Stairs detection and modelling using RGB-D images

Abstract

It is a great challenge for mobile robots to be able to overcome stairs. This could prove very useful in "Urban Search and Rescue" (USAR) scenarios, domestic service robots and other similar scenarios. This diploma thesis aims to present a method to detect stairs in an urban environment, describes them precisely in terms of numbers and dimensions and finds out the camera's parameters like pose, height and distance to stairs. In the domain of stair detection, our approach provides information for four types of situations with respect to stairs: (i) Approaching the first step. (ii) traversing the first step. (iii) traversing the Flight-of-Stairs. (iv) Landing. In this thesis we present an approach to detect and model stairs using 2D and 3D computer vision algorithms that process a stream of inputs of an RGB-D camera to detect stairs and to construct and maintain an incremental global stairs model from successive stairs observations.