@article{1310,
  author = {Intekhab Alam},
  title = {Real time Object Tracking Using Evolutionary Computation Approaches: State Space Exploration using Static & Dynamic Particle Like Sensors},
  journal = {Journal of Intelligent Computing},
  year = {2013},
  volume = {4},
  number = {3},
  doi = {},
  url = {http://www.dline.info/jic/fulltext/v4n3/1.pdf},
  abstract = {A Novel methodology has been proposed in this paper that enabled us to track objects of interest in video sequences at a much higher rates of 25-30 frames per second. This is particularly useful in applications where a real time intervention is of paramount importance for e.g. in safety critical applications and in video surveillance environment. This simplistic but very robust tracking algorithm is based on static and dynamic particle like sensors that also march along the object in video frames. The direction of propagation of these sensors is associated with different components including previous motion history and the maneuverability of the object in question and the track characteristics. A group of particles each of a pixel size bunch together to form a super particle and its mean position in the frame is used to direct other subordinate particles to explore the search space further to determine the exact boundary profiles of the object of interest. The super particle position is similar to groups best velocity profile in Particle Swarm Optimization. The main difference in this approach is that the birth of particles takes place in a circular like search space around the groups best position. This evolutionary approach inspired by biological life forms like flock of birds and a bee hive outperforms the standard particle filter based approach where the state dynamics are predetermined and are usually influenced by Newtonian Physics. Our approach dramatically reduces the number of iteration required in each frame and needs even less than 30 particles (300-500 in standard particle filter) to effectively track object of interest in a frame of medium resolution. This approach also recovered the object effectively in real time if it was partially or fully occluded for a significant number of frames and when it underwent large maneuverability which original mean shift algorithm had been unable to cope with.},
}