A novel zoom invariant video object tracking algorithm (ZIVOTA)

This paper describes a novel zoom-invariant video object-tracking algorithm (ZIVOTA). ZIVOTA extracts the object feature points and construct an affine-based model to predict the size and position of an object during the tracking process. The proposed affine model is zoom invariant, which makes it possible to track object with nonrigid size and shape. Compared to traditional frame difference and optical flow methods, ZIVOTA largely reduces the computational cost because it explores only relevant object feature points and processes smaller number of point instead of the full frame. Moreover, it is more accurate since affine transformation is viewpoint invariant.