Proposition d’une approche intelligente pour la reconnaissance d’actions humaines à partir d’image de vidéosurveillance

Abstract

In this memory, we propose a method of activity recognition, based on a hybrid model of Hidden Markov-Chain Separator (SVM-HMMs) which explicitly models the sequential aspect of the activities while exploiting geometric parameters, that we have proposed, the game of the experiments was playing on a video database KTH. We have shown that the use of this hybridization makes it possible to improve the performance of the recognition system. Keywords: Recognition of activities; Video surveillance, sequential data classification; Large margin separators; Hidden markov Model, video, recognition.