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

Can an inertial measurement unit, combined with machine learning, accurately measure ground reaction forces in cricket fast bowling?

This study examined whether an inertial measurement unit (IMU) could measure ground reaction force (GRF) during a cricket fast bowling delivery. Eighteen male fast bowlers had IMUs attached to their upper back and bowling wrist. Each participant bowled 36 deliveries, split into three different intensity zones: low = 70% of maximum perceived bowling effort, medium = 85%, and high = 100%. A force plate was embedded into the bowling crease to measure the ground truth GRF. Three machine learning models were used to estimate GRF from the IMU data. The best results from all models showed a mean absolute percentage error of 22.1% body weights (BW) for vertical and horizontal peak force, 24.1% for vertical impulse, 32.6% and 33.6% for vertical and horizontal loading rates, respectively. The linear support vector machine model had the most consistent results. Although results were similar to other papers that have estimated GRF, the error would likely prevent its use in individual monitoring. However, due to the large differences in raw GRFs between participants, researchers may be able to help identify links among GRF, injury, and performance by categorising values into levels (i.e., low and high).

The relationship between bowling intensity and ground reaction force in cricket pace bowlers

This study examined the relationship between perceived bowling intensity, ball release speed and ground reaction force (measured by peak force, impulse and loading rate) in male pace bowlers. Twenty participants each bowled 36 deliveries, split evenly across three perceived intensity zones: low = 70% of maximum perceived bowling effort, medium = 85%, and high = 100%. Peak force and loading rate were significantly different across the three perceived intensity zones in the horizontal and vertical directions (Cohen's d range = 0.14-0.45, p < 0.01). When ball release speed increased, peak force and loading rate also increased in the horizontal and vertical directions (η_p² = 0.04-0.18, p < 0.01). Lastly, bowling at submaximal intensities (i.e., low - medium) was associated with larger decreases in peak horizontal force (7.9-12.3% decrease), impulse (15.8-21.4%) and loading rate (7.4-12.7%) compared to decreases in ball release speed (5.4-8.3%). This may have implications for bowling strategies implemented during training and matches, particularly for preserving energy and reducing injury risk.

Automatic detection of passing and shooting in water polo using machine learning: a feasibility study

There is currently no efficient way to quantify overhead throwing volume in water polo. Therefore, this study aimed to test the feasibility of a method to detect passes and shots in water polo automatically using inertial measurement units (IMU) and machine-learning algorithms. Eight water polo players wore one IMU sensor on the wrist (dominant hand) and one on the sacrum during six practices each. Sessions were filmed with a video camera and manually tagged for individual shots or passes. Data were synchronised between video tagging and IMU sensors using a cross-correlation approach. Support vector machine (SVM) and artificial neural networks (ANN) were compared based on sensitivity and specificity for identifying shots and passes. A total of 7294 actions were identified during the training sessions, including 945 shots and 5361 passes. Using SVM, passes and shots together were identified with 94.4% (95%CI = 91.8-96.4) sensitivity and 93.6% (95%CI = 91.4-95.4) specificity. Using ANN yielded similar sensitivity (93.0% [95%CI = 90.1-95.1]) and specificity (93.4% [95%CI = 91.1 = 95.2]). The results suggest that this method of identifying overhead throwing motions with IMU has potential for future field applications. A set-up with one single sensor at the wrist can suffice to measure these activities in water polo.

Automated recognition of the cricket batting backlift technique in video footage using deep learning architectures

There have been limited studies demonstrating the validation of batting techniques in cricket using machine learning. This study demonstrates how the batting backlift technique in cricket can be automatically recognised in video footage and compares the performance of popular deep learning architectures, namely, AlexNet, Inception V3, Inception Resnet V2, and Xception. A dataset is created containing the lateral and straight backlift classes and assessed according to standard machine learning metrics. The architectures had similar performance with one false positive in the lateral class and a precision score of 100%, along with a recall score of 95%, and an f1-score of 98% for each architecture, respectively. The AlexNet architecture performed the worst out of the four architectures as it incorrectly classified four images that were supposed to be in the straight class. The architecture that is best suited for the problem domain is the Xception architecture with a loss of 0.03 and 98.2.5% accuracy, thus demonstrating its capability in differentiating between lateral and straight backlifts. This study provides a way forward in the automatic recognition of player patterns and motion capture, making it less challenging for sports scientists, biomechanists and video analysts working in the field.

External workload intensity in cricket fast bowlers across maximal and submaximal intensities: Modifying PlayerLoad and IMU location

Workload is a commonly accepted risk factor for injury among fast bowlers, however many methods exist to characterise workload. Recently, automated intensity-sensitive measures like PlayerLoad have been used to improve the estimation of workload in fast bowlers. The purpose of this study was to determine whether similar variables could be extracted from a single inertial measurement unit (IMU) that highly correlate with intensity, according to release speed. Eight elite and pre-elite bowlers participated in the study, with each bowler bowling one over each at 60%, 80% and 100% intensity and repeating this across two sessions (36 balls per participant). IMUs were placed on the upper-back and non-bowling wrist and maximum PlayerLoad from each delivery (PL_max) was compared to the accumulated value across each delivery (PL_acc). The strongest correlation with release speed was with PL_acc from the non-bowling wrist (R = 0.74), followed by PL_acc from the upper-back (R = 0.65) and PL_max from the upper back (R = 0.60). Consequently, an improved estimation of the intensity at which bowlers are working at could be gained by examining accumulated PlayerLoad values from an IMU on the non-bowling wrist.

Quantifying cricket fast bowling volume, speed and perceived intensity zone using an Apple Watch and machine learning

This study examined whether an inertial measurement unit (IMU) and machine learning models could accurately measure bowling volume (BV), ball release speed (BRS), and perceived intensity zone (PIZ). Forty-four male pace bowlers wore a high measurement range, research-grade IMU (SABELSense) and a consumer-grade IMU (Apple Watch) on both wrists. Each participant bowled 36 deliveries, split into two different PIZs (Zone 1 = 70-85% of maximum bowling effort, Zone 2 = 100% of maximum bowling effort). BRS was measured using a radar gun. Four machine learning models were compared. Gradient boosting models had the best results across all measures (BV: F-score = 1.0; BRS: Mean absolute error = 2.76 km/h; PIZ: F-score = 0.92). There was no significant difference between the SABELSense and Apple Watch on the same hand when measuring BV, BRS, and PIZ. A significant improvement in classifying PIZ was observed for IMUs located on the dominant wrist. For all measures, there was no added benefit of combining IMUs on the dominant and non-dominant wrists.

Quantification of the demands of cricket bowling and the relationship to injury risk: a systematic review

BACKGROUND

Bowling in cricket is a complex sporting movement which, despite being well characterised, still produces a significant number of injuries each year. Fast bowlers are more likely to be injured than any other playing role. Frequency, duration, intensity and volume of bowling, which have been generalised as measurements of workload, are thought to be risk factors for injuries. Injury rates of fast bowlers have not reduced in recent years despite the implementation of various workload monitoring practices.

OBJECTIVE

To identify the variables used to quantify frequency, intensity, time and volume of bowling; and evaluate relationships between these variables and injury risk.

METHODS

Six online databases were systematically searched for studies on fast bowling that included terms related to workload. Population characteristics, variables relating to demand and their relationship to standardised definitions of physical activity were extracted from all included studies.

RESULTS

Bowling workload is typically quantified through measures of frequency, duration, or indirect intensity, with few studies reporting on bowling volume.

CONCLUSIONS

When reported on, volume was often described using imprecise or insufficient measures of intensity. There is a need to develop more appropriate measures of intensity during bowling and improve the quality of evidence to inform on bowling programme management practices.

Is machine learning and automatic classification of swimming data what unlocks the power of inertial measurement units in swimming?

Researchers have heralded the power of inertial sensors as a reliable swimmer-centric monitoring technology, however, regular uptake of this technology has not become common practice. Twenty-six elite swimmers participated in this study. An IMU (100Hz/500Hz) sensor was secured in the participant's third lumbar vertebrae. Features were extracted from swimming data using two techniques: a novel intrastroke cycle segmentation technique and conventional sliding window technique. Six supervised machine learning models were assessed on stroke prediction performance. Models trained using both feature extraction methods demonstrated high performance (≥ 0.99 weighted average precision, recall, F1-score, area under ROC curve and accuracy), low computational training times (< 3 seconds - bar XGB and when hyperparameters were tuned) and low computational prediction times (< 1 second). Significant differences were observed in weighted average stroke prediction F1-score (p = 0.0294) when using different feature extraction methods and model computational training time (p = 0.0007), and prediction time (p = 0.0026) when implementing hyperparameter tuning. Automatic swimming stroke classification offers benefits to observational coding and notational analysis, and opportunities for automated workload and performance monitoring in swimming. This stroke classification algorithm could be the key that unlocks the power of IMUs as a biofeedback tool in swimming.

Can Machine Learning with IMUs Be Used to Detect Different Throws and Estimate Ball Velocity in Team Handball?

Injuries in handball are common due to the repetitive demands of overhead throws at high velocities. Monitoring workload is crucial for understanding these demands and improving injury-prevention strategies. However, in handball, it is challenging to monitor throwing workload due to the difficulty of counting the number, intensity, and type of throws during training and competition. The aim of this study was to investigate if an inertial measurement unit (IMU) and machine learning (ML) techniques could be used to detect different types of team handball throws and predict ball velocity. Seventeen players performed several throws with different wind-up (circular and whip-like) and approach types (standing, running, and jumping) while wearing an IMU on their wrist. Ball velocity was measured using a radar gun. ML models predicted peak ball velocity with an error of 1.10 m/s and classified approach type and throw type with 80-87% accuracy. Using IMUs and ML models may offer a practical and automated method for quantifying throw counts and classifying the throw and approach types adopted by handball players.

Can an inertial measurement unit (IMU) in combination with machine learning measure fast bowling speed and perceived intensity in cricket?

This study examined whether an inertial measurement unit (IMU), in combination with machine learning, could accurately predict two indirect measures of bowling intensity through ball release speed (BRS) and perceived intensity zone (PIZ). One IMU was attached to the thoracic back of 44 fast bowlers. Each participant bowled 36 deliveries at two different PIZ zones (Zone 1 = 24 deliveries at 70% to 85% of maximum perceived bowling effort; Zone 2 = 12 deliveries at 100% of maximum perceived bowling effort) in a random order. IMU data (sampling rate = 250 Hz) were downsampled to 125 Hz, 50 Hz, and 25 Hz to determine if model accuracy was affected by the sampling frequency. Data were analysed using four machine learning models. A two-way repeated-measures ANOVA was used to compare the mean absolute error (MAE) and accuracy scores (separately) across the four models and four sampling frequencies. Gradient boosting models were shown to be the most consistent at measuring BRS (MAE = 3.61 km/h) and PIZ (F-score = 88%) across all sampling frequencies. This method could be used to measure BRS and PIZ which may contribute to a better understanding of overall bowling load which may help to reduce injuries.

Biomechanical risk factors of lower back pain in cricket fast bowlers using inertial measurement units: a prospective and retrospective investigation

Objectives

To investigate spinal kinematics, tibial and sacral impacts during fast bowling, among bowlers with a history of low back pain (LBP) (retrospective) and bowlers who developed LBP in the follow-up season (prospective).

Methods

35 elite male fast bowlers; senior (n=14; age=24.1±4.3 years; height=1.89±0.05 m; weight=89.2±4.6 kg) and junior (n=21; age=16.9±0.7; height=1.81±0.05; weight=73.0±9.2 kg) were recruited from professional county cricket clubs. LBP history was gathered by questionnaire and development of LBP was monitored for the follow-up season. Spinal kinematics, tibial and sacral impacts were captured using inertial measurement units placed over S1, L1, T1 and anteromedial tibia. Bonferroni corrected pairwise comparisons and effect sizes were calculated to investigate differences in retrospective and prospective LBP groups.

Results

Approximately 38% of juniors (n=8) and 57% of seniors (n=8) reported a history of LBP. No differences were evident in spinal kinematics or impacts between those with LBP history and those without for seniors and juniors. Large effect sizes suggest greater rotation during wind-up (d=1.3) and faster time-to-peak tibial impacts (d=1.5) in those with no history of LBP. One junior (5%) and four (29%) seniors developed LBP. No differences were evident in spinal kinematics or impacts between those who developed LBP and those who did not for seniors. In seniors, those who developed LBP had lower tibial impacts (d=1.3) and greater lumbar extension (d=1.9) during delivery.

Conclusion

Retrospective analysis displayed non-significant differences in kinematics and impacts. It is unclear if these are adaptive or impairments. Prospective analysis demonstrated large effect sizes for lumbar extension during bowling suggesting a target for future coaching interventions.

Auto detecting deliveries in elite cricket fast bowlers using microsensors and machine learning

Cricket fast bowlers are at a high risk of injury occurrence, which has previously been shown to be correlated to bowling workloads. This study aimed to develop and test an algorithm that can automatically, reliably and accurately detect bowling deliveries. Inertial sensor data from a Catapult OptimEye S5 wearable device was collected from both national and international level fast bowlers (n = 35) in both training and matches, at various intensities. A machine-learning based approach was used to develop the algorithm. Outputs were compared with over 20,000 manually recorded events. A high Matthews correlation coefficient (r = 0.945) showed very good agreement between the automatically detected bowling deliveries and manually recorded ones. The algorithm was found to be both sensitive and specific in training (96.3%, 98.3%) and matches (99.6%, 96.9%), respectively. Rare falsely classified events were typically warm-up deliveries or throws preceded by a run. Inertial sensors data processed by a machine-learning based algorithm provide a valid tool to automatically detect bowling events, whilst also providing the opportunity to look at performance metrics associated with fast bowling. This offers the possibility to better monitor bowling workloads across a range of intensities to mitigate injury risk potential and maximise performance.