Shape-encoded particle filtering for object detection and tracking

This thesis discusses the problem of model-based object detection and tracking. We exploit the intensity gradient information of the object boundary or any prominent shape information inside the object.; We present an optimal two-dimensional shape operator, which is robust to noise and preserves all edge information. We first derive a one-dimensional optimal step edge operator, which minimizes both the noise power and the mean squared error between the input and the filter output. This operator is found to be the derivative of the double exponential (DODE) function. We then define an operator for shape detection by extending the DODE filter along the shape's boundary contour.; Object tracking has been handled by filtering methods such as linear or extended Kalman filters, which cannot properly handle nonlinearity or ambiguity of the motion. We present a novel solution to the problem by combining shape filtering with nonlinear state estimation, and solve it using particle filtering. The measurements are derived using the outputs of shape-encoded filters. The nonlinear state estimation is performed by solving the Zakai equation, and we use the branching particle propagation method for computing the solution. The unnormalized conditional density of the solution to the Zakai equation is realized by the weight of the particle. We first sample a set of particles approximating the initial distribution of the state vector conditioned on the observations, where each particle encodes the set of geometric parameters of the object. The weight of the particle represents geometric and temporal fit, which is computed bottom-up from the raw image using a shape-encoded filter. The particles branch so that the mean number of offspring is proportional to the weight. Time updating is handled by employing a second-order motion model, combined with local stochastic search to minimize the prediction error. The amount of diffusion is effectively adjusted using Kalman updating of the covariance matrix. The problem of occlusion is investigated using the phase-space analysis of particle dynamics. The effectiveness of this approach is illustrated by tracking human body parts. (Abstract shortened by UMI.)...