4 studies involving 246 patients were eventually included in the analysis. In these four studies, the combined sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, post-test probability and DOR were 90.0% (95% CI: 77–96), 97% (95% CI: 90–99), 31.08 (95% CI: 8.75–110.41), 0.11 (95% CI: 0.05–0.24), 89% (3%) and 288.81 (95% CI: 78.51–1062.48), respectively. The area under the SROC curve was 0.98 (95% CI: 0.97–0.99). Our meta-analysis showed that ultrasound can play an important role in the diagnosis of ACL injury. Because of its high sensitivity, high specificity and high diagnostic ability, ultrasound should be a part of the standard diagnostic work-up of an ACL injury.     

An accurate estimation of the horizontal acceleration of a rower’s centre of mass using inertial sensors: a validation

For a valid determination of a rower’s mechanical power output, the anterior–posterior (AP) acceleration of a rower’s centre of mass (CoM) is required. The current study was designed to evaluate the accuracy of the determination of this acceleration using a full-body inertial measurement units (IMUs) suit in combination with a mass distribution model. Three methods were evaluated. In the first two methods, IMU data were combined with either a subject-specific mass distribution or a standard mass distribution model for athletes. In the third method, a rower’s AP CoM acceleration was estimated using a single IMU placed at the pelvis.

Experienced rowers rowed on an ergometer that was placed on two force plates, while wearing a full-body IMUs suit. Correspondence values between AP CoM acceleration based on IMU data (the three methods) and AP CoM acceleration obtained from force plate data (reference) were calculated. Good correspondence was found between the reference AP CoM acceleration and the AP CoM accelerations determined using IMU data in combination with the subject-specific mass model and the standard mass model (intraclass correlation coefficients [ICC] >?0.988 and normalized root mean square errors [nRMSE]?3.81%). Correspondence was lower for the AP CoM accelerations determined using a single pelvis IMU (0.877?Based on these results, we recommend determining a rower’s AP CoM acceleration using IMUs in combination with the standard mass model. Finally, we conclude that accurate determination of a rower’s AP CoM acceleration is not possible on the basis of the pelvis acceleration only.     

Ground reaction forces and knee kinetics during single and repeated badminton lunges

Repeated movement (RM) lunge that frequently executed in badminton might be used for footwear evaluation. This study examined the influence of single movement (SM) and RM lunges on the ground reaction forces (GRFs) and knee kinetics during the braking phase of a badminton lunge step. Thirteen male university badminton players performed left-forward lunges in both SM and RM sessions. Force platform and motion capturing system were used to measure GRFs and knee kinetics variables. Paired t-test was performed to determine any significant differences between SM and RM lunges regarding mean and coefficient of variation (CV) in each variable. The kinetics results indicated that compared to SM lunges, the RM lunges had shorter contact time and generated smaller maximum loading rate of impact force, peak knee anterior-posterior force, and peak knee sagittal moment but generated larger peak horizontal resultant forces (Ps < 0.05). Additionally, the RM lunges had lower CV for peak knee medial-lateral and vertical forces (Ps < 0.05). These results suggested that the RM testing protocols had a distinct loading response and adaptation pattern during lunge and that the RM protocol showed higher within-trial reliability, which may be beneficial for the knee joint loading evaluation under different interventions.     

Machine-learning-based head impact subtyping based on the spectral densities of the measurable head kinematics

BackgroundTraumatic brain injury can be caused by head impacts, but many brain injury risk estimation models are not equally accurate across the variety of impacts that patients may undergo, and the characteristics of different types of impacts are not well studied. We investigated the spectral characteristics of different head impact types with kinematics classification.MethodsData were analyzed from 3262 head impacts from lab reconstruction, American football, mixed martial arts, and publicly available car crash data. A random forest classifier with spectral densities of linear acceleration and angular velocity was built to classify head impact types (e.g., football, car crash, mixed martial arts). To test the classifier robustness, another 271 lab-reconstructed impacts were obtained from 5 other instrumented mouthguards. Finally, with the classifier, type-specific, nearest-neighbor regression models were built for brain strain.ResultsThe classifier reached a median accuracy of 96% over 1000 random partitions of training and test sets. The most important features in the classification included both low- and high-frequency features, both linear acceleration features and angular velocity features. Different head impact types had different distributions of spectral densities in low- and high-frequency ranges (e.g., the spectral densities of mixed martial arts impacts were higher in the high-frequency range than in the low-frequency range). The type-specific regression showed a generally higher R² value than baseline models without classification.ConclusionThe machine-learning-based classifier enables a better understanding of the impact kinematics spectral density in different sports, and it can be applied to evaluate the quality of impact-simulation systems and on-field data augmentation.