Scalable Sketch-based Sport Video Retrieval in the Cloud

Authors
Ihab Al Kabary, Heiko Schuldt
Type
In Proceedings
Date
2020/8
Appears in
Proceedings of the International Conference on Cloud Computing (CLOUD'20)
Location
Honolulu, HI, USA
Publisher
Springer
Abstract

Content-based video retrieval in general and in sport videos in particular has attracted an increasing interest in the past few years, due to the growing interest in sports analytics. Especially sketch-based queries, enabling spatial search in video collections, are increasingly being demanded by coaches and analysts in team sports as an essential tool for game analysis. Although there has been great progress in the last years in the field of sketch-based retrieval in sports, most approaches focus on functional aspects and only consider just a very limited number of games. The problem is to scale these systems to allow for interactive video retrieval on a large game collection, beyond single games. In this paper, we show how SportSense, our sketch-based video retrieval system, can be deployed and scaled-out in the Cloud, allowing managers and analysts to interactively search for scenes of their choice within a large collection of games. In our evaluations, we show how the system can scale to a collection of the size of an entire season with response times that enable real-time analysis.