By Kapil Bardeja, CEO & Co Founder, Vehant Technologies
In this era, surveillance camera has become a common need of everyday living. It plays a vital role in numerous fields like news, sports, security surveillance, education, etc. Security surveillance cameras are installed at almost all public places, homes, airports, banks, shops, etc. and data is generated 24/7 with these cameras. How much ever complicated it may be, this data helps in retrieving useful and interesting information.
However, the process for this retrieval is time consuming. The users need to watch the video for extracting the useful information. Due to temporal nature of the video, it is much more tedious and complicated to extract meaningful information from them as compared to extraction from image content. Owing to the large size of the high dimensional data captured in the form of videos, it is further difficult to manipulate and manage the data and the recognition of anomalous or suspicious/ abnormal behaviour is quite difficult and time-consuming too. Subsequently, it demands high storage space, processing power and human intervention.
Video summarization is a ground breaking concept in the field of data extraction and manipulation. Video summarization is defined by researchers as the generation of a summary of extensive video content by detecting and presenting the material to potential users which is most informative and contains interesting information. As the name suggests, this technology extracts and summarizes the video content into a shorter version in order to present the informative and useful parts in a meaningful way thus helping in offering valuable insights and quick decision making.
It is important to organize the data so that it can be used for retrieval and analysis. Video summarization focuses on organizing the metadata in order to make it usable when retrieved. The technique deals with how to use the representation structure to provide the viewers top-down access using the summarization. The technique of video summarization has reduced the amount of human intervention and effort required to analyse a trail of events. Surveillance involves unscripted content and video summarization in unscripted content requires different extraction framework that only captures remarkable events that constitute the summary. One of the most common ways to accomplish it is to segment the input video and then detect objects such as Person, Vehicle etc. or detect motion in the video.
One of the methods used for effective video summarization is through background subtraction. Events are detected at run time for matching of a particular requirement. During retrieval, the database filters the matching criteria and the filtered segments are merged with a background for analysis. The background is dynamically updated with time. For instance, if 10 people are walking through a surveillance zone at different times, their video segments are filtered and merged with the background such that they seem to walk in the video at the same time. Such a summarized video would be helpful in observing and analyzing the events related to each of object through single summarized video. The technology takes care of the collision that may occur in the movements of the objects/people in the summarized video. However, such summarization is possible if the surveillance zone is not overcrowded. Although there are players who claim that compression of a 24-hour video is possible into a 15-minute video with this technique, but the results are not effective in a congested area. At traffic junctions or other surveillance areas, the technique may not be efficient in Indian scenario with large population and crowd congestion.
The number of video surveillance cameras has increased by a large amount in recent years. There is therefore a need to process the captured videos such that human operators can quickly review the activities recorded by a camera over a long period of time. Surveillance video data can be of great use. Besides the owner, these videos are often used by Police officials, security personnel and by Criminal Investigation
Department as a part of investigation as these videos can serve as evidence. Summarization is an efficient technique which produces a short skim of these surveillance videos by retaining the most prominent scenes and discarding the least important information.