Camillo J. Taylor

Camillo J. Taylor

Associate Professor

GRASP Laboratory

Computer and Information Science Department

University of Pennsylvania

Brief Resume

Contact Information:

3330 Walnut St, Levine Hall Rm 474 (GRW)Philadelphia, PA 19104-6309

email: cjtaylor@cis.upenn.edu

Tel: (215) 898 0376

Fax: (215) 573 2048

Office Hours: Tuesday and Wednesday 2-3PM

Current Courses:

CSE 313/Math 313 Computational Linear Algebra Spring Semester 2004 - NEW COURSE
CSE 401: Senior Design Spring Semester University of Pennsylvania
CSE 371: Digital Systems Organization and Design Spring Semester University of Pennsylvania
CIS 580: Machine Perception Fall Semester University of Pennsylvania

Research Interests:

Broadly speaking, my research interests can be divided into two areas: reconstructing and rerendering 3D scenes from 2D images and vision guided robotic systems. I have listed the projects I have undertaken in these areas below. Related publications can be found in the project pages and on my publication list.

Scene Reconstruction and Image Based Rendering

Structure and Motion from Line Segments

As part of my doctoral dissertation at Yale I developed an algorithm to recover the structure of a rigid environment from a set of straight line correspondences.

Facade project.

When I got to Berkeley I worked closely with Paul Debevec to apply these ideas to the reconstruction of architectural scenes as part of the Facade project.

Reconstruction of Linearly Parameterized Models from Single Images

Here's a link to some recent research on reconstructing polyhedral models from single photographs taken with an uncalibrated camera done with David Jelinek

Reconstructing Articulated Figures

This project dealt with the problem of recovering models of articulated figures, including humans, from single snapshots acquired with an uncalibrated camera. The resulting reconstruction algorithm can be used to recover stick figure models from newspaper photos or web site photos.

VideoMoCap

This project extends the work on recovering articulated models to deal with video imagery. This interactive system permits us to recover the motion of actors based on uncalibrated video sequences.

VideoPlus

A method for estimating the trajectory of a moving camera and the appearance of a scene from omnidirectional video sequences has been developed. The end result of our procedure is an omnidirectional video sequence where each frame is augmented with pose information and a sparse 3D model of the scene. The end user can then tour the scene by interactively navigating through the video sequence.

Motion Stereo for View Synthesis

In this work we employ epipolar plane image analysis to recover the positions of edge features in the scene. Once we have recovered the positions of these salient points we can use a morphing technique to synthesize new views of the scene.

Vision Guided Robots

	RJ the exploring robot As part of my dissertation research at Yale I developed algorithms that would enable a mobile robot equipped with a vision-based recognition system to carry out a systematic exploration of its environment. This robot was able to successfully navigate around our office complex searching for recognizable landmarks and building a vision based topological map.
	StereoDrive While at Berkeley I worked on the development of a vision-based lateral control system to control an autonomous motor car. This was a joint research project between the California PATH project and Honda Research and Development. The system we developed was successfully demonstrated at the NAHSC Demo in August 97.
	Multibots Recently we have been having lots of fun with teams of small inexpensive mobile robots equipped with omnidirectional cameras and other sensors. We have developed a number of applications based on these platforms including, manipulation, cooperative localization and coordinated map making.
	Urbie The Stair Climbing Robot This project involved developing methods for guiding a treaded robot up a staircase based on inputs from a camera.

Optimal Assignment of Conference Papers to Reviewers

In the course of helping to organize the 2006 edition of the IEEE Conference on Computer Vision and Pattern Recognition. I developed an algorithm that was used to help automate the assignment of reviewers to papers based on a set of affinities indicated by the conference area chairs. You can find a short paper detailing the theory behind this algorithm here. You can also download Matlab codes that implement the algorithm from the links below.

Paper Describing Assignment Algorithm

Matlab Codes:

	Structure and Motion from Line Segments As part of my doctoral dissertation at Yale I developed an algorithm to recover the structure of a rigid environment from a set of straight line correspondences.
	Facade project. When I got to Berkeley I worked closely with Paul Debevec to apply these ideas to the reconstruction of architectural scenes as part of the Facade project.
	Reconstruction of Linearly Parameterized Models from Single Images Here's a link to some recent research on reconstructing polyhedral models from single photographs taken with an uncalibrated camera done with David Jelinek
	Reconstructing Articulated Figures This project dealt with the problem of recovering models of articulated figures, including humans, from single snapshots acquired with an uncalibrated camera. The resulting reconstruction algorithm can be used to recover stick figure models from newspaper photos or web site photos.
	VideoMoCap This project extends the work on recovering articulated models to deal with video imagery. This interactive system permits us to recover the motion of actors based on uncalibrated video sequences.
	VideoPlus A method for estimating the trajectory of a moving camera and the appearance of a scene from omnidirectional video sequences has been developed. The end result of our procedure is an omnidirectional video sequence where each frame is augmented with pose information and a sparse 3D model of the scene. The end user can then tour the scene by interactively navigating through the video sequence.
	Motion Stereo for View Synthesis In this work we employ epipolar plane image analysis to recover the positions of edge features in the scene. Once we have recovered the positions of these salient points we can use a morphing technique to synthesize new views of the scene.