Study on Statistical Background Modeling for Visual Surveillance

• Main Authors: Cheng-Chun Chung, 鍾承君
• Other Authors: Ming-Yang Cheng
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/77158613559363237808

碩士 === 國立成功大學 === 電機工程學系碩博士班 === 96 === The main purpose of a visual surveillance system is to detect suspicious objects or erratic changes in the environment. In a visual surveillance system, object tracking and object classification rely on the accuracy of motion detection. Therefore, it is essential to quickly and accurately identify a moving object in an intricate environment. Background subtraction is commonly used in motion detection. The system must update the alteration of the environment into the background model quickly and accurately for best performance. A suitable background model for changing environments was developed in this thesis. In general, there are two methods of constructing a statistical background model. One is the parametric method, which includes the Gaussian mixture model and the spatial distribution of Gaussian. The other is the nonparametric method, which includes the kernel density estimation and the k-nearest neighbors method. In this thesis, several experiments have been conducted to provide a performance comparison between the Gaussian mixture model and kernel density estimation.