Contour - based tracking with Particle Filters – The Condensation Algorithm

From Wikipedia: definition of Visual Tracking Visual tracking is the process of locating a moving object (or several ones) in time using a camera. An algorithm analyzes the video frames and outputs the location, optionally in real time. A visual tracking algorithm is based on a motion model which describes how the image of the target changes depending on a vector of motion parameters. When the target is a rigid 3D object, the motion model defines its aspect depending on its 3D position and orientation. The role of the tracking algorithm is to analyse the video frames in order to estimate the motion parameters. These parameters characterize the location of the target. Applications of real-time 3D tracking cover many fields of interest, for which commercial products are increasingly becoming available. Today there are several known approaches for this task, and we consider here some advanced techniques, by organizing them into • Point-based tracking: Tracking of single point features, followed by least-squares object pose estimation. • Contour-based tracking: Detection of the object boundary line (e.g. active contours, or Condensation algorithm) as it deforms with the roto-translation of the object in space • Template-based tracking: Registration of the whole object surface, given as a triangular mesh, together with its texture (i.e. the surface appearance) Lecture Slides from our WS06/07 Course " Advanced 3D Tracking Methodologies " provide also useful material for this Seminar. They are available at Please visit also the following Webpage for further interest (NOTE: the page is still " work in progress ")