Core Concepts
This work presents a real-time and accurate method for estimating the spin of table tennis balls using an event camera.
Abstract
The authors propose a method for estimating the spin of table tennis balls using an event camera. The key contributions are:
Tracking the ball in real-time and extracting events generated by the ball's logo.
A real-time and accurate method for estimating the table tennis ball's spin magnitude and axis.
Evaluation using a ball spinner to show the potential of the approach and a deployment with a ball thrower to demonstrate the method in a real setup.
The method works by first tracking the ball using the Exponential Reduced Ordinal Surface (EROS) event representation. This allows for continuous and asynchronous updates from the event stream. The ball's position, velocity, and radius are estimated using a Kalman filter.
Next, the events generated by the ball's logo are extracted based on the estimated ball properties. Optical flow is then computed on these extracted logo events to infer the ball's spin magnitude and axis.
The authors evaluate their method in two settings: a ball spinner and a ball thrower. With the ball spinner, they achieve a spin magnitude mean error of 10.7 ± 17.3 rps and a spin axis mean error of 32.9 ± 38.2°. When deployed with the ball thrower, their method achieves a success rate slightly above a state-of-the-art frame-based approach, with comparable spin magnitude and axis estimation errors.
The authors also discuss the limitations of their approach, noting that the method struggles with certain logo designs and orientations. They suggest potential solutions, such as using a higher resolution event camera or a longer focal length lens.
Stats
The ball spinner experiments reported a spin magnitude mean error of 10.7 ± 17.3 rps and a spin axis mean error of 32.9 ± 38.2°.
The ball thrower experiments reported a success rate slightly above a state-of-the-art frame-based approach, with a spin magnitude mean error of 10.7 ± 17.3 rps and a spin axis mean error of 32.9 ± 38.2°.
Quotes
"Spin plays a pivotal role in ball-based sports. Estimating spin becomes a key skill due to its impact on the ball's trajectory and bouncing behavior."
"Event cameras do not suffer as much from motion blur, thanks to their high temporal resolution. Moreover, the sparse nature of the event stream solves communication bandwidth limitations many frame cameras face."