Comparative Analysis of Tracking Algorithms through Video

Kalman filtering was introduced by R.E. Kalman in 1960 as a way to predict the state of system that was subject to noisy measurements. The measurements were assumed to have Gaussian noise which make the actual state somewhere in the middle but unknown. Since the first implementation the Kalman filter has been used extensively in signal processing and later introduced into object tracking. Using state equations from physics on a moving object a newly predicted state can be estimated based on time between observations the only difference between signal processing and object tracking is working in an additional dimension.
The Particle filter (sequential monte carlo (SMC)) was introduced in 1993 by Gordon et al. in the paper ’Novel approach to nonlinear/non-Gaussian /Bayesian state estimation’ which discussed a new way for state space estimation by continuously resampling an estimated distribution and reducing the error based on actual observations. The inherent advantage to this method is that it requires no knowledge of how the object motion needs to be modeled.
In this paper I intend to do a comparative analysis of algorithms on video data and how well they are able to track an object in motion as a saved video or if given the video as though it were sequenced in real-time. Another aspect of exploration is the determination of which algorithm performs better under variable frame rates or how slow can each frame rate be before there is a severe lack of data to track objects in a scene accurately