|
III.1. Object Registration and Tracking.
III.2. Fusion of frequency and
spatial domain data for motion estimation and segmetnation
Research in digital video
technology is constantly evolving, as new communications systems are
increasingly using multimedia content. Thus, the analysis of multiple motions in
video is particularly important, as it allows the characterization of the
sequence, and its segmentation into differently moving entities. Numerous
methods have been developed for the motion estimation in video, using
information from the spatial or spectral domain. We investigate new approaches
for the analysis of multiple motions in video, which integrate Fourier and
spatial domain information.
The tasks of interest are finding the number of moving objects, velocity
estimation, object tracking, and motion segmentation. The proposed, hybrid
approach, performs the motion estimation based on frequency domain information,
but also uses spatial information for precise object localization. Unlike
existing frequency domain methods, its use is not limited to constant
translational motions, but can also address the problem of roto-translational
and non-constant motions. The validity, effectiveness, and potential of the
proposed approaches is verified through experiments with both
synthetic and real video sequences.
Another task of interest tackled by our group is the
estimation of the positions of multiple translating objects in a video sequence,
via a different hybrid approach to motion estimation.
We suggest minimizing the squared error in both the spatial and frequency
domains. We show that, both theoretically and experimentally, the spatially
global nature of FT leads to a motion estimation error that is much lower than
that obtained via spatial motion estimation. On the other hand, spatial analysis
is useful for accurate segmentation. We thus describe a new, hybrid approach
combining the above two estimates of motion and segmentation. We examine the
robustness of minimizing the error terms in both domains, both theoretically and
experimentally. Experiments with real and synthetic sequences demonstrate the
capabilities of the proposed algorithm.
There is great interest in the detection and analysis of
repetitive motions, as they are appear in numerous applications. Their analysis
can lead to useful results in recognition and classification, concerning human
and animal activities, or even dynamic objects. It also plays a central role in
characterizing the motions in a video as purely periodic, purely translational,
or a combination of both. Existing work on periodic motions takes place in the
spatial domain and involves the extraction of motion trajectories via feature
tracking or region matching. This often requires human intervention, or is
otherwise computationally costly, and often unreliable. Also, current methods
cannot deal with motions that are not purely periodic without some form of
pre-processing. We develop a novel approach for the detection and analysis of
periodic trajectories over a static background, which avoids the shortcomings of
existing spatial approaches by combining time-frequency and spatial analysis. It
is able to detect periodicities that are pure or superposed on translational
motions, thus allowing the characterization of the motions in a video. We show,
both theoretically and experimentally, that our method is robust under Gaussian
deviations from strict periodicity as well. The capabilities and improvements of
this approach on existing methods are shown through experiments on both
synthetic and real sequences.
III.3. 3D Motion and Structure from Image Motion, computed from a video
sequence showing a scene with moving objects and taken from a still or moving
camera.
III.4. 3D Motion and Structure from Two-View motion (Temporal
Stereo) , based on the disparity between two images
taken from one camera but at two different times.
III.5. 3D Surfaces and
Motion from multiple video sequences, showing moving objects and
taken from multiple cameras.
II.6. 3D Surfaces from
Active Stereo, Focus, Vergence, Zoom and Aperture, from the
parameter values assumed during active acquisition of video
sequences from multiple cameras.
III.7. Integrated Analysis
Guided Synthesis of Video Sequences, for 3D based compression,
augmentation and synthesis of video sequences, aimed at enhancing
the perception of specific visual characteristics, e.g.,
motion.
|
