News Highlight

Title: Toward practical approaches to understanding scenes from images and 3D point clouds

Date: April 09, 2012

Time: 2:00pm-3:30pm

Location: Babbio 321

Speaker: Prof. Martial Hebert (http://www.cs.cmu.edu/~hebert/), Department of Computer Science, Carnegie Mellon University

Abstract:

I will first give an overview recent and current work in the cmu vision group in the area of scene understanding, briefly discussing ideas on combining statistical and reasoning techniques, incorporating external, non-visual knowledge, and discovering models from unsupervised data. I will then focus of one aspect of this work in which we want to generate efficiently semantic segmentation of scenes. A typical approach to this type of problems uses a global model, e.g., a based on a graphical model formulation, to estimate the "optimal" label configuration in the image. We proposed a couple of years ago an alternate formulation in which the problem is decomposed into smaller, tractable partial labeling problems. I will show how we extended this initial approach to 3D point clouds and, more recently, how we are moving these ideas to robotic applications, and current extension to time sequences and other developments.

Biography:

Martial Hebert is a Professor in the Robotics Institute at Carnegie-Mellon University. His interest includes computer vision, especially recognition in images and video data, model building and object recognition from 3D data, and perception for mobile robots. His group has developed new approaches for object recognition and scene analysis in images and in video sequences. Recent development include scene understanding for mobile robots from images and 3D point clouds, and object discovery from unsupervised data.