Object tracking with spatio-temporal blob

Catherine Achard, Ghilès Mostafaoui, Maurice Milgram
2005 IAPR International Workshop on Machine Vision Applications  
We propose to develop a tracking algorithm of objects or humans, based on kinematics, with a fix ed monochromatic camera, without any knowledge on the sequence: siz e, shape or number of objects are unknown and can evolve with time. For this purpose, we first make a motion detection, then, as we suppose that people move locally in a consistent way and thus draw a regular trajectory in the spatio-temporal space, we modeliz e locally the 3D moving points by a mix ture of Gaussians where each
more » ... ian represents a trajectory . These points described only by their coordinates ( x , y , t) are clustered with the Classification Ex pectation-Max imiz ation ( CEM) algorithm. We will shown on tracking results, that this method based only on kinematics, manages the number of objects to track, occlusions and poor segmentations ( over-or undersegmentation) .
dblp:conf/mva/AchardMM05 fatcat:wdjfpfig2faqrmth6amrz3wuum