Abstract: Taking advantage of the invariant characters for translation and rotation of the Log-polar transformation, a scale tracking method based on the ellipse log-polar transform was investigated to resolve the problem of scale change object tracking. Firstly, a saliency weighted mean shift was presented to locate the objects spatial orientation, then the candidate object region was translated to the ellipse log-polar transformation and integrated along the scale axis. The new proposed method estimated the targets scale parameters according to the maximum correlation coefficient in the transform domain. Experimental results demonstrated that the algorithm could adapt to the objects scale changes and the tracking error was lower. Compared with the traditional, it had a better robustness.

Key words: object tracking, salient weighting, scale variant, mean shift, ellipse log-polar transform, correlation matching