Object-Based Video Archive Summarization
Habiba Yasser Adel, Ramez M. Elmasry, Mohammed A.‐M. Salem · 2023
Nowadays, footage from surveillance cameras has a huge amount of data and is very long and exhausting to watch. This is where our problem lies, as the need to efficiently explore these videos quickly is rising. So video summarization aids people in exploring videos efficiently by capturing the most important frames in a video. In this paper, we provide a system that summarizes long video footage into short videos displaying the object of interest to the user. The results are very promising, as the system shows the original video where the objects are being tracked, and the customized video shows the summarized movement where the object of interest is the only thing in the video tracked in all the frames of the original video and last is the summarized video which is after extracting the keyframes from the generated customized video, and this video only shows the essential parts of the original video related to this object. There is a percentage of error due to the usage of the Yolo model and the tracking algorithm which are not very accurate as every system has a percentage of error, for static object videos, the duration decreased till it reached 0 seconds, and the compression ratio is 0.005. For nonstatic object videos, the duration on average decreased the half and the size decreased by more than half of the original size. Regarding the summarized video, the duration, size, and compression ratio compared to the customized video did not change much but there was a huge difference if compared to the original video.