This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: FOREGROUND SEGMENTATION IN SURVEILLANCE SCENES CONTAINING A DOOR Andrew Miller and Mubarak Shah Computer Vision Lab at University of Central Florida Orlando, Florida 32816 [email protected] / [email protected] ABSTRACT We propose a new method for performing accurate background sub- traction in scenes with a door, like a building entrance or a hall- way. This kind of scene is common in surveillance applications, yet the sporadic motion of a door causes problems for existing systems that falsely report the door as foreground. Our method models the scene’s appearance by storing a set of gaussian pixel distributions corresponding to a discrete sample of the door’s range of motion. All of the pixels in the image are dependent on the position of the door, so we use the joint probability for all of them to estimate the maximum-likelihood position of the door. We then perform back- ground subtraction using the specific appearance model indexed by our estimated position. We show that our algorithm accurately seg- ments the foreground region in several actual indoor and outdoor surveillance settings. 1. INTRODUCTION Some of the most popular surveillance settings are centered around a door, such as a building entrance or an office hallway. Unfortu- nately, doors present difficult problems to most surveillance systems since they tend to violate basic assumptions about the nature of a background. Doors move relatively infrequently compared to cam- era noise or jittering clutter objects, and their appearance can change dynamically as they sweep out different angles, possibly reflecting a light source at the camera. When a person dynamically occludes a moving door, both the person and the door will be lumped together as a single foreground object. Therefore we propose a solution that exploits a different prop- erty of the background to distinguish it from the foreground. The position of the door is a parameter of the entire scene, so we can use the joint evidence from every pixel in the region to determine the position of the door. Once we recover the position of the door, we unambiguously know appearance of the background and can per- form background-subtraction just as easily as if the background were static. 2. RELATED WORK The goal of a background-subtraction approach is distinguish the background from the foreground by utilizing some discriminating characteristic between them. Most systems rely on the assumption that the foreground will usually appear different from the foreground and treat each pixel as an independent sensor. Pfinder was the original work in statistical background model- ing, using a single RGB gaussian for each pixel in the background . Pixels whose values differ substantially from the background model are marked as foreground. This system is simple and works effectively in scenes with a static background Stauffer and Grimson  have developed a Mixture of Gaus- sians (MoG) background model which has become very popular be- cause of its flexibility and stability in complicated scenes.cause of its flexibility and stability in complicated scenes....
View Full Document
- Spring '08
- Pixel, RGB color model, Door, Bayesian probability, The Doors, Prior probability