王依婷
The demand for high-quality sports event filming has grown significantly with advancements in streaming technologies and the widespread use of multimedia platforms. However, capturing dynamic team sports like basketball remains a challenge for static camera setups. While traditional multi-camera systems operated manually by multiple camera operators are effective, they are costly, labor-intensive, and require skilled operators. Commercial automated systems offer alternatives, but they often depend on specialized hardware and struggle in complex environments, such as areas with moving spectators or when multiple events are held simultaneously. This paper presents a real-time automated filming system for basketball, utilizing a single PTZ (Pan-Tilt-Zoom) camera to track player movements and dynamically adjust camera angles. We explore both position-based and motion-based methods, discussing their strengths and weaknesses and the contexts in which they are most suitable. The proposed system offers a cost-effective, flexible solution that can adapt to the dynamic nature of basketball games and the complexities of various environments, while relaxing the constraints of fixed camera placements. To ensure fair comparisons between tracking methods, we developed a simulation environment based on pre-recorded 360° basketball footage. This environment enables us to capture human-controlled camera angles as ground truth data, providing a consistent basis for testing. Additionally, we constructed a dataset of annotated basketball videos with camera angles, addressing a gap in publicly available resources for research in automated camera control. Comprehensive evaluations, including objective metrics and user studies, demonstrate the system’s ability to capture footage that aligns with human filming behavior. This work not only presents a practical solution for automated sports filming but also lays the groundwork for future research in dynamic camera control.
MIS Lab