Tracking People and their Faces
Our system localises and tracks peoples' faces as they move through a scene. It integrates the following techniques:
Faces are tracked robustly by integrating motion and model-based tracking. Here are some MPEG movies to view:
- Tracking Collette and her extracted face sequence .
- Tracking Jon with low resolution and poor lighting conditions
- Tracking two people simultaneously: lock is maintained on the faces despite unreliable motion-based body tracking.
This work forms part of the EPSRC Integrated Machine Vision (IMV) Project:
"Real-time Target Identification for Security Applications"
Motion estimation employs spatio-temporal filtering.
Moving people are modelled as Gaussian motion clusters and tracked using Kalman filters.
A model of facial appearance is used to detect and track faces. This model takes the form of a neural network trained
on thousands of images of faces and `face-like' patterns.
Stephen McKenna,
stephen@computing.dundee.ac.uk