Developing a Method for 3D Scene Understanding Using Image Sequence

Abstract

Indoor scene understanding is a challenging problem in computer vision. To achieve an accurate solution for this task, a model that can exploit discriminating information between different scene categories and objects is necessary. This thesis presents a framework for scene understanding which includes several components of learning models, segmentation, object recognition and tracking. A comprehensive study for supervised learning models for recognizing indoor scenes is presented. The study compares between several “Shallow Learning” models against the recent approach “Deep Learning”. Furthermore, the robustness of methods is tested against environment changes such as: contrast degradation, additive blurring and additive noise. A segmentation method is proposed for object