GPI Seminar Series: Rafiq Sekkal

Rafiq Sekkal, Joint hierarchical and multiresolution image segmentation and door detection and tracking for indoor localization 
Friday February 22nd, at 12:00, Seminar Room D5-007

Abstract:

First, we develop a new scalable segmentation algorithm called JHMS (Joint Hierarchical and Multiresolution Segmentation) characterized by region-based hierarchy and resolution scalability. As a pyramidal approach, the image is considered as a set of images at different levels of resolution. The segmentation is based on split-and-merge technique: from a quadtree portioning, a hierarchical merging process is performed to recover homogeneous regions. Multiresolution implies that a segmentation of a given level is reused in further segmentation processes operated at next levels so that to insure contour consistency between different resolutions. Each level of resolution provides a Region Adjacency Graph (RAG) that describes the neighborhood relationships between regions within a given level of the pyramid. Region label consistency is preserved thanks to a dedicated projection algorithm based on inter-level relationships.

Second part of my works focus on landmarks extraction for indoor localization. Actually, doors represent meaningful landmarks for robot localization. In that way, we develop vision based door detection and tracking framework based on geometrical features. Actually, doors are well characterized by their doorposts, and respect a certain rectangular shape, so the aim is to find shapes in the image that fit well the doors' model. We start first by estimating the global structure of the environment; indeed, a wall/floor segmentation is applied in order to determine the door search areas. After that, we retrieve doors among all lines extracted from the image, and those that are located on walls. For each detected door, a 2D tracker is initialized; the tracking is based on the estimation of doorposts motion.