An Approach to Real-Time Region Detection Algorithm Using Background Modeling and Covariance Descriptor

A region detection algorithm using robust image descriptor that abstracts objects from moving picture using Gaussian mixture background modeling on a moving object and covariance matrices are formatted using region variance magnitudes is presented.

Abstract

In this paper, we present a region detection algorithm using robust image descriptor. It abstracts objects from moving picture using Gaussian mixture background modeling on a moving object. And feature image is comprised of spatial properties and statistical properties on abstracted objects, and covariance matrices are formatted using region variance magnitudes. By using it to detection region, this paper puts a proposal for outstanding real-time region detection algorithm to diverse situations with strong noise, rotation and illumination. To estimate performance evaluation of proposed algorithm, we conduct an experiment on region detection in diverse traffic circumstance moving picture. And then we get a great performance result on interested region detection.