A Fast Approach for Background Subtraction on Spatial and DCT Domain

碩士 === 元智大學 === 資訊工程學系 === 98 === Because of the hardware limitation of embedded system, we present a background subtraction algorithm that can be provided with high speed, low complexity, less space and no ghost effect. First, we only need to decompress the video to DCT domain, then build the refer...

Full description

Bibliographic Details
Main Authors: Yen-Kuan Lin, 林彥寬
Other Authors: 林啟芳
Format: Others
Language:zh-TW
Published: 2010
Online Access:http://ndltd.ncl.edu.tw/handle/91934130126703601313
Description
Summary:碩士 === 元智大學 === 資訊工程學系 === 98 === Because of the hardware limitation of embedded system, we present a background subtraction algorithm that can be provided with high speed, low complexity, less space and no ghost effect. First, we only need to decompress the video to DCT domain, then build the reference background with running moving average. With reference background, we use counters to emulate the method of median. Checking every DC value of all DCT blocks. If the difference is bigger than the threshold value, we increase the counter of the block. When the counter of block is bigger than a half of sliding window, we replace the DCT block of reference background with the current DCT block, and then set the counter to zero. Repeat these steps, we can not only have the advantages of running moving average, but also achieve to build reference background without ghost effect. We propose the method for fast background subtraction on embedded system that could apply to intelligent analysis.