Web site: http://www.prometheus-fp7.eu
Prometheus: Prediction and interpretation of human behavior based on probabilistic structures and heterogeneous sensors.
We officially started the Prometeus project in January of this year. The kickoff meting took place in Chamrousse France were I could confirm that the project promises a wide and enthusiast scientific cooperation.
"Prometeus proposes to research probabilistic inference algorithms within the paradigm of recursive Bayesian estimation to the problem of online tracking of multiple people in a scene, and the identification of the interaction amongst them and with the environment. The core research components of the proposal are in the representation of uncertainties arising from multiple modalities including laser, visual, acoustical and infrared, the fusion of information gathered from such diverse range of sensors into a coherent mechanism that makes predictions about interactions and the coupling between modeling high level behaviour of people in a scene with signal processing issues of sensor fusion and tracking. Bayesian inference will provide the core architectural framework to carry out the above research via a rigorous mathematical framework. Sequential estimation algorithms around the notion of particle filtering and approximations that are required to handle estimation variance issues will be explored. The overall task is based on a hierarchy which at low levels consists of processing of robust detection of humans, and the expansion of this into abstractions as 2D shapes and 3D wholes of articulated limbs. The subsequent levels are responsible for the localization and tracking of humans on an extended time basis. Higher level processing includes making predictions about intended actions of humans in the scene via a combination of data-driven (i.e., classifications learnt from an annotated database), continuous time (vector autoregressive processes) and utility-driven models. The ultimate goal of the project will be to demonstrate a framework that allows the interpretation of human behaviour based on integration of sensor information and a priori statistical models of human appearance, motion and behaviour.
Specifically, the PROMETHEUS project will:
deploy a dispersed network of perceptual modalities that will allow an integration of sensing views andefficient analysis of human-motion analysis, including situations with several people in the scene
set-up a synergetic network of heterogeneous modalities that provide complementary perceptual information
implement a probabilistic framework for the fusion of heterogeneous information to allow the detection,localization and tracking of humans
train advanced probabilistic networks to recognize human bahavioural and interaction patterns based on sensor-logs of the heterogeneous sensors
provide short-time prediction of human goals based on the trained models of behaviour
set up a prototype that will be tested on two scenarios, where one consists of a network of confined spaces with focus on the analysis of individual behaviour, and one of a large space with focus on overall directions and densities of crowd motion."
RSS 2.0