Illumination eigenspace and its application for robust video surveillance
Yasuyuki Matsushita Ko Nishino Katsushi Ikeuchi Masao Sakauchi
Effects of illumination, such as shadows and reflections, cause one of the most confusing effects on detection and tracking of moving objects. Especially, large-area shadows cast on moving objects by surrounding buildings and sudden weather changes are harmful for robust video surveillance. In this work we present a method to normalize those illumination effects, i.e. eliminating shadows and brightness changes of moving objects, as a preprocessing stage of video surveillance. We propose illumination eigenspace to model the variation of illumination, and utilize the generated model to directly normalize illumination components in input images. Finally, we confirm the usefulness of our method by evaluating errors of tracking results.
Publications
1. Y. Matsushita, K. Nishino, K. Ikeuchi and M. Sakauchi, ``Handling Shadow and Illumination for Video Surveillance'', in Proc. of First European Conf. on Color in Graphics, Image and Vision (CGIV '02) ,pp153-158, Apr. 2002.2. Y. Matsushita, K. Nishino, K. Ikeuchi and M. Sakauchi, ``Robust Tracking with Normalizing Illumination'', in Proc. of IPSJ SIG-ICII'02, 2002 (to be appeared)(in Japanese)
3. Y. Matsushita, K. Nishino, K. Ikeuchi and M. Sakauchi, ``Illumination eigenspace and its application for robust video surveillance'', in Proc. of IPSJ Symposium Series MIRU'02 (to appear) (in Japanese)
Left: obtained from image sequences from 10 days(10:00-14:00)
Right: obtained from 120 days(7:00-15:00) including sunny, cloudy and rainy day.
Fig 2. Left image shows the result over the sequence without our normalization
method,
and the right image shows the result with the normalization method applied.