Emergent Multi-View Fidelity in Autonomous UAV Swarm Sport Injury Detection

TL;DR

This paper demonstrates that UAV strategies aimed at capturing all collision incidents in sports naturally lead to multi-view coverage, enhancing the accuracy and redundancy of collision detection without needing drone communication.

Contribution

It reveals that UAV strategies designed for incident coverage inherently produce multi-view fidelity, improving collision detection robustness in sports monitoring.

Findings

Most collisions are captured by multiple drones

Multi-view coverage improves detection accuracy

No drone-to-drone communication is needed

Abstract

Accurate, real-time collision detection is essential for ensuring player safety and effective refereeing in high-contact sports such as rugby, particularly given the severe risks associated with traumatic brain injuries (TBI). Traditional collision-monitoring methods employing fixed cameras or wearable sensors face limitations in visibility, coverage, and responsiveness. Previously, we introduced a framework using unmanned aerial vehicles (UAVs) for monitoring and real time kinematics extraction from videos of collision events. In this paper, we show that the strategies operating on the objective of ensuring at least one UAV captures every incident on the pitch have an emergent property of fulfilling a stronger key condition for successful kinematics extraction. Namely, they ensure that almost all collisions are captured by multiple drones, establishing multi-view fidelity and redundancy, while not requiring any drone-to-drone communication.