BadminSense: Enabling Fine-Grained Badminton Stroke Evaluation on a Single Smartwatch
BadminSense is a smartwatch-based system for fine-grained badminton stroke evaluation that aims to provide amateur players with professional-quality coaching feedback without requiring expensive external equipment. The system uses a single commercial smartwatch on the dominant wrist to segment and classify four stroke types, predict stroke quality on a 5-point Likert scale, and estimate shuttle impact location on the racket string area. The key innovation is enabling fine-grained quality assessment beyond simple activity recognition, targeting the gap between basic fitness tracking and professional coaching.
BadminSense presents a well-motivated and technically sound contribution to sports HCI, achieving promising offline results: 91.43% stroke classification accuracy, 0.438 MAE for quality rating (5-point scale), and 12.9% average impact location error. However, the work is limited by a small dataset (12 participants, 848 strokes), lack of longitudinal validation, and several unverified claims about generalizability. The formative methodology is solid, but the evaluation lacks quantitative validation against expert ground truth during real gameplay.
The formative exploration with badminton players is rigorous and well-documented, yielding actionable design requirements (DR1-DR4) and implementation insights. The ink-marking methodology for impact location ground truth is clever and reproducible. The two-step stroke segmentation using IMU+acoustic fusion achieves 99.41% accuracy with 0.23% false positive rate. The choice of stroke-specific evaluation strategies (I1) is technically justified and shows thoughtful feature engineering. The dataset is open-sourced, supporting reproducibility.
Several limitations undermine the paper's claims. First, the 'user-independent' evaluation uses leave-3-user-out cross-validation (only 9 train/3 test), which is insufficient for robust generalization claims with $N=12$ total participants. Second, the comparison to Silent Impact (Park et al., UIST 2024) is misleading: BadminSense claims 91.43% accuracy versus Silent Impact's 88.2%, but Silent Impact uses the PASSIVE arm (non-racket arm), which is a significantly harder sensing problem than BadminSense's dominant-arm placement. The paper also lacks validation that the 0.438 rating error (on a 1-5 scale) is perceptually meaningful - does a 0.5-point difference matter to learners? Finally, the rule-based feedback generation (I3) was not evaluated against actual coaching advice quality.
The empirical evidence supports basic feasibility but has gaps. The ICC reliability scores for ratings (0.790-0.865) establish reasonable inter-rater agreement. However, the paper overstates novelty: wrist-worn stroke classification was demonstrated by Anand et al. (2017) and Ganser et al. (2021) prior to this work. The quality rating task is novel but lacks a validated baseline - what do 'good' vs 'poor' strokes look like in sensor space? The 12.9% impact location error is reported on normalized coordinates, but without knowing the physical racket dimensions, this is hard to interpret. The comparison to racket-mounted sensors (SmartDampener) is fair but the comparison to Silent Impact is inappropriate due to different anatomical placement.
The paper demonstrates good open science practices: the dataset (848 strokes with labels) is publicly available at https://github.com/taizhouchen/BadminSense_Dataset. However, critical implementation details are missing: exact SVM hyperparameters for classification, the specific SVR kernel and $C$ values for quality rating, and threshold values ($21$ for peak detection is mentioned but not justified). The 'heartbeat-based network synchronization' for temporal alignment is described but not validated. The system requires a Windows server with Python/Scikit-learn - no docker/containerization is provided. Perhaps most critically, the real-time latency is reported as 'approximately one minute' for post-session analysis, but no per-stroke processing latency is given, making real-time feasibility assessment impossible. Left-handed player support is explicitly excluded, limiting generalizability.
Evaluating badminton performance often requires expert coaching, which is rarely accessible for amateur players. We present adminSense, a smartwatch-based system for fine-grained badminton performance analysis using wearable sensing. Through interviews with experienced badminton players, we identified four system design requirements with three implementation insights that guide the development of BadminSense. We then collected a badminton strokes dataset on 12 experienced badminton amateurs and annotated it with fine-grained labels, including stroke type, expert-assessed stroke rating, and shuttle impact location. Built on this dataset, BadminSense segments and classifies strokes, predicts stroke quality, and estimates shuttle impact location using vibration signal from an off-the-shelf smartwatch. Our evaluations show that
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.