Color based Human Detection and Tracking Algorithm using a Non-Gaussian Adaptive Particle Filter

In this paper, an algorithm is presented that can detect and isolate moving skin colored pixels, or more generally humans, from a static background, and then track them using a non-Gaussian recursive Bayesian Particle filter. Particle filter (PF) is an adaptive filter based on sequential Monte Carlo methods, and represents probability densities in terms of particles. It is of vital importance that the elements of non-linearity and non-Gaussianity are included so that the physical system being modeled can be more and more close to the real world, and all the estimations and analysis carried out could hold practical. Key aspects of our tracking algorithm such as background removal with illumination compensation, skin color detection and Particle filter implementation have been explained and demonstrated. This is followed by the results of our tracking algorithm and a conclusion.