Noise tolerance analysis of marginalized particle filter for target tracking

In this present world even though technology has improved tremendously, real time target tracking is still considered to be an important and challenging research area. In target tracking the aim is to estimate the kinematic state of an observed object. Particle filter offers a general solution for such problems, however the main concern is its computational complexity which increases quickly with state dimension. By extracting the states which are linear from the system dynamics, this problem can be solved. Marginalized Particle filter can be obtained by an efficient combination of particle filter and kalman filter. It exploits the linear substructure and its analytical relationship among state variables which is available in the model, so as to improve the efficiency and accuracy of a regular particle filter. This paper analyzes the performance of filters based on the effect of measurement noise level. Simulation is done in the case of a typical target tracking application using both particle filter and marginalized particle filter. From the analysis it can be concluded that, the marginalized particle filter which is better than particle filter in terms of complexity and performance, is found to be more tolerant to measurement noise levels.