Object Recognition and Content Abstraction of Surveillance Video

Abstract

Video content abstraction is a useful tool for the object and event acquisition in long-term video. To solve this problem, the video processing is divided into three phases including foreground extraction, key frame selection, and object recognition module design. Firstly, foreground pixels are extracted from the original video. Then the semantic category information of foreground objects are also extracted. Finally, concerned video content are automatically synthesized into the produced video. Experiments show the practical effect of video abstraction in different public datasets.