High-intensity activities in sports like basketball can result in fatigue without proper recovery. This study introduces a collaborative framework that leverages Computer Vision (CV) and Machine Learning for evaluating jump landings and predicting athletic readiness by modelling Countermovement Jumps (CMJs) biomechanical aspects. Seventeen female collegiate basketball athletes of Sacred Heart University (SHU), CT, USA, participated in weekly CMJs over a 26-week season. Through CV-driven semantic analysis of videos, the framework identifies the crucial initial contact and maximum flexion point during jump landings and extracts kinetic and kinematic features of the lower extremities. Next, an inferential analysis is conducted to understand the relationship between these features and the CMJ-driven reactive strength index modified (RSImod) score, which measures fatigue and athletic readiness. An XGBoost regressor, trained on the past week’s data, then predicted the RSImod score for the following week, which resulted in an MSE of 0.020 and an R2 of 0.892. Using SHapley Additive exPlanations (SHAP), the framework offers interpretable feedback, aiding coaches in creating personalised training programs and optimising athletic performance while minimising injury risks.
1 Introduction
Competitive sports like basketball involve jump-landings that involve quick, frequent, and intense movements during practice and games. An athlete’s fatigue and asymmetric postures cause increased joint torque, resulting in greater chances of injuries and deteriorated performance [16][8].
In competitive sports, characterising and evaluating an athlete’s fatigue and readiness non-invasively is crucial for helping coaches and fitness professionals tailor training programs. Vertical jumping is an essential motor skill in basketball [9] and can be used to characterise fatigue. The Countermovement Jump (CMJ) test is an efficient, non-invasive method for assessing jumping performance, neuromuscular fatigue, and readiness. The Reactive Strength Index modified (RSImod) is a CMJ-driven metric that reflects lower extremity explosiveness, combining jump height and contact time to evaluate an athlete’s capacity for generating maximum vertical impulse quickly [4] [10]. Studies show RSImod’s positive association with strength and endurance performance and a negative association with acceleration and change of direction speed [12]. Research also indicates that RSImod is correlated with net impulse and jump height [20], with performance worsening when countermovement depth is limited (less) [19]. Biomechanical analyses of drop jumps and simulations highlight the importance of knee flexion [13], knee kinematics, stance width [22], trunk stability [7], and foot rotation [6], all contributing to enhanced jump performance.
Traditional methods adopted by sports coaches for evaluating athlete jump performance involve manual monitoring and correcting them during training sessions. However, this approach is labour-intensive, error-prone, non-scalable, non-flexible, opaque, and has inter-expert variability [11]. In contrast, recent progress in Machine Learning (ML) and Computer Vision (CV) technologies has transformed sports analytics by enabling non-intrusive monitoring of athletes and developing data analytic models and their analysis through collaborative efforts involving coaches, biomechanical researchers, and data scientists [24]. CV techniques can accurately measure joint movements and forces, providing precise kinematic data crucial for performance analysis and injury prevention without the need for intrusive markers [2]. ML models integrated with CV significantly enhance the detection of movement patterns and prediction of potential injuries by identifying subtle asymmetries often missed by traditional methods [3]. Real-time evaluation systems offer immediate feedback and comprehensive data analysis, optimising training regimens and enabling quick adjustments during sessions [17]. Additionally, integrating artificial intelligence (AI) in CV frameworks allows for the rapid and accurate analysis of large volumes of video data, improving the prediction of injury risks and developing personalised training programs [5]. For instance, a study on sports assist systems collected and analysed sensor data from athletes’ limbs during training and incorporated ML models to identify errors or suggest improvements in athletic movements [24]. Another study utilised cameras to record jump exercises performed by athletes and designed a CV-based framework that analysed lower-limb asymmetries during jump-landing for injury risk assessment [21].
Current studies primarily focus on quantitative fatigue analysis in team sports [20][19][13]. However, no research has assessed athletes’ fatigue and game readiness regarding biomechanical features - lower-extremity joint mechanics in motion and load. This research helps characterise the relationship between the biomechanical features (kinetic-kinematic) extracted during jump-landings and RSImod. The quantitative analysis helps predict athletes’ readiness in RSImod. As shown in Figure 1, we employ a CV-based framework for biomechanical assessment of athletes’ jump landings to predict athletic readiness. The framework uses motion to analyse jump-landing biomechanics and it looks at overall movement quality, symmetry, coordination, and fluidity during jump-landing. As an outcome, it predicts an athlete’s RSImod score for the following week. This comprehensive evaluation aids in developing effective training schedules and game strategies that optimise performance while minimising the risk of injury.
Thus, this study’s primary aims (PAs) are (PA-I): design of a CV-based framework for biomechanical assessment of jump landings to extract lower-extremity kinetic-kinematic features, (PA-II): examining the impact of fatigue on these features and their relationship with the CMJ-derived RSImod score of athletes, (PA-III): prediction of the RSImod score for the following week based on the biomechanical feature using ML algorithms, (PA-IV): plotting SHapley Additive explanations (SHAP) values to offer coaches detailed insights into the specific impact of each biomechanical feature for individual athletes.
Figure 1:
Figure 1: Proposed framework for assessment and prediction of RSImod.
2 Proposed Framework
Figure 1 illustrates a detailed workflow for analysing athletic performance through video data to provide actionable feedback to coaches. The process begins with capturing frontal and lateral videos of an athlete performing the countermovement jumps. These videos are processed using Google’s MediaPipePose [14] with the BlazePose model, which detects 33 key joint coordinates, generating a 2D joint coordinate skeleton for the athlete. The skeleton is then analysed to identify key frames, specifically focusing on critical moments such as initial contact and maximum knee flexion in both the frontal and lateral views.
Once these key frames are identified, feature extraction is performed, yielding distinct kinetic and kinematic biomechanical features that capture various aspects of the athlete’s movements. These features are then subjected to a relationship assessment to determine their correlation with the RSImod. An XGBoost regressor, an ensemble ML algorithm, predicts RSImod values based on the extracted features.
2.1 Subjects
This study involved 17 female basketball athletes of the women’s basketball team at SHU University, with the following demographic characteristics: mean age of 21.00 years (±3.00), height of 174.21 cm (±19.27), and body mass of 73.98 kg (±11.52), competing in AABB Division I. The research received approval from the Institutional Review Board (IRB), and all subjects were provided with information about the study and consented before participating (IRB approval number 170720A on 9/14/2020).
2.2 Procedures
Figure 2:
Figure 2: Camera Set-up for video recording.
We used two Panasonic LUMIX FZ80 - 4K digital cameras (with an 18.1-megapixel sensor, 60X zoom DC VARIO lens with a focal range of 20-1200mm and an aperture range of F2.8-5.9), offering a high resolution and stabilisation for recording the videos from frontal and lateral planes. The athletes performed CMJ over dual force plates (FDlite, Vald Performance, Brisbane, QL, AUS) with a sampling rate of 1000 Hz), as a part of their regular training routine. These cameras were placed 136 inches in front of and to the side of the jump spot. Figure 2 shows the camera setup for video recording. Each CMJ session was scheduled on the same day of the week, typically on Monday, to maintain consistency. Athletes were directed to maximise jump height and minimise ground contact time. After placing their hands on their sides and executing a deep squat, they performed a maximal jump. Three CMJs were performed following a low-intensity warm-up. The highest RSImod score from the three CMJs in a session was considered for analysis.
2.3 Pose Estimation
Initially, pose estimation is performed using Google’s MediaPipe Pose [14]. This lightweight model runs on GPU and CPU devices, ensuring flexibility and high performance. The model generates 33 pose landmarks, each characterized by two coordinates—x and y determine its position in the frame and visibility, indicating the confidence of the estimated key point. Figure 3 shows the 33 body landmarks detected. We utilised 16 body landmarks (11-16 and 23-32) for biomechanical feature extraction.
Figure 3:
Figure 3: Pose estimation landmarks: 0 - nose, 1 - left eye (inner), 2 - left eye, 3 - left eye (outer), 4 - right eye (inner), 5 - right eye, 6 - right eye (outer), 7 - left ear, 8 - right ear, 9 - mouth (left), 10 - mouth (right), 11 - left shoulder, 12 - right shoulder, 13 - left elbow, 14 - right elbow, 15 - left wrist, 16 - right wrist, 17 - left pinky, 18 - right pinky, 19 - left index, 20 - right index, 21 - left thumb, 22 - right thumb, 23 - left hip, 24 - right hip, 25 - left knee, 26 - right knee, 27 - left ankle, 28 - right ankle, 29 - left heel, 30 - right heel, 31 - left foot index, 32 - right foot index [14].
We applied normalisation and scaling techniques to ensure that the model remains impartial to subject size and appearance variations, facilitating broad applicability without reliance on specific details of the input skeleton. The normalised values are calculated once and applied across all frames.
•
Vertical axis (y) normalisation: Divide the given value by the sum of vertical distances between landmarks (11-left shoulder, 23-left knee), (23-left knee, 25-left hip), and (27-left ankle, 25-left knee)—representing the height of the subject’s torso and leg.
•
Horizontal axis (x) normalisation: The values are divided by the sum of horizontal distances between landmarks (15 -right waist, 13-right elbow), (13-right elbow, 11-right shoulder), (11-right shoulder, 12-left shoulder), (12-left shoulder, 14-left elbow), and (14-left elbow, 16-left wrist)—representing the width of the torso and both upper extremities.
2.4 Key Frame Identification
The initial contact (IC) frame is critical in CMJ analysis as it marks the moment of ground contact, enabling precise timing measurements, jump height calculation, and delineation of jump phases. On the other hand, the frame with maximum knee flexion KFmax is crucial for biomechanical analysis. In the next step, we used Algorithm 1 and 2 to extract these key (IC and KFmax) video frames.
Figure 4 demonstrates KFmax identification over a sample video. It shows two plots, one for each knee. The line in blue marks the original values, while the orange marks the rolling average (size = 3). For example, in both cases, we observe that the two consecutive intersection points pair between these lines with the highest distance in frame numbers 6 and 8. The distance between point 8 and its consecutive points (7, 9) is longer than 6 and its consecutive points (5, 7). Hence, opting for the point with the lowest x value (of 8’s neighbours), we conclude frame 7 as the KFmax frame.
Figure 4:
Figure 4: Sample rolling window plot for KFmax identification: (a) right knee, (b) left knee. The x-axis show frame numbers, and the y-axis shows knee flexion angle; blue marks the original values, and orange marks the rolling average (size = 3).
2.5 Biomechanical Feature Extraction
After contact with the ground, the reaction to the load adds to the body weight. Thus, the kinematic results are described by the lower angular movements of the body. In the next stage, we performed lower extremity kinetic-kinematic feature extraction. (PA-I). Equation 1 shows the formula for cosine angle calculation (for knee flexion angle) and 2 for Euclidean distance calculation (for stance width).
\(\overrightarrow{L_{23, 25}}\) denotes the vector between the landmarks 23 (left hip) and 25 (left knee).
•
\(\overrightarrow{L_{25, 27}}\) denotes the vector between the landmarks 25 (left knee) and 27 (left ankle).
2.5.1 Knee Flexion Angle (frontal and lateral):
(1)
Compute the left knee flexion angle using the cosine of coordinates for points (23-left hip, 25-left knee, 27-left ankle).
(2)
Compute the right knee flexion angle using the cosine of coordinates for points (24-right hip, 26-right knee, 28-right ankle).
2.5.2 Countermovement Depth (lateral):
Determine the countermovement depth by calculating the hip keypoint’s vertical displacement (23-left knee or 24-right knee) from the start to the lowest point during the countermovement.
2.5.3 Hip Flexion Angle (frontal and lateral):
(1)
Compute the left hip flexion angle using the cosine of coordinates for points (11-left shoulder, 23-left hip, 25-left hip) (refer equation 1).
(2)
Compute the right hip flexion angle using the cosine of coordinates for points (12-right shoulder, 24-right hip, 26-right knee).
2.5.4 Trunk Flexion Angle (frontal and lateral):
(1)
Compute the left trunk flexion angle using the cosine of coordinates (shoulder, hip, knee) for points (11, 23, 25).
(2)
Compute the right trunk flexion angle using the cosine coordinates for points (12, 24, 26).
2.5.5 Lateral Trunk Flexion Angle (frontal):
Determine the angle formed by two lines: one from the hip midpoint to the top of the frame and the other connecting the hip midpoint to the shoulder midpoint.
2.5.6 Ankle Plantar Flexion (frontal and lateral):
Assess the angle between a reference line from the heel to the bottom of the frame and a line connecting the heel (29 or 30) and footindex (31, 32) key points.
2.5.7 Stance Width (frontal):
Calculate the ratio of the Euclidean distance between the left and right shoulder key points (11, 12) to the distance between the left and right ankle key points (27, 28) as shown in equation 2. Stance width could be normal \((1)\), narrow (> 1), and wide (< 1).
Examine the y-coordinate of the left and right footindex (31 or 32) to assess whether the feet landed simultaneously. If the feet land simultaneously, we assign the value 1 else 0.
2.5.9 Foot Orientation (frontal):
Assess the angle between a reference line from the heel to the bottom of the frame and a line connecting the heel and footindex key points. The orientation can be toe-in (internal rotation), toe-out (external rotation), or neutral (no rotation).
Figure 5:
Figure 5: Biomechanical feature extraction: (a) Foot orientation, (b) Knee flexion and trunk flexion angles, (c) Lateral trunk flexion angle, and (d) Hip flexion angle, countermovement depth.
2.6 Visualisation over Kinovea
Figure 5 visually represents some of the features. Figure 5(a) depicts the calculation of foot rotation (internal or external), which reflects foot orientation. The right foot of the athlete appears externally rotated, while the left, however slightly internally rotated, is in a near-neutral orientation position. Figure 5(b) demonstrates the knee and trunk flexion angles. Figure 5(c) shows lateral trunk flexion, where the athlete has turned towards the right, resulting in a 3° flexion. Figure 5(d) demonstrates the hip flexion angle and relative countermovement depth (297.71px) as the difference between the hip and the reference keypoint on the force plate. We used the software Kinovea (www.kinovea.org, version 0.8.25) [18] to visualise angles and distances.
2.7 Athletic Readiness Prediction
Next, we conducted statistical analysis to assess and visualise the marginal effects of the extracted features on the RSImod score. It was observed that there was an imbalance in the distribution of data records, with only 38% attributed to high and very high RSImod scores. In comparison, the remaining 62% belonged to low and moderate RSImod scores. Hence, we augmented the dataset using the data balancing technique - Synthetic Minority Over Sampling with Gaussian Noise (SMOGN) [1] and then trained an XGBoost regressor over the dataset for RSImod prediction based on these extracted features.
2.8 Model Interpretations
We generated SHAP [15] values for these features to provide detailed individual insights into the specific impact of each biomechanical feature for athletes to coach. SHAP excels in identifying joint effects and interactions, providing a comprehensive view of how biomechanical elements collaborate, which is valuable for capturing complexity in biomechanical analyses, and contributing to individualised and global understandings of feature contributions to RSImod scores.
3 Results and Discussion
A total of 84 countermovement jumps videos were subjected to analysis. The RSImod score was quantified in mm/ms (from force plate) as an outcome of the CMJ assessments conducted weekly over a duration spanning from week 1 to week 26. The videos were recorded at a resolution of 1920×1080 pixels. Each jump session was approximately 30 seconds long. The frame rate set was 70 fps, resulting in approximately 210 frames (average 200) per session.
The mean RSImod score was 0.37, with a standard deviation (SD) of 0.08. To validate the effectiveness of our developed framework, we computed the RSImod score from the CMJ videos (addressing PA-I). Specifically, we quantified the jump height by measuring the vertical displacement of the hip keypoint (23 or 24) from the starting position to the peak of the jump. Frames corresponding to take-off and landing moments were pinpointed by analyzing body posture or velocity changes, with timestamps recorded for precise timing. The contact time was then computed by subtracting the take-off timestamp from the landing timestamp. Using the following equation, we computed the RSImod scores.
This resulted in a mean RSImod score of 0.34 with 0.05 SD. Notably, the correlation (r) between force plate-measured and framework-measured RSImod scores was a high 0.94, affirming the framework’s accuracy. Table 1 details the statistical significance of the numeric features for the RSImod score in terms of Pearson’s correlation coefficient (r) and the effect size (ES). ES is a quantitative measure of the strength or magnitude of a relationship or difference between variables in a study. It was used to quantify the real-world impact of biomechanical features on athletic performance regarding the magnitude of association [23].
Table 1:
Feature
Pearson’s Correlation Coefficient (r)
Effect Size (ES)
Countermovement Depth
0.781
Large
Knee Flexion Angle
0.683
Large
Trunk Flexion Angle
0.672
Large
Hip Flexion Angle
0.670
Large
Ankle Plantar Flexion
-0.581
Medium
Lateral Trunk Flexion
-0.721
Large
Table 1: Feature correlation and effect size.
3.1 Inferential Analysis (addressing PA-II):
Knee Flexion Angle:
Figure 6 shows the relationship between knee flexion angle and RSImod. The analysis reveals a positive correlation between increased knee flexion and higher RSImod scores. Optimal knee flexion, implying a deeper squat position during landing, positively influences jump height. It is observed that, on average, Initial knee flexion typically ranges from 20° to 30°, with the majority of data points falling within the +1SD range of RSImod scores. At maximum flexion, data points predominantly range from 60° to 70°. RSImod increases with knee flexion, although a decrease is above 85°.
Figure 6:
Figure 6: Relationship assessment knee flexion angle vs RSImod (+).
Countermovement Depth.
Countermovement depth positively correlates with RSImod as shown in Figure 7. Deeper countermovement allows for increased potential energy storage, enhancing force production and leading to a more robust and explosive upward movement during the jump.
We observe a positive correlation between increased Hip Flexion and elevated RSImod scores in Figure 8. Adequate hip flexion is identified as a contributing factor to efficient energy storage and release, ultimately leading to a more powerful jump. The initial hip flexion typically ranges from 30° to 45°, with most data points falling within the +1SD range of RSImod scores. At KFmax, an increase in RSImod is noted with higher hip flexion (160°-170°). However, above 1SD, some cases exhibit below-average RSI scores, suggesting the need for optimal hip flexion.
Figure 8:
Figure 8: Relationship assessment hip flexion vs RSImod (+).
Trunk Flexion Angle.
We observe a positive correlation between increased Trunk Flexion and elevated RSImod scores in Figure 9. The observation suggests that bending forward is crucial in engaging key muscle groups, facilitating elastic element preloading, and enhancing the force for upward propulsion during the jump. Data analysis indicates that initial trunk flexion typically ranges from 0° to 20°. Both values above and below 1SD result in low RSI scores. At KFmax, the observed range extends from 60° to 80°, with values below and above this range associated with low RSI scores. Figure 9 shows the relationship between trunk flexion angle and RSImod.
Figure 9:
Figure 9: Relationship assessment trunk flexion vs RSImod (+).
Ankle Plantar Flexion.
Ankle plantar flexion negatively correlates with RSImod (Figure 10). Excessive ankle plantar flexion negatively impacts force production efficiency during the push-off phase, emphasising the importance of optimal ankle plantar flexion for power generation.
Figure 11 shows that lateral trunk flexion negatively correlates with RSImod. There must be minimal lateral movement during jump-landings, ideally close to zero degrees. Excessive lateral movement disrupts force distribution, compromising the coordinated engagement of muscles and elastic elements involved in the countermovement.
Figure 11:
Figure 11: Relationship assessment lateral trunk flexion vs RSImod (-).
Stance Width.
A shoulder width to a slightly wider stance enhances force transfer, optimizing force production and biomechanical balance for an improved RSImod (Figure 12). A narrow stance may hinder RSImod, limiting the body’s capacity for efficient force generation and transfer in the countermovement, potentially resulting in a low RSImod score.
Figure 12:
Figure 12: Relationship assessment stance width vs RSImod.
Foot Symmetricity.
In Figure 13, We observed a positive correlation with RSImod. Symmetric landing with equal force distribution on both feet contributes to balanced force distribution, optimizing force production and transmission mechanics during the countermovement.
Figure 13:
Figure 13: Relationship assessment foot symmetricity vs. RSImod (+).
Foot Orientation.
It is observed in Figure 14 that individuals with a neutral or slightly outward foot orientation tend to achieve higher RSI scores. Optimal foot orientations positively impact force transmission, while less ideal foot orientations, such as toe-in positions, may be associated with disruptions in force generation and transmission, potentially leading to poor RSI scores.
Figure 14:
Figure 14: Relationship assessment foot rotation vs RSImod.
3.2 RSImod Prediction (addressingPA-III)
As a preprocessing step, we applied the SMOGN [1] technique to the dataset to address the imbalance issue, resulting in an augmented overall sample size of 219 records. After the rebalancing process, 50% of the records were now associated with high and very high levels of athlete readiness, and the remaining 50% represented low and moderate levels. We employed a stratified 10-fold cross-validation approach for model generalization. The dataset was partitioned into training (70%) and testing (30%) sets. Specifically, 154 records were utilised for training, and 65 were reserved for testing. The performance evaluation of the XGBoost regressor (with hyperparameter values colsamplebytree = 0.8, gamma = 0.1, maxdepth = 5, minchildweight = 1, nestimators = 200; as identified using grid search), trained using RSImod as the target feature, yielded a Mean Squared Error (MSE) of 0.020 (ideal 0) and adjusted R2 of 0.892.
Table 2:
Study
Qualitative Result
Quantitative Result
Key Finding
Limitation
Padua et al. (2009)
Identifies landing errors in vertical jumps.
Validity: α = 0.92; ICC: inter = 0.85, intra = 0.90.
Correlation with force plate readings r=0.94; adjusted R² of 0.892, MSE of 0.020, and MAE of 0.015
Validates jump mechanics in real-time for readiness
Requires computational resources for deployment
Table 2: Comparison of proposed framework with state-of-the-art techniques.
Table 2 summarises a comparison of proposed work with state-of-the-art in terms of qualitative and quantitative outcomes. The proposed study achieved a mean RSImod score of 0.34 with a strong correlation (r = 0.94) between force plate and framework measurements, surpassing the inter-rater reliability of 0.85 reported by Padua et al. (2009). Significant effect sizes for biomechanical features, particularly countermovement depth (0.781) and lateral trunk flexion (-0.721), underscore the framework’s robustness in evaluating jump mechanics. Additionally, the application of machine learning to predict RSImod, with an adjusted R² of 0.892, introduces a powerful new method for assessing athlete readiness in real time.
Figure 15:
Figure 15: SHAP value explanation of model outcomes - features are listed by importance, with positive SHAP values indicating a positive impact on the prediction and negative values indicating a negative impact.
3.3 Interpretability and Explanation (addressing PA-IV)
We plotted SHAP values to break down the model’s prediction into contributions from each feature, quantifying each feature’s impact on athletic readiness prediction [15]. Figure 15 presents model-average SHAP values. Significant features, such as trunk flexion angle at maximum knee flexion, hip flexion angle, countermovement depth, and ankle plantar flexion, correlate with biomechanical principles widely recognised in sports science. The emphasis on trunk flexion at KFmax is crucial as it reflects the forward bending of the upper body, engaging essential muscle groups and enhancing elastic element preloading for potent upward propulsion. The negative impact of excessive trunk flexion after landing aligns with coaching principles that discourage such post-landing movements. Adequate hip flexion is recognised for efficient energy storage, supporting optimal power generation.
In contrast, the caution against excessive hip flexion echoes established knowledge on avoiding inefficiencies in force production. The positive correlation of countermovement depth with RSImod resonates with the understanding that a deeper countermovement allows for increased potential energy storage, facilitating a more potent upward movement. The inverse correlation of ankle plantar flexion with RSImod emphasises the significance of maintaining proper ankle positioning for efficient force production during the push-off phase of a jump.
3.4 Feedback to coach
The study presents outcomes to coaches using visual dashboards and individual athlete profiles, featuring SHAP value plots and historical RSImod trends for quick overviews. Comparative charts highlight performance patterns, and video overlays show biomechanical adjustments during jumps. These tools support the creation of individualized training programs, aid in identifying features affecting RSImod scores for injury prevention, and enable efficient allocation of training resources by focusing on critical biomechanical factors that influence athletic readiness.
4 Conclusion
This CV-driven semantic analysis of videos is a significant improvement, providing an innovative and holistic approach to sports performance evaluation compared to conventional, manual-centric techniques. By deciphering the intricate relationship between biomechanical features and RSImod, athletes and coaches can fine-tune training regimens to enhance reactive strength. It optimises performance and plays a pivotal role in injury prevention. With RSImod prediction and SHAP explanations, the individualised nature of this predictive approach ensures that training programs are tailored to an athlete’s unique characteristics, fostering targeted improvements.
The framework, on average, can analyse a single video in approximately 90 seconds (with the potential for further reduction on more advanced hardware configurations) and predict an RSImod score with an MSE of 0.020 and R2 of 0.892. Jump tests, a standard practice in sports like basketball, volleyball, and soccer for assessing athlete readiness, can benefit from this framework. The suggested two-camera arrangement is convenient for implementation in athletic training facilities, expanding the system’s applicability.
We believe our approach of continuously monitoring the biomechanical features and their impact on the RSImod (weekly) facilitates real-time adjustments, enabling athletes to track progress and make informed decisions about their training journey. Coaches gain valuable insights into biomechanical features, empowering them to provide precise feedback and interventions. As the framework becomes more robust with the coming seasons’ data, it could replace force plates as we could predict athletic readiness (RSImod score) for athletes only based on their biomechanical features.
References
[1]
Pedro Branco, Luís Torgo, and Rita P Ribeiro. 2017. Smogn: A pre-processing approach for imbalanced regression. In Proceedings of the First International Workshop on Learning with Imbalanced Domains: Theory and Applications. 36–50.
Steffan L Colyer, Marcus Evans, Darren P Cosker, and et al.2018. A review of the evolution of vision-based motion analysis and the integration of advanced computer vision methods towards developing a markerless system. Sports Medicine - Open 4, 24 (2018).
Ryan J Cotton, Austin DeLillo, Andrew Cimorelli, Khushbu Shah, John D Peiffer, Sumit Anarwala, and et al.2023. Markerless motion capture and biomechanical analysis pipeline. arXiv preprint arXiv:https://arXiv.org/abs/2303.10654 (2023).
Stefano David, Bruno Bačić, Christoph Richter, and Martin Mundt. 2023. Editorial: Artificial intelligence to enhance biomechanical modelling. Frontiers in Sports and Active Living (2023).
Lindsay J DiStefano, Stephen W Marshall, Darin A Padua, Kelli Y Peck, Andreas I Beutler, Sarah J de la Motte, and et al.2016. The effects of an injury prevention program on landing biomechanics over time. American Journal of Sports Medicine 44, 3 (2016), 767–776.
Rodrigo G Gheller, Jane Dal Pupo, Jeruza Ache-Dias, Dalmar Detanico, Johnny Padulo, and Sidnei G Dos Santos. 2015. Effect of different knee starting angles on intersegmental coordination and performance in vertical jumps. Human Movement Science 42 (2015), 71–80.
Lian Guo, Jing Zhang, Yao Wu, and Li Li. 2021. Prediction of the risk factors of knee injury during drop-jump landing with core-related measurements in amateur basketball players. Frontiers in Bioengineering and Biotechnology 9 (2021), 738311.
Adam Heishman, Braden Brown, Benjamin Daub, Richard Miller, Eduardo Freitas, and Michael Bemben. 2019. The influence of countermovement jump protocol on reactive strength index modified and flight time: contraction time in collegiate basketball players. Sports 7, 2 (2019), 37.
Arnon Hershkovitz and Orit Karni. 2018. Borders of change: A holistic exploration of teaching in one-to-one computing programs. Computers & Education 125 (2018), 429–443.
Paul Jarvis, Anthony Turner, Paul Read, and Chris Bishop. 2022. Reactive strength index and its associations with measures of physical and sports performance: A systematic review with meta-analysis. Sports Medicine 52, 2 (2022), 301–330.
Kristof Kipp, Mitchell T Kiely, Michael D Giordanelli, Patrick J Malloy, and Christian F Geiser. 2018. Biomechanical determinants of the reactive strength index during drop jumps. International Journal of Sports Physiology and Performance 13, 1 (2018), 44–49.
Marinos Konstantinou, Georgios Retsinas, and Petros Maragos. 2023. Enhancing action recognition in vehicle environments with human pose information. In Proceedings of the 16th International Conference on Pervasive Technologies Related to Assistive Environments. 197–205.
Aniruddha Lalwani, Amey Saraiya, Akshat Singh, Anushri Jain, and Tanmoy Dash. 2022. Machine learning in sports: A case study on using explainable models for predicting outcomes of volleyball matches. arXiv preprint arXiv:https://arXiv.org/abs/2206.09258 (2022).
Gregory D Myer, Dennis A Chu, Jessica L Brent, and Timothy E Hewett. 2008. Trunk and hip control neuromuscular training for the prevention of knee joint injury. Clinics in sports medicine 27, 3 (2008), 425–448.
Bhaskar Thimmegowda Naik, Mohd Farukh Hashmi, and Nitin Dutt Bokde. 2022. A comprehensive review of computer vision in sports: Open issues, future trends and research directions. Applied Sciences 12, 9 (2022), 4429.
Norsyahida Mohd Nor Adnan, Mohd Nazri Abd Ab Patar, Hee Lee, Shoichi I Yamamoto, Lee Jong-Young, and Jamilah Mahmud. 2018. Biomechanical analysis using Kinovea for sports application. In IOP Conference Series: Materials Science and Engineering, Vol. 342. IOP Publishing, 012097.
Andre S Salles, Vasilios Baltzopoulos, and Juergen Rittweger. 2011. Differential effects of countermovement magnitude and volitional effort on vertical jumping. European Journal of Applied Physiology 111, 3 (2011), 441–448.
Alberto Sánchez-Sixto, Andrew J Harrison, and Pablo Floría. 2018. Larger countermovement increases the jump height of countermovement jump. Sports 6, 4 (2018), 131.
Srishti Sharma, Srikrishnan Divakaran, Tolga Kaya, Christopher Taber, and Mehul S Raval. 2023. A Framework for Biomechanical Analysis of Jump Landings for Injury Risk Assessment. In 2023 IEEE 28th Pacific Rim International Symposium on Dependable Computing (PRDC). IEEE, 327–331.
Yuh-Jen Tsai, Chien-Chi Chia, Pei-Ying Lee, Li-Ching Lin, and Yu-Ling Kuo. 2020. Landing kinematics, sports performance, and isokinetic strength in adolescent male volleyball athletes: Influence of core training. Journal of Sport Rehabilitation 29 (2020), 65–72.
Jian Zhang, Jian Sun, Jing Wang, and Xiaoguang Yue. 2021. Visual object tracking based on residual network and cascaded correlation filters. Journal of Ambient Intelligence and Humanized Computing 12 (2021).
Most work on predicting the outcome of basketball matches so far has focused on National College Athletics Association Basketball NCAAB games. Since NCAAB and professional National Basketball Association, NBA basketball have a number of differences, it ...
ICCIP '16: Proceedings of the 2nd International Conference on Communication and Information Processing
Practicing and playing a sport causes athletes' bodies to adapt to the movements they regularly perform. Unfortunately, this can cause muscle imbalances, which might impair performance or worse, cause an injury. It is always best to find the root cause ...
MMSports '23: Proceedings of the 6th International Workshop on Multimedia Content Analysis in Sports
Advanced data collection technologies, computational tools, and sophisticated algorithms have a revolutionary impact on sports analytics on various aspects of sports, from athletes performance to fan engagement. Computer Vision (CV) and Deep Learning (...
