A video information driven football recommendation system

Abstract

Designing a football recommendation system requires collecting physical, technical and tactical information from football games. However, traditional technical and tactical statistics of football still depend on manual numbering, which is a huge labor consumption. Though GPS (Global Positioning System) devices could be applied to collect football data, they are very expensive and are forbidden in many football games. To solve these problems, we utilize video tracking to capture physical and tactical information of football players and propose a football recommendation system through combining players’ tracking techniques with recommendation algorithms. In our proposed system, the YOLOv2 (You Only Look Once version 2) algorithm and improved KCF (Kernelized Correlation Filter) method are applied to obtain and analyze the location information of football players. The proposed system could automatically recognize and track players according to match videos instead of using wearable GPS devices. Compared with GPS, the experimental results have shown that the data obtained from our system are much closer to reality and have lower standard deviation.