Automated video analysis for social science research

Handbook of computational social science, vol. 2: Data science, statistical modeling, and machine learning methods
,
pp. 386-398
,
New York
,
Routledge
,
2022,

Uwe Engel, Anabel Quan-Haase, Sunny Liu,
Nyhuis, Dominic, Tobias Ringwald, Oliver Rittmann, Thomas Gschwend, Rainer Stiefelhagen, Lars E. Lyberg
ISBN: 978-1-032-07770-3 (print), 978-1-003-02524-5 (online)
DOI: 
10.4324/9781003025245
Abstract

Widespread digitization is a disruptive force for the social sciences by generating vast data on human behavior. While social scientists have made increasing use of the opportunities that result from the widespread availability of digitized text, scholars have only begun to tap the more general potentials of the universal digitization of data, specifically by analyzing digitized images, as well as audio and video data. To highlight some of these potentials, the chapter discusses the state of the art in automated video analyses. First, we provide an overview of technical solutions for common classification scenarios in video data, such as object detection, pose detection, activity recognition, and person identification. Next, we discuss some current studies that use audio and video data to study social science phenomena. In the third part of the chapter, we discuss a sample application on video footage from a German state-level parliament in greater detail. The chapter closes with a discussion of potentials and challenges for automated video analysis in social science research.