Foreground Detection Using Spatiotemporal Projection Kernels

Foreground detection is at the core of many video processing tasks. In this paper, we propose a novel video foreground detection method that exploits the statistics of 3D space-time patches. Efficient and accurate foreground detection relies to a large extent on reliable background modeling, where common and expected background changes are characterized. In this paper we characterize 3D space-time patches by means of the subspace they span. As the complexity of real-time systems prohibits performing this modeling directly on the raw pixel data, we propose a novel framework in which spatiotemporal data is sequentially reduced in two stages. The first stage reduces the data using a cascade of linear projections of 3D space-time patches onto a small set of 3D Walsh-Hadamard (WH) basis known for its energy compaction of natural images and videos. This stage is efficiently implemented using the Gray-Code filtering scheme [4] requiring only 2 operations per projection. In the second stage th...

• Type: Conference
• Year: 2012
• Where: cvpr
• Authors: Y. Moshe, H. Hel-Or, and Y. Hel-Or
• Keywords: Background Subtraction, Foreground/background Segmentation, Video Analysis