OBJECT MODELLING, DETECTION AND LOCALISATION IN MOBILE VIDEO: A STATE-OF-THE-ART - Download this document for free, or read online. Document in PDF available to download.

1 ENSTA ParisTech U2IS-RV - Robotique et Vision

Abstract : This report is part of the state-of-the-art deliverable of the ITEA2 project SPY -Surveillance imProved System: Intelligent situation awareness-whose purpose is to develop new urban surveillance systems using video cameras embedded within mobile security vehicles. This report is dedicated to the problem of finding objects of interest in a video. -Object- is understood in its familiar i.e. semantic sense: e.g. car, tree, human, road

. and the system is supposed to automatically find the location of such objects in the captured video. To be consistent with the project technological level, we shall exclude the -developmental- approaches, where the system does not know the objects in advance, and constructsincrementally its own internal representation. We then suppose that the system operates with a provided representation of the objects and its environment that has been constructed learned off-line, and that may evolve on-line. Such representation includes a set of object classes that the system is then expected to recognize and localise in every image, either by attributing accordingly a label to every location in the image task referred to as -semantic segmentation-, or by localising –more or less precisely– instances of each class in the video and tagging every image accordingly referred to as -semantic indexing-. We thus present a state-of-the-art of the video analysis methods for object and environment modelling and semantic indexing or segmentation with respect to the corresponding model.

Keywords : object detection object modelling

Author: Antoine Manzanera -

Source: https://hal.archives-ouvertes.fr/


Related documents