Summarizing the performances of a background subtraction algorithm measured on several videos

  • 2020-02-13 17:35:34
  • Sébastien Piérard, Marc Van Droogenbroeck
  • 10

Abstract

There exist many background subtraction algorithms to detect motion invideos. To help comparing them, datasets with ground-truth data such as CDNETor LASIESTA have been proposed. These datasets organize videos in categoriesthat represent typical challenges for background subtraction. The evaluationprocedure promoted by their authors consists in measuring performanceindicators for each video separately and to average them hierarchically, withina category first, then between categories, a procedure which we name"summarization". While the summarization by averaging performance indicators isa valuable effort to standardize the evaluation procedure, it has notheoretical justification and it breaks the intrinsic relationships betweensummarized indicators. This leads to interpretation inconsistencies. In thispaper, we present a theoretical approach to summarize the performances formultiple videos that preserves the relationships between performanceindicators. In addition, we give formulas and an algorithm to calculatesummarized performances. Finally, we showcase our observations on CDNET 2014.

 

Quick Read (beta)

loading the full paper ...