Adam Baumberg's Home Page
- Electronic Mail: amb@comp.leeds.ac.uk
- Office Phone: +44 113 233 6816
- Fax: +44 113 233 5468
- Snail Mail:
Contents
The aim of my work is to automatically process sequences of images
taken from a camera viewing outdoor pedestrain scenes. The system is
designed to extract the positions of the moving pedestrains in the
scene and follow each person throughout the sequence. Such a system
has applications in the areas of automated surveillance, animation and
human motion analysis. A prototype tracking system has been
implemented, running in real time (30 frames per second) on modest
hardware (a 100Mhz workstation).
This work is an example of the tracking and analysis of non-rigid
motion and the methods that have been developed are applicable to a
range of problems in computer vision.
Home Page
Publications
- Learning Flexible Models from Image Sequences, in proc. of
the European Conference on Computer Vision (ECCV), Stockholm, May 94
- An Efficient Method for Contour Tracking using Active Shape
Models, in proc. of IEEE Workshop on Motion of Non-rigid and
Articulated Objects, Texas, Nov 94
- An Adaptive Eigenshape Model, in proc. of the British
Machine Vision Conference, Birmingham, Sept 95
- Learning Spatiotemporal Models from Examples, in proc. of
the British Machine Vision Conference, Birmingham, Sept 95
(awarded prize for best scientific paper at BMVC95)
Research Area
Home Page
Demos
Background
The following image sequences were obtained by applying our
"pedestrian tracking" system on image sequences from various outdoor
scenes. In the first two cases the camera was fixed (i.e. stationary
with fixed zoom, apperture etc) making the tracking task more
straightforward. The input images are shown with the computer's
interpretation of the sequences (the position and shape of each
person) superimposed in colour. Each tracked pedestrian is represented
by a different coloured outline.
The system is based on a 2D flexible contour model of the silhouette
of a pedestrian. The model was generated automatically from
training image sequences.
- Tracking a pedestrian with a fixed camera
- Tracking several pedestrians with a fixed camera
- Tracking with a moving camera
- Tracking with a zooming camera
Publications
Home Page