Background Modeling and Subtraction Reasearch Page

We have implemented a robust real-time background subtraction algorithm based on codebook approach. Our method is more forgiving for camera and image noise than the methods utilizing a single modality for each background pixel. The algorithm adapts background model to the changing backgrounds and is tolerant to illumination changes and background clutter. Special emphasis is made on dealing with the compression artifacts produced by MPEG-like compression algorithms.

The results presented below are from two datasets: OTCBVS shot outdoors with one RGB and one thermal camera and CAVIAR shot indoor with RGB camera

 

This is the novel model of the scene background where each pixel is represented as a multi-modal distribution. The number of modalities is dynamically changing for both color and thermal input. We demonstrate how to eliminate the influence of shadows with this type of fusion. Based on this background model we introduce a pedestrian tracker designed as a particle filter. It includes a number of informed reversible transformations to sample the model probability space in order to maximize our model posterior probability.