Analysis and pattern detection on large amounts of annotated sport motion data using standard SQL

Abstract

This paper proposes a inexpensive and flexible way to analyze large amounts of sport motion data, which are generated by automated motion tracking methods and complemented with manual annotations. A database, obtained by tracking and annotating over 100 squash plays was used. The goal was to find a way to automatically detect certain kinds of play, activities, certain predefined scenarios and to generate various statistics about these activities - without hard-coding them in the executable code. We found that SQL-enabled databases provide a flexible and scalable solution to this problem. The examples of actual SQL queries for sport analysis are presented in the paper along with short tutorial on particular aspects of SQL language, which were exploited in our solution.