List of Past Events
Adventures in Visual Odometry: Egomotion estimation for robotics and wearable applications
Dr. James Bergen
Monday, April 03, 2006, 02:00pm - 03:00pm
Sarnoff Labs, Princeton
Accurate position sensing is a critical element of mobile robotic systems and also plays an important role in coordination and information sharing in multi-entity operations involving humans. The developments of GPS and Inertial Navigation Systems have provided powerful methods for determining location and motion but are not applicable in all situations. Indoor or otherwise signal-impaired environments do not easily allow use of GPS and highly accurate inertial systems are (so far) too heavy, expensive, and power-hungry for many applications. Visual navigation using video cameras to estimate egomotion can operate in a large range of different environments and provides an interesting alternative to GPS and inertial systems. Visual analysis can be used in conjunction with GPS or inertial sensors where appropriate. Egomotion estimation involves analysis of the apparent motion of static objects seen by a camera to determine incremental displacements and rotations of that camera. Visual position sensing further requires that these incremental relative pose estimates be accumulated accurately to produce an estimate of overall change in position with respect to some reference. For many applications of interest these estimates must be produced in real-time and with low latency. I will describe Sarnoff�s activities in visual egomotion determination for autonomous robotic mobility (�visual odometry�) as well as some ongoing activities involving human-wearable sensors.
Dr. James Bergen