Summary: | 碩士 === 淡江大學 === 電機工程學系碩士班 === 98 === Since there are more than one million surveillance cameras in public space of a city, we have faced the age of surveillance system. The most painful task is retrieving those volumes of video and trying to find some available capacity to store them in. The conventional way to shrink the video is compression, however this method causes some disadvantages, such as saving miserable space with lossless compression, and poor evidence with lossy compression. The conventional way to retrieve video is abstraction and index, but it is incompatible with video archiving because it needs more capacity.
This thesis presented a solution that can have it in both ways. The approach based on Video Synopsis uses the LL sub-band which is obtained from Symmetric Mask-based Discrete Wavelet Transform to down-sample the resolution of frames, so that the computing coast of the following step in this proposed flow could be reduced easily, and the high frequency noise could be decreased in the meanwhile. Then some algorithms are proposed to extract foreground objects. The key part is that the thesis proposed an algorithm to calculate the spatial information for showing objects simultaneously which originally occurred at different times, just like the stroboscopic effect.
According to the experimental result, this solution has some merits: good efficiency to handle real-time video archiving with 23 FPS in VGA resolution, lossless recording, only need low storage capacity with about below 15% space of original video, the duration of the generated video is about 15% run-time of original one, and suitable for outdoor environment.
|