Testing automated video analysis to capture discrimination in Germany

The project develops and tests new methods for observing social interaction in public spaces. Its aim is to make subtle forms of discrimination measurable—forms that are expressed not through words, but through body movements, spatial distance, or gaze behavior.

To this end, automated 3D video analysis techniques originally developed in movement research and robotics are employed. In an experimental field design in Berlin, passersby were filmed as they walked past two male actors—once with visible Muslim markers (galabija) and once without. AI-based image analysis was used to collect precise movement data, such as the distance between individuals, walking speed, and upper-body rotation.

The project investigates whether discrimination is also expressed through physical avoidance—for example, when people unconsciously keep greater distance from certain groups. The findings show that, on average, passersby maintained a greater distance from individuals with Muslim markers, although the strength of this effect varied depending on the context.

Beyond these substantive findings, the project makes an important methodological contribution. It demonstrates that automated 3D video analysis can be a powerful tool for social research when combined with clear ethical standards, anonymization techniques, and transparency mechanisms. In this way, 3D Social Research opens up new perspectives for studying discrimination, social interaction, and social cohesion.

Previous studies on discrimination in public spaces have largely relied on observations, surveys, or experiments with limited objectivity and scope. Subtle forms of nonverbal exclusion—such as physical distancing or avoidance—have been difficult to capture reliably. The 3DSR project addresses this gap by using automated 3D video analysis to precisely measure behavioral patterns in real urban environments. This approach makes discrimination accessible not only as a subjective experience but also as an observable and quantifiable phenomenon. At the same time, the project expands the methodological toolkit of social science by incorporating AI-based approaches that provide new insights into social dynamics.

The project pursues two main objectives:
First, to demonstrate that automated 3D video analysis is an innovative and ethically responsible tool for measuring social inequality in public spaces.
Second, to empirically assess whether—and to what extent—subtle discrimination against Muslims is reflected in physical distance and nonverbal behavior.

In addition, the project contributes to the development of new methodological standards for the use of AI-based observation in social science, with the goal of enabling more precise, transparent, and context-sensitive measurement of societal discrimination in the future.

The study is based on a field experiment conducted in Berlin, in which passersby were recorded using a 3D camera while walking past two male confederates. Three experimental conditions varied the confederates’ appearance:

Two white men without Muslim markers
Two Muslim men with visible religious markers (galabija)
The same Muslim men without these markers

The 3D video data were analyzed using an AI-based motion analysis (OpenPose algorithm, Cao et al. 2021). For each person and each frame, the positions of 21 body points were recorded, enabling precise measurements of interpersonal distance, walking speed, and upper-body rotation. Block randomization and standardized camera positioning ensured that differences in passersby behavior could be attributed to the experimental conditions.

This design combines innovative methodology with classical experimental social research, creating a basis for objectively and reproducibly analyzing physical expressions of discrimination.

The results show that automated 3D video analysis can be reliably used to measure subtle forms of discrimination in public spaces. In the Berlin field experiment, passersby on average maintained greater physical distance from confederates with visible Muslim markers than from white, non-Muslim comparison individuals. This effect was particularly evident in busy street sections, while it was weaker or even reversed in quieter residential areas.

In addition to replicating findings from U.S.-based research (Dietrich & Sands 2023), the study makes a methodological contribution: it demonstrates that AI-based 3D analysis can precisely and anonymously capture nonverbal micro-reactions—such as distance, movement speed, and body orientation. This makes discrimination observable as embodied spatial behavior and allows it to be empirically and quantitatively assessed.

• https://www.econstor.eu/bitstream/10419/323244/1/I4R-DP250.pdf 
•  www.nature.com/articles/s41562-023-01589-7eeing