A Robust Texture-Based Background Subtraction for Moving Object Detection in Video Sequences

• Main Authors: Chia-Yen Liu, 劉佳諺
• Other Authors: C.S. Ouyang
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/79968480937959420754

碩士 === 義守大學 === 資訊工程學系碩士班 === 95 === In this thesis, we propose a robust approach of texture-based background subtraction for moving object detection in video sequences. There are three phases in our approach: feature extraction, foreground detection, and background model updating. Firstly, we propose a method of local fuzzy patterns to describe the local texture features of pixels in video frames. Then, a fuzzy histogram is calculated for describing the local texture distribution of each pixel and its corresponding neighboring pixels. To classify each pixel into background or foreground, we calculate the dissimilarities between the fuzzy histograms of the pixel and its corresponding background models with the method of Euclidean distance. Finally, we use the dissimilarities to decide whether the background models of the pixel should be updated. Compared with the approach proposed by M. Heikkil? et al.[11], experimental results show that our approach has better performances in the conditions of shadow and noise, and produces more accurate detection results of moving objects.