The Robotics Institute
 Search the site
RI | Research | Labs | 3D Computer Vision Group

Text only version of this site

[Lab image] 3D Computer Vision Group
This lab is no longer active.

Head: Martial Hebert
Contact: Martial Hebert (hebert@ri.cmu.edu)

Mailing address:
Carnegie Mellon University
Robotics Institute
5000 Forbes Avenue
Pittsburgh, PA 15213

Location:
NSH A401
(412) 268 7931

Associated center: VASC

For more information, see this lab's homepage.

Jump to: Lab Description | Personnel | Projects | Publications

Lab Description

This lab has been superseded by the Vision and Mobile Robotics Lab

Our group studies fundamental problems of 3-D computer vision with a concentration on applications of modeling, recognition, and free-form surface matching. We investigate these problems over a range of physical scales and with a variety of sensors. To efficiently solve surface matching problems, we have developed novel representations, and we are working on algorithms to incorporate machine learning and artificial intelligence techniques into the recognition process.

The domains we consider cover a range of scales, including small-scale object modeling and recognition, medium-scale modeling of building interiors, and large-scale environment mapping and localization. We exploit the fact that many aspects of these problems that are independent of scale, which allows us to use a common base of general algorithms. At the same time, we explore the issues specific to each domain.

From a sensing standpoint, we employ a variety of devices capable of measuring 3-D data. Current sensing systems include desktop laser rangefinders, field-deployable scanning and single-line laser rangefinders (mounted on autonomous vehicles), stereo camera systems, and a 3-D sonar system.

For representing free-form surfaces, we primarily use triangular meshes. Our group has developed new representations to aid in solving surface matching problems, including harmonic maps and spin-image signatures. These representations are examples of surface signatures, meaning they distinctively encode local properties of arbitrary points on a free-form surface.

Past members

Name Title Email Address
Owen's personal homepage Owen Carmichael Visiting Researcher
Samuel Drulhe Visiting Scholar
Martial's personal homepage Martial Hebert Professor hebert@ri.cmu.edu
Daniel's personal homepage Daniel Huber Systems Scientist dhuber@cs.cmu.edu
Andrew's personal homepage Andrew Johnson PhD Student, RI
Taku Osada Visiting Scholar
Dennis's personal homepage Dennis Strelow PhD Student, CS
Scott Thayer Systems Scientist
Dongmei Zhang PhD Student, RI

Past projects

3D Object Recognition - We are applying our techniques for surface representation and matching to object recognition problems; for example, we have used 3D object recognition as part of the Artisan system for interior mapping.
3D Terrain Mapping - We are developing methods for building large-scale, topographic maps of unstructured outdoor environments.
Advanced Sensor Based Defect Management at Construction Sites - This research project builds on, combines and extends the advances in generating 3D environments using laser scanners.
Automatic 3D Modeling from Range Images - A system for generating 3D models of real-world objects without manual or mechanical aids.
Exploitation of 3-D Data - The E3D project will develop technology to detect, characterize and recognize vehicular targets in 3-D data.
Humanoid Vision - We are adding visual recognition and navigation to Honda's humanoid robots
MARS2020 - This project seeks to develop softwares needed to program autonomous mobile robots in partially known, changing, and unpredictable environments.
Model Building - Surface registration is applied to a variety of problems, including object modelling and mapping of large areas.

Recent publications [View all 39 publications]

The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.
For updates and comments, please see these instructions.
This page maintained by robotwebmaster@ri.cmu.edu