Effective motion field description based on affine models and global motion information

In this paper we study the possibility to estimate reliable motion field by considering both a global motion, due to camera parameters changes and a local motion, due to the displacement of the image objects. By considering two images In and In-k a first motion field is estimated using a block matching algorithm. Thanks to this information, the global motion parameters (horizontal/vertical pan and zoom factor) are estimated. One of the two images is then compensated by the estimated global motion. A combination of a block matching and a differential algorithm is used to obtain a dense local motion field. Simulation results indicate that the detection and compensation of the global motion are essential for good motion filed estimation and motion compensated prediction. Moreover the local motion field is used as input for a segmentation algorithm based on affine model, in order to detect the moving object present in the scene.