Sign in

Visual surveillance, Crowd behavior, Tracking, Automatic analysis, Covid-19.

The paper presents a number of recently developed methods for automatic categorization of socio-cognitive crowd behavior from video surveillance data. Sadly and unexpectedly, the recent pandemic outbreak of Covid-19 created a new important niche for such tasks, which otherwise have been rather associated with police monitoring public events or oppressive regimes tightly controlling selected communities. First, we argue that a recently proposed (see [31]) general socio-cognitive categorization of crowd behavior well corresponds to the needs of social distancing monitoring. It is explained how each of four proposed categories represents a different level of social (ir)responsibility in public spaces. Then, several techniques are presented which can be used to perform in real time such a categorization, based only the raw-data inputs (i.e. video-sequences from surveillance cameras). In particular, we discuss: (a) selected detection and tracking aspects for individual people and their groups, (b) practicality of data association combining results of detection and tracking, and (c) mid-level features proposed for neural-network-based classifiers of the behavior categories. Some illustrative results obtained in the developed feasibility studies are also included.