Cultural behaviors analysis in video sequences

In this paper, we investigate the cultural aspects of different populations from video sequences. For that, we proposed a model that considers a series of characteristics of the pedestrians and the crowd, such as distances and speeds and performs the mapping of these characteristics in personalities, emotions, and cultural aspects. The model called Big4GD consists of four dimensions of geometric characteristics and seeks to describe the behavior of pedestrians and groups in the crowd. We performed a study of group behavior in a controlled experiment and focused on differences in two attributes that vary across cultures: (walking speed and personal distance) in three countries (India, Brazil, and Germany). We use the Fundamental Diagram theory that determines the relationship between the density and speed of individuals. We use Computer Vision methods to detect and track individuals through video sequences by generating their positions and speeds as a function of time. With these data, we analyze emergent walking speeds and densities while considering the personal distance of each individual and the neighbor in front of him/her. Our results show that human behavior is more similar in highly dense populations, i.e., individuals behave like a mass when presented with limited free personal space. The opposite result is also relevant: cultural differences can be observed at low and moderate densities, and such assumptions can be applied to computational interfaces and simulations, games, and movies. Besides, we present GeoMind, a software we developed to detect a series of characteristics from pedestrians. We also performed a practical case-study using GeoMind focusing on event detection in video sequences.