American Football is one of the most popular sports in the world. As an amateur football player and a computer science student i have combined two of my interests to finish my studies.

There are a lot of great projects and papers about object tracking in sports like soccer, tennis, handball etc. I wrote a software which is based on OpenCV. OpenCV provides functions for object-tracking and for user-interfaces.

The focus at this project was how to create a low-cost system for tracking the movements of American Football Players. So i've started to analyze some recorded football games of my team. Ive played around with camera perspectives (sideline, sideline-aslant, and from the endzone) and the tracking algorithms Boosting, Multiple Instance Learning (MIL), MedianFlow and Tracking-Learning-Detection (TLD). 

Ive combined ten plays from every perspective with the tracking-algorithms above - 120 tracking-attempts at all. The best results were made with the perspective from the endzone in combination with MIL and Boosting. The algorithms TLD and Median-Flow failed at every attempt. The start-position of the players is marked as a square, the direction and the movements are displayed as dotted lines.

The result of the tracking is not perfect, but good enough to figure out what the players did at the certain play.