Indoor Scene Understanding

In computer vision, holistic indoor scene understanding from images is a complex and important task that requires solving several subtasks simultaneously. Advanced indoor scene understanding systems can be used to build personal robots that can clean rooms and to build automated surveillance systems to improve public safety. In this research, we consider the problem of indoor scene understanding from both RGB and depth images. We are interested in distinguishing the major planes from objects that appear in a scene and our goal is to classify each pixel from an indoor image as one of the following categories: walls, the ceiling, the oor or objects. We present how to construct descriptors from RGB and depth images and demonstrate that images with depth can be used to improve the performance of our system. Our recognition system is based on support vector machines and a Markov random eld. In addition, we explore how to utilize global consistency to improve the performance of our system. In particular, we propose two ltering algorithms that can improve the performance over state-of-the-art computer vision systems