A Background Estimation Algorithm Based on Parallelization Concept

• Main Authors: LIN, ZHAO-HAN, 林昭翰
• Other Authors: CHEN, YUNG-YAO
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/5br3yj

碩士 === 國立臺北科技大學 === 自動化科技研究所 === 107 === Video background initialization technique has been widely used in our life. For example, a clear background image is required for finding the moving objects in video surveillance or video objects tracking. In this paper, we proposed a background estimation algorithm based on parallelization concept method. First, the difference between successive frames are computed and with the edge of each frame and use connectivity to construct binarized images that are used to determine backgrounds pixels in each frame. Next, multi-directional pixel information and the constructed binarized images are jointly utilized to search the candidate background pixels based on priors. We then fuse a clean and complete background image from those candidate pixels quickly. Experimental results show that the proposed technique is able to construct a perfect background image quickly.