Surveillance Video Summarization based on Target Object Detection

The recent trend increases the use of surveillance cameras in many of the private and public premises, which causes the number of surveillance videos to grow exponentially. The information gained from these surveillance videos not only helps the owner of the property, but also helps in crime investigations for police and security officials. Though there are several applications of such videos, yet their storage, management and retrieval processes are still challenging. Hence, it is important to develop an efficient technique to describe a long video into a shorter video with semantic information by eliminating the redundant and unimportant frames. This technique makes the larger video to shrink in length for efficient storage and also helps the users to attain a complete knowledge of the video by only watching the shorter video, without spending more time in watching the original longer video. To achieve this objective, this paper proposes a video summarization technique for summarizing the surveillance videos by extracting the target object using YOLO then discarding the remaining frames and finally combining the extracted key frames into a single video. This method first detects the target object related in the original video frames and then eliminates the remaining frames that are irrelevant without prominent objects, resulting in video with only the key frames which are into the interest of the user, finally those frames are combined to form a summarized video.