This image processing algorithm is part of the PERCEPTION Project. The goal of this project is to study automated scene recognition problem for missions such as situation assessment, surveillance or intelligence, for systems moving and performing autonomous actions (e.g. mobile robots) or for decision-aid systems. As far as dynamic environments are concerned, object detection, classification and recognition are needed. A further step consists in finding what is going on in this environment, that is to be able to say what the objets are doing and what their purposes are.

In our case, image sequences from the VIGILE II data campaign, many different sensors (infrared and panchromatic cameras) are looking at the environment (a parking area). Whatever the spectral mode, infrared or panchromatic, the numerical process has to detect moving objects in corresponding images sequences. The moving object detection task is based on a spatio-temporal edge detector. As a matter of fact, the information provided by three consecutive images is sufficient to tell the static edges of the scene from the real moving edges. Moving edges location is done by comparing gradient of the current image to gradients of previous and next images.

The ultimate step of numerical processing task is to recognize the detected objects. Decentralized data fusion classification methods are used to achieve good performances. Tracking of moving objects is also performed by the system.

And This is Important Too :

The data collecting campaign VIGILE II was done for the purpose of the Perception Project. VIGILE II database is built upon several scenarii. Each one is based on realistic situations to study main difficulties of the perception fonction (e.g. sensor saturation, object occlusion or disappearing, concerted action, activities conflict...).
In this frame, a generic simulation software platform is performed at this time. It has been developed exclusively upon VIGILE II scenarii.
Data synchronization was performed during recording of image sequences. The available sensors are :
panchromatic CCD cameras, wide or restricted fields of view,
color CCD cameras, wide or restricted fields of view,
infrared cameras, 3-5 and 8-12 micrometers, wide or restricted fields of view,
panoramic radar, 94 GHz wavelength.
Until today, only panchromatic and infrared data are available (not radar data).