Deep learning Computer-Vision projects applied to real-time football games.
Football players are detected in live real-time video streams, identified, tracked across the field and their actions and intensions detected. Statistics of the results are extracted and provided to the football teams analytics departments.
Different algorithms employed, with a focus on accuracy and real-time (custom short fully conv networks, custom Mask-rcnn, time-series, etc).
Work done by Adrian Balan @ Footovision, France and presented at DS3 2018
