Probabilistic Hierarchical Framework for Clustering of Tracked Objects in Video Streams

Riad Hammoud 1 Roger Mohr 1
1 MOVI - Modeling, localization, recognition and interpretation in computer vision
GRAVIR - IMAG - Graphisme, Vision et Robotique, Inria Grenoble - Rhône-Alpes, CNRS - Centre National de la Recherche Scientifique : FR71
Abstract : Currently it is impossible to automatically achieve high level video indexing as suggested for instance by MPEG-7. Part of the problem is the identification of moving objects in the framework of a video. It represents such a challenging problem due to the variable appearance of objects over time. This paper studies how to automatically classify tracked objects in a video based on estimated Gaussian mixture density functions of each tracked object. In the color feature space, the variability of each tracked object is modeled separately by an Gaussian mixture where the appropriate number of Gaussians is determined by the Integrated Classification Likelihood criterion. Then, the Ascendant Hierarchical Classification technique is used to identify the clusters of tracked objects. Thus, the problem is different to conventionnal classification in that the underlying measurements are not feature vectors but density estimates. The proposed framework is evaluated on the Avengers TV movie, segmented into 2749 individual objects of seven different clusters.
keyword : MOVI
Document type :
Conference papers
Complete list of metadatas

https://hal.inria.fr/inria-00548294
Contributor : Thoth Team <>
Submitted on : Tuesday, December 21, 2010 - 4:38:10 PM
Last modification on : Wednesday, April 11, 2018 - 1:54:19 AM
Long-term archiving on: Tuesday, March 22, 2011 - 2:37:42 AM

File

hammoud_IMVIP2000a.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : inria-00548294, version 1

Collections

IMAG | CNRS | INRIA | UGA

Citation

Riad Hammoud, Roger Mohr. Probabilistic Hierarchical Framework for Clustering of Tracked Objects in Video Streams. Irish Machine Vision and Image Processing Conference (IMVIP '00), Aug 2000, Belfast, United Kingdom. pp.133--140. ⟨inria-00548294⟩

Share

Metrics

Record views

175

Files downloads

123