BlurBall: Joint Ball and Motion Blur Estimation for Table Tennis Ball Tracking

TL;DR

This paper introduces a new labeling strategy that places the ball at the center of the blur streak and explicitly annotates blur attributes and introduces BlurBall, a model that jointly estimates ball position and motion blur attributes.

Abstract

Motion blur reduces the clarity of fast-moving objects, posing challenges for detection systems, especially in racket sports, where balls often appear as streaks rather than distinct points. Existing labeling conventions mark the ball at the leading edge of the blur, introducing asymmetry and ignoring valuable motion cues correlated with velocity. This paper introduces a new labeling strategy that places the ball at the center of the blur streak and explicitly annotates blur attributes. Using this convention, we release a new table tennis ball detection dataset. We demonstrate that this labeling approach consistently enhances detection performance across various models. Furthermore, we introduce BlurBall, a model that jointly estimates ball position and motion blur attributes. By incorporating attention mechanisms such as Squeeze-and-Excitation over multi-frame inputs, we achieve state-of-the-art results in ball detection. Leveraging blur not only improves detection accuracy but also enables more reliable trajectory prediction, benefiting real-time sports analytics.

Figures (15)

• Figure 1: Motion blur frequently appears in broadcast footage but is typically disregarded. Yet, it offers valuable cues for estimating the ball’s velocity. The blue cross denotes the classical labeling approach, which introduces asymmetry and ambiguity to the detection task. We propose a refined annotation strategy: relabel the ball center to correspond to the middle of the blur (red cross) and include a directional blur label (green line) to capture motion information better.
• Figure 2: Example scenes from our dataset, showcasing a diverse range of contexts to ensure comprehensive coverage.
• Figure 3: Motion blur labeling schematic. Conventional labels mark the front blur edge $\mathbf{p}_1$, which may be confused with $\mathbf{p}_2$ without motion context. We propose labeling the blur center $\mathbf{p}_b$ with its half-length $l$ and orientation $\theta$.
• Figure 4: Distribution of the half-lengths $l$ of the motion blur streak in our dataset. The maximum recorded value was 73 pixels.
• Figure 5: Left: Label and target heatmap. Right: Model prediction overlay with estimated ball center (red) and blur (green).