Spherical approach: Objects Tracking in Catadioptric System Using The GGD Distribution

The objects tracking applications are ubiquitous and largely discussed in computer vision field. Nevertheless, the toughest challenge encountered in the catadioptric system is images geometry deformations. Nevertheless, these images have significant distortions that that bring us to take them into account during processing. In this paper, we propose an adapted tracking according to spherical coordinates over the 2D image. We present a novel content based omnidirectional image provided by the catadioptric system for object tracking model using a Spherical object tracking (SOT). Our statistical approach uses a new search algorithm called Ring Search (RSA), the Generalized Gaussian Distribution (GGD) as well as the Kullback-Leibler Distance (KLD). In SOT, we split images into several macro-blocks according to spherical coordinates, each macro-block are modelled by the GGD parameters. After that, we compute the similarity between the current bloc and the candidate one onto the search zone in the next frame. Experimental results in spherical space show that the SOT outperforms efficiently the adaptive Mean-Shift in visual tracking in term of Spatial Overlapping Estimation (SOE) and the ROC curve.