Combining depth with appearance images for object detection using online learning

Abstract

This Master's Thesis presents a new developed algorithm that merges Appearance and Depth Images in order to perform the detection of objects in scenes. This new algorithm uses the approach of the Online Learning Random Ferns, proposed by Villamizar et al at [1]. Although Villamizar's detector takes only into account appearance information (RGB), the new developed algorithm will also use depth images for creating a robust Object Detector. In this way, it has been studied the best method for combining both sources of information in the same algorithm (appearance and depth). Several approaches to solve this, have been created and tested through a set of experiments, studying its effects over different scenarios and parameters configuration.