Motivation
Advances in camera miniaturization and mobile computing have made it feasible to capture and process photos and video from cameras worn on a person's body and "looking out" at the world. Such an egocentric perspective of visual computing naturally ties into everyday life and provides a level of detail and ubiquity that may well exceed what is possible from environmental cameras. For instance, a lapel or shoulder-mounted camera is particularly well-suited to monitor and track daily ego-activities by recognizing objects being handled, gestures performed via hand motion and social interactions with other people. Robust and inexpensive solutions to egocentric vision would have immediate and high-impact applications in healthcare, education, entertainment and human-resource management.
The egocentric perspective casts many classical vision problems under a new
light. For instance, objects do not appear in isolated, well positioned photos.
They are embedded in a dynamic, everyday environment and constantly interact
with one another and with the wearer. Human activities are not recognized from
a passive observer at a distance, but from an active participant. Motion
analysis and tracking would be essential for egocentric video but camera
movements and viewpoints differ substantially from those in surveillance or in
movies.
The egocentric perspective provides many constraints that may help simplify vision problems. For instance, key parts of the body, such as hands and arms, are observed at consistent scales and orientations. Objects of interest are typically well-positioned in the view and often handled by the wearer. Meanwhile, the egocentric perspective also introduces a variety of challenges. For instance, visual signals from a wearable camera are poor in quality and limited in both resolution and field-of-view. Large and uncontrolled camera movements, along with motion blur, need to be accommodated. Systematic occlusions by hands and arms may also occur often. Performance requests may be challenging: mobile computing demands interactive processing speeds and low power consumption.
Call for Papers
The goal of this workshop is to call for a converged effort to understand the opportunities and challenges emerging in egocentric vision, to identify key tasks and evaluate the state of the art, and to discuss future directions. We invite submissions in all fields of vision that explore the egocentric perspective, including, but not limited to:
- Egocentric object detection, recognition and categorization
- Feature detection, tracking and matching in egocentric video
- Motion analysis, object tracking and scene segmentation with moving cameras
- Human motion, gesture and event recognition
- Localization and visual SLAM in everyday environments
- Machine learning techniques in egocentric vision
- Online learning and modeling of objects and scenes
- Data collection, benchmarking and performance evaluation
- Integration of egocentric vision with other sensors
- Applications of egocentric vision in daily life
Important Dates
| March 28, 2009 | Deadline for paper submission |
| April 10, 2009 | Notification of decision |
| April 15, 2009 | Camera-ready copies due |
Keynote Speakers
| Takeo Kanade | Carnegie Mellon University |
(list to be completed)
Workshop Organizers
| Martial Hebert | Carnegie Mellon University |
| Matthai Philipose | Intel Research Seattle |
| Xiaofeng Ren | Intel Research Seattle |
Contact Information
Intel Research Seattle
1100 NE 45th Street, 6th Floor
Seattle, WA 98105
U.S.A.
| Phone: | 1-206-545-2523 |
| Fax: | 1-206-633-6504 |
| Email: | xiaofeng.ren@intel.com |