Paranoid Shelter is a recent installation / architectural device that fabric | ch finalized later in 2011 after a 6 months residency at the EPFL-ECAL Lab in Renens (Switzerland). It was realized with the support of Pro Helvetia, the OFC, the City of Lausanne and the State of Vaud.
It was initiated and first presented as sketches back in 2008 (!), in
the context of a colloquium about surveillance at the Palais de Tokyo in
Being created in the context of a theatrical collaboration with french writer and essayist Eric Sadin around his books about contemporary surveillance (Surveillance globale and Globale paranoïa --both published back in 2009--), Paranoid Shelter
revisits the old figure/myth of the architectural shelter, articulated
by the use of surveillance technologies as building blocks.
Additionnal information on the overall project can be found through the two following links:
A compressed preview and short of the play by NOhista.
On the first
technical drawings and sketches of the Paranoid Shelter project, the
entire system was just looking like a (big) mess of wires, sensors
and video cameras, all concentrated on a pretty tiny space where humans
will have difficulties to move in. The entire space is consciously
organised around tracking methods/systems, the space being delimited
by 3 [augmented] posts which host a set of sensors, video cameras and
microphones. It includes networked [power over ethernet] video cameras,
microphones and a set of wireless ambient sensors (giving the ability
of measuring temperature, O2 and CO2 gaz concentration, current
atmospheric pressure, light, etc...).
Based on a real-time analysis of major
sensors hardware, the system is able to control DMX lights, a set of
two displays (one LCD screen and one projector) and to produce sound
through a dynamically generated text to speech process.
All programs were developed using
openFrameworks enhanced by a set of dedicated in-house C++ libraries
in order to be able to capture networked camera video flow, control
any DMX compatible piece of hardware and collect wireless Libelum sensor's
data. Sound analysis programs, LCD display program and the main
program are all connected to each other via a local network. The main
program is in charge of collecting other program's data, performing
the global analysis of the system's activity, recording system's raw
information to a database and controlling system's [re]actions
The overall system can act in an
[autonomous] way by controlling the entire installation behavior
while it can also be remotely controlled when used on stage,
in the context of a theater play.
Collecting all sensor's flows is one of
the basic task. Cameras are used to track movements, microphones
measure sound activity and sensors collect a set of ambient
parameters. Even if data capture consists in some basic network based
tasks, it is easily raised to upper complexity level when each data
collection should occur simultaneously, in real-time, [without,with]
a [limited,acceptable] delay. Major raw data analysis have to occur
directly after data acquisition in order to minimize the time-shift
in the system's space awareness. This first level of data analysis
brings out mainly frequencies information, quantity of activity and
2D location tracking (from the point of view of each camera). Every
single piece of raw information is systematically recorded in a
dedicated database : it reduces system's memory footprint (by keeping
it almost constant) without loosing any activity information. From
time to time the system can access these recorded information in its
post-analysis process, when required, mainly to add a time-scale
dimension on the global activity that occurred in the monitored
space. Time isolated information can be interpreted in a rough and
basic way, while time composition of the same information or a set of
information may bring additional meanings by verifying information
consistency over time (of course, it could be in a negative or a
positive way, by confirming or refuting a first level deduced
activity information). Another level of analysis can be reached by
taking in account the spacial distribution of sensors in the overall
installation. The system is then able to compute 3D information
getting an awareness of activities within the space it is monitoring.
It generates a second level of data analysis, spatialised, that will
increase the global understanding of captured data by the system.
Recorded activities are made available
to the [audience,visitors] through a wifi access point. Networked
cameras can be accessed in real time, giving the ability to humans to
see some of the system's [inputs]. Thus, network activity is also
monitored as another sign of human presence, the system can then
[detect] activity elsewhere than in its dedicated space.
Whatever how numerous are collected
data, the system faces a real problem when it comes to the
interpretation of these data while not having benefit of a human
brain. Events that are quite obvious to humans, do not mean anything
to computers and softwares. In order to avoid the use of some
artificial neural networks simulation (which may still be a good
option to explore), I have decided to compute a limited set of
parameters, all based on previously analysed data, only computed
lately when the system may decide to react to perceived activities.
It defines a kind of global [mood] of the system, based on which it
will [decide] whether to be aggressive (from a human point of view)
by making the global tracking activity [noticeable] by humans
evolving in the installation's space, or by focusing tracking sensors
on a given area or by trying to enhance some sensor's information
analysis, whether to settle in a kind of silent mode.
Moreover, the evolution of these
parameters are also studied in time, making the [mood] evolving in
a human way, increasing and decreasing [analogically]. System's
[mood] may be wrong or [unjustified,weird] from a human point of
view, but that's where [multi-dimensional] software becomes
interesting. Beyond a certain complexity, by adding computation
layers on top of each over, having written every single line of code
does not allow the programmer to predict precisely what next system's
[re]action will be.
We did reach here monitoring system
limitations which is obviously [interpretation,comprehension]. As long as automatic
system can not correctly [understand] data, humans will need to be in
the loop, making all these monitoring systems quite useless [as
expert system], except for producing an enormous quantity of data
that still need to be post-analysed by a human brain. As the system
is producing an important set of heteregeneous data, a set of rules
may suggest to the system some sort of data correlation. These rules
should not be too [tights,precises] in order to avoid producing
obvious system's interpretation, while keeping them slightly [out of
focus] may allow [smart,astonishing] conclusion being produced. So
there's rooms here for additional implementation of the data analysis
processes that can still completely change the way the entire
installation [can,may] behave.