Abstract
The golf swing is a complex movement requiring considerable full-body coordination to execute proficiently. As such, it is the subject of frequent scrutiny and extensive biomechanical analyses. In this paper, we introduce the notion of golf swing sequencing for detecting key events in the golf swing and facilitating golf swing analysis. To enable consistent evaluation of golf swing sequencing performance, we also introduce the benchmark database GolfDB, consisting of 1400 high-quality golf swing videos, each labeled with event frames, bounding box, player name and sex, club type, and view type. Furthermore, to act as a reference baseline for evaluating golf swing sequencing performance on GolfDB, we propose a lightweight deep neural network called SwingNet, which possesses a hybrid deep convolutional and recurrent neural network architecture. SwingNet correctly detects eight golf swing events at an average rate of 76.1%, and six out of eight events at a rate of 91.8%. In line with the proposed baseline SwingNet, we advocate the use of computationally efficient models in future research to promote in-the-field analysis via deployment on readily-available mobile devices.
Introduction
Analyzing a golf swing is a staple of the sport, but it has historically been a choice between two extremes: expensive, marker-based motion capture systems used in labs, or manual video analysis by a professional instructor. While computer vision has mastered tracking the flight of the ball, understanding the mechanics of the human body during the high-speed swing is significantly more difficult. To automate this, computers must identify exactly when specific events occur—such as the "Top" of the backswing or the moment of "Impact."
The GolfDB Dataset
To address the lack of standardized data for this task, we introduce GolfDB, the first high-quality video database dedicated to golf swing sequencing.
- Scale and Diversity: The dataset contains 1,400 HD videos (over 390,000 frames) sourced from YouTube, featuring 248 professional golfers. It covers both men and women using various clubs (drivers, irons, wedges) from both "face-on" and "down-the-line" perspectives.
- Annotations: Each video is manually annotated with 10 frames: the start/end of the clip and eight key swing events (Address, Toe-up, Mid-backswing, Top, Mid-downswing, Impact, Mid-follow-through, and Finish). It also includes bounding boxes and metadata like player name and club type.
Methodology: SwingNet
We propose SwingNet, an end-to-end deep learning architecture designed for computational efficiency and potential deployment on mobile devices.
SwingNet utilizes MobileNetV2 as a spatial feature extractor. Because static frames can be ambiguous (e.g., "Mid-backswing" and "Mid-downswing" look remarkably similar in a single image), the model incorporates temporal context via a Bidirectional LSTM (Long Short-Term Memory) network. This allows the model to analyze the "flow" of time to identify events accurately. Performance is evaluated using the Percentage of Correct Events (PCE) metric.
Results and Discussion
SwingNet achieves an average PCE of 76.1% across all eight events. When excluding the more subjective "Address" and "Finish" events, accuracy increases to 91.8%. Key findings include:
- Temporal Context is Essential: The Bi-LSTM is critical for resolving visual ambiguities between different swing phases.
- Transfer Learning: The model required pre-trained ImageNet weights to converge, highlighting the importance of general visual features in domain-specific tasks.
- Efficiency: On a standard desktop CPU, the model processes a frame in 10-13ms. This speed suggests that real-time analysis on modern smartphones is entirely feasible with further optimization.
Conclusion
GolfDB and SwingNet provide a foundation for automated golf biomechanics. Future work will involve combining sequencing with human pose estimation and club tracking. By identifying key frames like the "Top" of the swing, future applications could automatically measure the "X-factor" (hip-to-shoulder rotation difference), providing elite-level data to golfers on the practice range via mobile devices.
Summary updated