A New Foveal Cartesian Geometry Approach used for Object Tracking

Foveal vision has been used as a way for sampling and reducing the amount of data in Cartesian images for vision systems. For this sampling, there are different approaches as the Log Polar Transform, the Exponential Cartesian Geometry and the Foveal Wavelet Transform between others. In this paper a new approach to obtain the foveal sampling and its application to single object tracking is presented. The approach uses the log polar formulation for the sampling but preserves the Cartesian properties of the information in the original image. In this way, it is possible to overcome the problems of nolinearity of the Log Polar Geometry. Furthermore, it allows an easier object location as the hierarchical processing used in the Exponential Cartesian Geometries and it is easy to implement because it does not imply complex operations for the sampling of the images and the recovery of the original image through the foveated image, contrary to the Foveal Wavelet Transform. The proposed geometry has been tested in diverse images sequences where a single object is tracked successfully by appearance based methods which demonstrate the effectiveness of the proposal.