Skip to Main Content
Probabilistic visual tracking has been an active research area in the computer vision community in the last few years. Recently, the popular approach to analyze human motion is the use of particle filtering and its extension which mainly based on the Bayes' rule. It has been pointed out that an essential structure involved in the particle filter is quite similar to that in the genetic algorithm. In the sampling stage and resampling stage of the particle filter, particles are drawn from the prior probability distribution of the state evolution. Consequently, the algorithm demands a large number of particles and computationally expensive. In this paper, we elaborate on the relationship of the particle filter and genetic algorithm, then we replace the "Evolve" step of the particle filter by the mutation and crossover operators in the GA to solve the conventional Monte Carlo methods problems. Experiments with tracking real image sequences are made to compare the performance of the two algorithm.