Motion estimation of deformable target over infrared camera video in coarse resolution with possible frame-out of target by particle filter

Motion estimation of deformable target over infrared camera video in coarse resolution with possible frame-out of the target by an elaborated particle filter has been proposed. Target state consists of location factor and shape factor of the target, where the location factor is position of the center of gravity of the target over the image plane, and the shape factor is a set of connected pixels that forms appearance of the target over the image plane. The location factor is formulated in real number space, while the shape factor is formulated in pixel based manner. State space model has been formed by a system model to represent a smooth motion of the target and smooth change of its shape, and by an observation model to evaluated likeliness of the state based on an appearance of the target over the image plane. We have employed a mixture model for the location factor of the system model consisting of second order difference equation for smooth change and uniform distribution over the image plane to cope with abrupt change of the location due to lack of frame rate in the video. For the smooth change of the target shape, we propose to use a Markov Chain Monte Carlo move on the set of connected pixels. Experiments on some real video images explore the performance of the proposed method through a prototyped implementation by simplifying the method with rough approximation of the target shape with a rectangle.