Automatic football video production system with edge processing

Automatic video production of sports aims at producing an aesthetic broadcast of sporting events. This is an enabler of low-cost solutions for TV-like streaming of sports events. We present a new video system able to automatically produce a smooth and pleasant broadcast of football games using a camera rig composed of three fixed 4K cameras. The system automatically detects and localizes the main action of a football game and frame it, so it yields a professional cameraman-like production of the football event. We compute the actionness of a football game using a ball detector and an occupancy map representation of players based on a saliency map. The action framing is computed through geometrically pitch modeling. The whole video processing pipeline is done in the edge using smartphones and we report a 25 FPS throughput.

Automated summary

Automatic video production of sports aims to create aesthetic broadcasts of sporting events, providing low-cost solutions for TV-like streaming. A new video system is presented in this study, which uses three fixed 4K cameras to automatically produce smooth and pleasing broadcasts of football games. The system detects and localizes the main action of a football game, resulting in a professional cameraman-like production of the event.