Gaussian-Based Codebook Model for Video Background Subtraction

Abstract

As an effective method of background subtraction, codebook model suffers from unacceptable false negative detection rate in many situations due to its quantization criterion. In this paper, we propose an improved codebook model to solve this problem. Instead of using the original quantization criterion, we quantize the temporal series of the observations at a given pixel into codewords based on the Gaussian distribution assumption. We have performed this approach in our surveillance system for outdoor scenes and achieved excellent detection results.