We discover quasi-static objects, objects that are stationary during
some interval of observation, across image sequences acquired by any
number of completely uncalibrated cameras using only temporal (no
spatial) information.
Brandon Sanders (Ph.D. Thesis Work)
Randal Nelson (Thesis Advisor)
Rahul Sukthankar (Summer Advisor 2001)
Dana Ballard (Thesis Committee)
Chris Brown (Thesis Committee)
Kyros Kutulakos (Thesis Committee)
A. Murat Tekalp (Thesis Committee)
Brandon Sanders
OD Results on a 3 camera sequence in which the spoon and can arrive
simultaneously, the keyboard is always partially occluded by either
the bowl or the can, and the can and keyboard mutually occlude each
other. Temporal information alone is sufficient to group the pixels
in and across the uncalibrated cameras.
The Quasi-static Object Discovery project is part of a larger effort
to equip computers with spatial awareness. Our temporal object
discoverer (TOD) provides percepts of objects to other processes that
reason about relationships between objects and actions involving
objects. TOD discovers quasi-static objects, objects that are
stationary during some interval of observation, within image sequences
acquired by any number of completely uncalibrated cameras.
In this project we ignore spatial information (for example
neighborhood relationships) and concentrate on a theoretically well
founded treatment of temporal information. Our framework ignores
distracting motion, correctly deals with complicated occlusions, and
naturally groups observations across cameras. The sets of 2D masks we
recover are suitable for unsupervised training and initialization of
object recognition and tracking systems.
For each pixel (sensel) we generate a signature that encodes the
pixel's temporal structure. Using the set of temporal signatures
gathered across views, we hypothesize a global schedule of events and
a small set of objects whose arrival and departures explain the
events. Under certain observability conditions, we provably create a
valid and complete schedule of all quasi-static objects in the scene.
Likewise, our Quasi-Static Labeling algorithm generates the
maximally-informative mapping of pixels' observations onto the objects
they stem from.
http://www.cs.rochester.edu/~sanders/objectdiscovery
Object Discovery, OD, vision, TCC, temporally coherent, temporally
coherent cluster, Brandon Sanders, Rahul Sukthankar, Randal Nelson,
Segmentation, Quasi static, temporal, home, homepage, home page, URCS,
University of Rochester Computer Science Department, Computer Vision