Skip to Main Content
There is an increasing need in biometric deployment, especially gait, for extraction in unconstrained scenes. Extraction in such circumstances is problematic and requires noise-resistant algorithms. We describe novel methods that enable generic extraction of moving objects, but particularly walking people, from large outdoor video databases. Combining the techniques into a preprocessing chain, we apply them to the NIST Gait Challenge database. This produces visually good extracted data suitable for biometric use. Analysis of the output by multiple gait biometrics yields encouraging recognition results, which approach those obtained from laboratory quality data.