Abstract:
The extended Kalman filter (EKF) and unscented Kalman filter (UKF) are
standard techniques for performing recursive nonlinear estimation. However, these
methods are far from being optimal. Non-Gaussian and non-linear estimation problems
are resolved in an optimal way (approximately) using recursive Monte Carlo methods or
Particle filters with better performance as compared to that of the EKF and UKF. In
multiple targets tracking, a difficulty arises in keeping track of the individual objects as
they come close to each other and then separate and also when new objects come into
view and/or existing ones disappear. Association of the available measurements to the
correct objects is the fore most important aspect in multiple objects tracking problems.
The data association process is helpful in such cases that helps tracker keep correct track
of individual objects. This thesis presents a tracking method based on the particle filter
and the probabilistic data association (PDA) hypothesis calculations. This algorithm is
used to estimate the position and size of multiple targets in noisy video sequences, thus
showing that it is able to answer the data association problem. The uncertainty of the
measurement origin can be handled by the proposed algorithm. A data association
iii
technique based on the joint probabilistic data association (JPDA) is utilized. A
comparison of tracking with classical Kalman Filter with PDAF, SSRLS with PDAF and
Particle filter with JPDA is carried out. Performance comparison of tracking using
Particle filters with and without JPDA is also demonstrated through simulations.
In most of the practical applications, it is observed that the recorded videos have
some noise which may be due to bad weather (light, wind, etc.) or due to problems in
sensors. Moreover, when a video is transmitted or stored from one data storage location
to another loss of some information and addition of some noise generally takes place. A
comparison of tracking with various noise levels is also made through simulations.
