# LES AUTEURS : Marcos Zúñiga, François Brémond, Monique Thonnat

## RESUME:

We propose a new object classification approach for monocular video sequences, which allows to classify objects modelled in a way that is independent from the relative position between the object and the camera, and object orientation, considering a pin-hole camera model. To achieve this independence, a new generic object model which represent object as parallelepiped has been proposed, giving good estimates of its dimensions and proposing visual reliability measures for these dimensions. These measures give a representation of the visibility of the estimated
dimension and have been principally proposed to aid posterior phases of the video understanding process, as dimensional estimation on tracking or confidence degree of dimensions for primitive state detection. The method obtains the 3D model predictions using a set of 2D moving regions (obtained in a segmentation phase), the perspective matrix transform (obtained from camera calibration) and predefined 3D models of expected objects on the scene. After classification, a merging step is performed to improve the classification performance by assembling 2D moving regions with better 3D model probability when together. This approach has shown promising results
on object classification, obtaining very high detection rates for very complicated situation and accomplishing with real-time constraint.

Mots clé: video interpretation, object classification, 3D object model, reliability measure

## BibTeX reference:



@InProceedings{zuniga2006,
author = {M. Z\'u$\tilde{\mathrm n}$iga and F. Br\'emond and M.Thonnat},
title = {Fast and Reliable Object Classification in Video Based on a 3D
Generic Model},
booktitle = {Proceedings of the International Conference on Visual
Information Engineering (VIE2006)},
address = {Bangalore, India},
year = {2006},
month = {26-28 September},
pages = {433-440},
}



Dernière mise à jour : 3/10/06
Catherine.Martin@sophia.inria.fr