ACTIVITY SHAPES : Analysis methods of video-recorded human activity in a co-visible space

The aim of this research is to develop two methods to help us understand the fundamental distinctionsamong human activities in terms of spatial occupancy. To characterize the features of the distribution ofhuman activities in a space (and over time), we introduce the concept of “activity shapes.” To obtain adistinctive analysis of activity shapes, we ran an experiment in which a group of six adults shared a fully covisiblespace and sequentially performed three specific activities characterized as eccentric, concentric, ordistributed. We video recorded the three scenarios using overhead cameras that allowed us to closely mapparticipants’ positions on the floor layout, obtaining the data in two formats: 1) a sequence of images fromthe overhead videos, automatically stored and pre‐computed to extract and aggregate motion; and 2) adataset of individuals’ identification and positions over time, manually annotated after repeatedobservations of the videos. Using the images sequence, we qualitatively analyzed the activity shapes usingViz‐A‐Vis, a tool for visualizing of activity through computer vision (Romero et al., 2008; 2011). Using thedataset, we performed two analyses: 1) the geometry and the topology of the activity shapes; and 2) theirspatiotemporal configurations, introducing the use of statistical analysis of space occupancy patterns. Whileit is not possible to generalize to all activity conditions from these three samples, we discovered sometendencies in the activity shapes. Our findings revealed several main distinctions in terms of geometry,topology, dispersion, gravitation, and clustering; supporting the development of the methods presented inthis work and directions of future implementation of these analyses in more complex spaces and scenariosthat complement space syntax analysis.QC 20160407