Answers are grounded in this paper and cite their sources.
A real-world dataset, a with a total of 132 cricket deliveries were recorded. These are captured by three cameras and a radar, where one of the cameras has the radar positioned below it. The sensors are calibrated, and this calibration is also used in the simulated dataset of 200 deliveries. The real-world dataset had a bowling machine bowl the balls for consistency. The ball used was a plastic hockey ball wrapped in tin foil so that the radar could detect it. Sticky tape was then wrapped over the tin foil so that the bowling machine would be able to bowl the ball without ripping off the tin foil. The bowling machine was set to bowl the ball at 32 m/s.The data is the post-processed data, where the ball is detected in the videos and radar heat maps. The dataset has two folders, one for the real-world and one for the simulated dataset. In the real-world dataset there are the final calibration of the cameras and radar, the radar config, and the detected ball for the sensors for each of the videos. The simulated dataset has the simulated ball state space, the camera and radar projections, and the detected ball in the simulated camera and radar projections. This can also be used to optimise the placement of the sensors on/around the pitch.
No takes yet. Share an insight, caveat, or question.
Dylan Kamstra (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: