|
(See ISLA for the new IAS website. Most maintenance activities on this old website have stopped.)
Image Sequence Analysis
Observing and analyzing
moving objects is the goal of Image Sequence Analysis.
Motion estimation
The extraction of image motion, the 2D motion vectors a.k.a. optical flow,
from a black & white image sequence is reasonably understood. However,
for color images it is not clear how to handle the multi-channel character
of the color bands. In this project we further investige the fusion of
feature matching techniques and optical flow techniques in color space.
Structure from motion
From the image motion a reconstruction of the relative depth
to the environment is generally possible. We have found the affine model
to be particulary feasible for structure from motion. Our extension is
based on taking the temporal dynamics in the optical flow into account.
Motion segmentation
Motion based segmentation is the extraction of 'objects' on the basis of
their motion. Grouping of similar regions with similar motion is the basis
for motion segmentation. We investigate the fusion of color edges and motion
edges.
Motion recognition
Motion recognition is the classification of motion type. This implies a
model and a classifier. Recognition is based on both kinematic and dynamic
models. Typical techniques for classification of sequences are Neural Networks
and Hidden Markov Models.
Observing human and animal motion
Applications such as
- video databases
- wireless virtual reality interfaces
- smart rooms
- low bandwidth video compression
- security monitoring
are just some examples of systems which have a common
need to track and interpret human behavior.
In this project we focus on this area of human detection, localization,
tracking, gesture and gait recognition.
People
References
Dev97a "Confidence Measures for Image Motion Estimation", IROS 97.
Dev97b "Visual Navigation on the temporal development of the Optical
Flow", RWC 97.
Dev98 "Where are you driving to? Heading direction from optical flow",
ICRA 98.
Sponsors
This project is sponsored by the Ministry of Economic Affairs, as an IOP.
Partners in this project are Noldus Information BV and Eagle Vision.
|
|