An Efficient Algorithm for Multi-Target Human Motion Tracking

Effective human motion tracking is necessary for all surveillance based biometric applications. Human motion being highly non-linear and non-modular has always been a challenging and difficult task. To make it more challenging, in real scenarios,there are always multiple targets to be tracked. These multiple targets do not behave independent of each other and often show interestingly complex inter-dependence. In this work particle filter based multiple target tracking system is proposed which along with tracking the individual profiles also models the inter-connectivity among the profiles of multiple modalities. Posterior probabilities are used to design ‘multiple hypothesis tracking framework’ and for performing actual tracking, ‘Markov Chain Monte Carlo’ methods are used. The system was tested using moviedb and labdb databases and accuracy of 91.7% was obtained.