Foreground Objects Segmentation in Videos with Improved Codebook Model

Abstract

Extraction of foreground objects in real-time is a significant topic for applications in computer vision. Most of the proposed techniques use background subtraction technique to detect moving or static foreground objects in the scene. Despite ongoing lots of research, the domain has not reached mature status and needs more advanced and improved solutions. In this proposed system, background subtraction is done by improved codebook model-based method to get segmented foreground objects. In background modeling, the L*a*b* color space is used instead of RGB color space. This method has been tested with standard datasets and the accuracy of segmentation results are also evaluated. The experimental results demonstrate that the proposed method perform well under difference background subtraction challenges such as dynamic background, shadow, illumination changes and bad weather.