Predicting object occupancy on the floor from RGBD images of indoor scenes

Abstract

This thesis presents an approach to predict the occupied area on the floor in an image of an indoor scene. The goal is to be able to obtain navigable areas even in cluttered indoor environments. This algorithm could be used in the field of robotics where robots need to navigate through a room while being mindful of the surrounding objects. The results are quite close to the ground truth, as exemplified by the false positive, false negative, precision and recall rates. Using this algorithm improves the label predictions.