An analogic CNN-algorithm of pixel level snakes for tracking and surveillance tasks

Abstract

This paper addresses the application of the pixel level snakes for the segmentation of moving objects. This kind of active contour techniques can handle multiple contours simultaneously without time-processing penalty as well as to manage appropriately the topologic transformations among them when this is required. The implementation into a CNNUM or a specific purpose CNN platform gives solution to the speed requirements of this kind of tasks. Particularly, we show an analogic CNN-algorithm which meets all the constrains imposed for the current CNNUM hardware implementations.