Background image generation by preserving lighting condition of outdoor scenes

In the Sensing Web project, we have proposed the concept of “Henshin” camera, which outputs only privacy-protected information in order to open and share the information on the Internet. The information should not contain appearances of people in the captured scene but just their positions and the background image of the scene. In this paper, a novel method is proposed to generate the verisimilar background image which well expresses the weather and the lighting condition of the scene. This method collects a huge number of images by super long term surveillance, classifies them according to their time, and applies the eigenspace method so as to reproduce the background image without any appearances of the people in it. We experimentally evaluate the results of this approach using data of several surveillance cameras.