Space/Crowd Interactions -Tracking people's motion in architecture

Paloma Gonzalez Rojas

Final project for the MIT class 4.566 Advanced Project in Digital Media
Development: September-December, 2013
Instructor: Prof. Takehiko Nagakura

Movie info: 480x320 pixels, mp4, 3 minutes playing at 3x speed

This project was made possible thanks to the courtesy of Nick Guillian and Prof. Joe Paradiso who generously granted me access to the recording data from their previous research project titled "Gestures Everywhere."

A 640x480 pixel version is available here at YouTube

Project Overview

The understanding of space relies on motion, one might say, as we experience space by crossing it. When people move in space, they leave an unseen trace that has not been possible to accurately record and reproduce until very recently.
In 2010, the Kinect with depth sensor was first released as a special camera for gaming console, and in 2011 Microsoft released its SDK and generated interest in research areas that use people's walking-paths and gestures. In 2012, a research group of the MIT Media Lab explored the use of Kinects by placing 25 Kinects in their building, and simultaneously recorded every path of a person walking in front of the Kinect sensors across its interior space. In September 2013, I started working with the data collected by this group, and realized exceptional values of the data, which might constitute the only complete database of anonymized people tracked inside a building readily available today.
For achitects, this is certainly an advancement because understanding space-human interaction is a fundamental part of architectural design. In nature, our traces become visible when their turning into trails, but we usually leave little visible trace in buildings until now. Jacques Derrida (1997) refers to the concept denominated “trace”, as a “mark of the absence of a presence, an always-already absent present”. By creating the visualizations of the trace data, my target is to bring this absence, the traces of moving people to presence.
This project consists of processing, visualizing and analysing the data generated in 8 months of recording, at 8 of the 13 locations with 25 cameras inside the Media Lab Building, in order to create a catalog of space-crowd interactions

A set of 8 locations were selected to study people's behavioral reactions in various architectural context. The illustration shows three examples of different spatial forms and functions.

Top: Stairs and landing
Middle: Elevator hall
Bottom: Gallery on a mezzanine floor

Fig 2. Position data

The positions of people recorded with the kinects appear as triangles since the Kinect’s sensing range is triangular in a 3.5 x 3.5 meter area. The colors differentiate sampled people. The visualizations show some emergent patterns of behavior in each architectural situation, which sometimes do not fit with our pre-conception of what people do in a specific place. This research provides important observations to account for further development.

Fig 3. Skelton data

The skeleton data retrieve the information of detailed postures of people in space. For instance, you can tell if a person looked down from a gallery floor over the handrail, or leaned back against the handrail. At a location with a bench, a person may be found sitting on it or just standing by it.

Fig 4. Visualizations

Future development

This experiment encountered several drawbacks such as insufficient KINECT range to properly analyze an architectrually meaningful spatial situation. Use of multiple Kinects or new version of Kinect with wider range would be a solution. Further development will focus on retrieving statistical data in more architectural settings and forming the catalog of human reactions to various spatial context in built environment, especially those designed with distinct architectural forms .