A Study of the Automatic Object Extraction Algorithm on the Image Sequence Under Scaling Ttransformations

The work is devoted to the problem of moving objects extraction in a video sequence. The main feature of the problem considered is the presence of geometric transformations. The novelty of the proposed approach is considering the impact of scaling transformations to the image observation model. Geometric transformations are caused due to the spatial movement of the image sensor. To evaluate transformation parameters, a multi-template approach is used. Then object extraction is performed using spatio-temporal background subtraction algorithm. The results of experimental studies on an extended video sequence dataset are presented. For test purposes, we use natural video sequences taken with a camera in the visible wavelength range. In the process of research, the influence of the standard deviation of the scale distortion estimation on the quality of object localization is considered. The research shows high efficiency of the proposed algorithm.