A Real-Time, Vision-Based System for Badminton Smash Speed Estimation on Mobile Devices

Diwen Huang · 2025

Read A Real-Time, Vision-Based System for Badminton Smash Speed Estimation on Mobile Devices by Diwen Huang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-05334.txt.

Introduction

Measuring a badminton smash with an ordinary phone camera is the goal of Diwen Huang in a solo 2025 sports technology paper. Shot speed and angle give players useful feedback, but the gear that captures them has been costly and complex, shutting out amateurs.

Huang trains a compact object detector to find the shuttlecock in video and pairs it with a Kalman filter, a tracker that smooths noisy positions over time. A video-based method with space and time scaling then converts the tracked path into speed automatically. Everything is packaged in a simple mobile app. Affordable analysis of this kind could help recreational players train with the feedback once reserved for professionals.

Catalog record

CREATORDiwen Huang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE17 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-05334

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Real-Time, Vision-Based System for Badminton Smash Speed Estimation on Mobile Devices Diwen Huang Abstract Performance metrics in sports, such as shot speed and angle, provide crucial feedback for athlete development. However, the technology to capture these metrics has historically been expensive, complex, and largely inaccessible to amateur and recreational players. This paper addresses this gap in the context of badminton, one of the world’s most popular sports, by introducing a novel, cost-effective, and user-friendly system for measuring smash speed using ubiquitous smartphone technology. Our approach leverages a custom-trained YOLOv5 model for shuttlecock detection, combined with a Kalman filter for robust trajectory tracking. By implementing a video-based kinematic speed estimation method with spatiotemporal scaling, the system automatically calculates the shuttlecock’s velocity from a standard video recording. The entire process is packaged into an intuitive mobile application, democratizing access to high-level performance analytics and empowering players at all levels to analyze and improve their game. 1 Introduction The proliferation of data analytics has revolutionized modern sports, offering athletes and coaches unprecedented insights into performance. Key metrics such as shot speed have become fundamental for training, strategy, and talent identification. Despite these advancements, access to the requisite technology remains a significant barrier for those outside professional circuits, primarily due to prohibitive costs and operational complexity. While several efforts have sought to democratize sports technology, these innovations have not been adequately extended to badminton. Badminton is a global sport, with participation levels that rank it among the top two most popular sports worldwide [ 1 ] . Nevertheless, consumer-grade products for performance analysis are conspicuously scarce. The primary available tool, the radar gun, is not only exorbitant but also cumbersome to configure and operate. Furthermore, standard radar guns often struggle to consistently detect the badminton shuttlecock due to its small size and low radar cross-section, which impedes functionality [ 2 ] . This paper presents a novel solution that circumvents these barriers. We have developed a system that is free, user-friendly, and universally accessible to anyone with a modern smartphone. By harnessing the power of computer vision and sophisticated tracking algorithms, our application provides a reliable method for measuring the pinnacle metric of badminton performance: the smash speed. 2 Background and Related Work Arguably the most celebrated and crucial metric in badminton is the speed of the smash. Existing solutions for its measurement vary drastically in accessibility and accuracy. 2.1 Professional Tournament Technology In elite international tournaments, systems like Hawk-Eye are employed to provide instantaneous and highly accurate data. This technology utilizes multiple high-speed cameras positioned around the court to triangulate the shuttlecock’s position in 3D space. While representing the gold standard in accuracy, this solution is confined to professional venues and is entirely unachievable for the vast majority of players. 2.2 Radar Guns The radar gun is the most common alternative. It measures speed by emitting radio waves and analyzing the Doppler shift in the reflected waves from a moving object. However, high-precision sports radar guns cost thousands of dollars and require careful positioning to yield accurate results. Cheaper consumer models often fail to reliably detect the shuttlecock, as its non-metallic feather or nylon construction provides a poor reflective surface for radar waves. 2.3 Video-Based Analysis A more accessible but laborious method involves recording a video and manually tracking the shuttlecock frame by frame using software. After identifying the shuttlecock’s pixel coordinates in consecutive frames, one can apply kinematic equations and a known reference distance to calculate its real-world speed. This process is exceedingly technical, time-consuming, and prone to human error, making it impractical for regular use. Beyond 2D speed measurement, other research has focused on the more complex challenge of full 3D trajectory reconstruction from a single camera view, using geometric constraints and physics-based models to estimate the shuttlecock’s flight path in three-dimensional space [ 3 , 4 ] . Our work, however, prioritizes accessibility and immediate feedback by focusing on the peak 2D velocity, which is the most critical metric for player training. Our proposed solution directly addresses the shortcomings of these methods. It does not require professional status, thousands of dollars in equipment, or hours of tedious manual analysis. Instead, it offers an automated, on-demand tool for any player with a phone. Figure 1: The end-to-end system pipeline. The user uploads a video, which is processed frame-by-frame by the YOLOv5 model. Detections are filtered and smoothed by the Kalman filter, and the resulting trajectory is used to calculate the peak smash speed. 3 Methodology Our system provides an end-to-end pipeline for automated speed calculation using video-based kinematic speed estimation (see Figure 1 ). The core of our method involves spatiotemporal scaling and robust shuttlecock tracking via a custom-trained computer vision model, further refined with a state-estimation filter. 3.1 Data Collection and Model Training Our methodology relies on a strict two-dimensional analysis, which requires the camera to be positioned perpendicular to the shuttlecock’s primary plane of motion, as shown in Figure 2 (a). This orthogonal viewpoint effectively eliminates the z-axis (depth) from calculations, simplifying the problem to x and y coordinates. Consequently, established public datasets for ball tracking, such as TrackNetV2 [ 5 ] , which are typically captured from broadcast angles (Figure 2 (b)), were unsuitable for our specific geometric constraints. (a) Correct Perpendicular View (b) Unsuitable Broadcast View Figure 2: Comparison of camera views. (a) The required perpendicular view for our 2D analysis. (b) A typical broadcast view, unsuitable due to perspective distortion. To address this, we first curated and annotated a custom dataset of 15,000 images, all captured from the required perpendicular perspective. This ensures the model is trained on data that mirrors the intended use case. We then used this dataset to train a YOLOv5 object detection model [ 6 ] . YOLOv5 was chosen for its excellent balance of speed and accuracy in real-world applications and its proven efficacy when trained on relatively small, custom datasets. You Only Look Once (YOLO) is a family of single-pass object detection models. Unlike two-stage detectors, YOLO treats object detection as a single regression problem, directly from image pixels to bounding box coordinates and class probabilities [ 7 ] . Upon completion of training, the model achieved a precision of 93% , a recall of 87% , and a mean Average Precision at an IoU threshold of 0.5 ([email protected]) of 91% , demonstrating a high of accuracy in identifying the shuttlecock. 3.2 Automated Speed Calculation Program The program, developed in Python, orchestrates the entire analysis process. 3.2.1 Spatiotemporal Scaling The user initiates the process by uploading a video and establishing a real-world scale factor. To correct for perspective distortion, the reference line for scaling must align with the player’s depth on the court. The user selects two points on the court line (e.g., the rear doubles service line or the front service line) that is parallel to the camera’s viewing plane and at the same distance from the camera as the player’s point of impact. This ensures the shuttlecock and the reference line are in the same plane of motion, making the scale factor consistent. With a known real-world distance, d_{real} , the scale factor S_{f} is computed from the pixel distance d_{pixel} : S_{f}=\frac{d_{real}}{d_{pixel}}\quad[\text{meters/pixel}] 3.2.2 Shuttlecock Detection and

Read the complete text