Abstract

After a brief overview of our current activities in user-centred processing, data-driven behaviour recognition and 3D face analysis, I will analyse a number of object representations that are suitable for localising objects and look at the trade-off between descriptiveness and robustness to scaling, rotation and partial occlusions. Next, I will cover the advantages and the limitations of deterministic target tracking approaches and then motivate the use of more efficient and accurate video tracking algorithms. In particular, I will present a filtering framework with feedback from scene contextual information that is based on the Probability Hypothesis Density (PHD) filter. This PHD filter can deal with a varying number of targets and reduces the growth in complexity with the number of targets from exponential to linear. In addition to this, the PHD filter spatially modulates its strength based on learned contextual information and can compensate for missing detections and reduce noise and clutter. This framework can be combined with any object detectors that generate positional and dimensional information of objects of interest. I will discuss the performance improvement due to the introduction of the context modelling through experimental results on a large video dataset using a standard evaluation protocol. (Joint work with E. Maggio)
About Andrea Cavallaro 
Dr. Andrea Cavallaro is Reader in Multimedia Signal Processing at Queen Mary University of London. He received the Ph.D. in Electrical Engineering from the Swiss Federal Institute of Technology (EPFL), Lausanne, Switzerland, in 2002 and the Laurea (Summa cum Laude) in Electrical Engineering from the University of Trieste in 1996. Dr. Cavallaro was awarded a Research Fellowship with British Telecommunications (BT) in 2004/2005; the Royal Academy of Engineering teaching Prize in 2007; three student paper awards on target tracking and perceptually sensitive coding at IEEE ICASSP in 2005, 2007 and 2009; and the best paper award at IEEE AVSS 2009.  Dr. Cavallaro is Associate Editor for the IEEE Signal Processing Magazine, the IEEE Transactions on Multimedia, the IEEE Transactions on Signal Processing and for the EURASIP Journal on Information Security. He served as elected member of the IEEE Signal Processing Society, Multimedia Signal Processing Technical Committee (2006-2009), and as Guest Editor for several journals including Computer Vision and Image Understanding (Elsevier) the International Journal of Computer Vision and the IEEE SPM. He also served as General Chair for IEEE/ACM ICDSC 2009, BMVC 2009, M2SFA2 2008, SSPE 2007, and IEEE AVSS 2007. Dr. Cavallaro was Technical Program chair of the European Signal Processing Conference (EUSIPCO 2008), of WIAMIS 2010 and has been a member of the organizing/technical committee of several conferences, including ACM Multimedia, IEEE ICME, IEEE ICIP, SPIE VCIP, IEEE AVSS, ACM/IEEE ICDSC, ECCV-VS, and PETS. He has authored more than 100 papers, including 10 book chapters.