In the World of Watson, Intelligent Video Management Drives the Value of Video
November 1, 2017
Oh, for the good old days. Used to be one could read a memo, or a magazine piece, and add insight with the simple application of a sticky note and some archaic calligraphy that I hear is no longer taught in schools. Readers instantly benefitted from this highly relevant wisdom.
As the prevalence of content distributed on physical media has declined, many forms of digital communication include the equivalent of the ubiquitous sticky note. In fact, some of those annotations even look like sticky notes, complete with fonts that simulate bad handwriting.
Such annotation is great for static pieces of content like documents, emails, and photographs. But what about the most popular form of digital communication today – video?
Sure, a comment can be added to a video file. However, such a comment may not make any sense, at least for long. That’s because video is a dynamic visual medium. The information conveyed actually lives in the fourth dimension – time. Comments make most sense when they are aligned with the portion of the video to which they apply. And so time-based metadata, and the need for intelligent video management, was born.
Time-based metadata, whether automatically generated by an analytics system or generated by a human, is stored completely in context with the video file. Repositories that support time-based metadata are able to retrieve the metadata and position the video to the exact point where the metadata makes sense.
The simplest example is audio transcription. A search of a video repository may return a number of results where a particular word or phrase exists in the transcribed audio track of a video file. For longer videos, it’s absolutely necessary for the player infrastructure to position the video to the location containing the word or phrase. The emerging world of intelligent video analytics adds even more need for time-based metadata. For example, a facial recognition application may recognize a hundred different individuals over the course of an hour-long video. The simple fact that a particular person appears somewhere in the video may be germane in some situations, but knowing, and seeing, exactly where he appears on the screen truly is the relevant information.
Consider the video an unmanned drone might record as part of a regular pipeline inspection. The entire inspection could involve days of recording. Reviewing all of that footage would be impossible and defeats the purpose of using an unmanned system in the first place. However, video analytics systems can and will identify scenes of interest within that video. Again, by using appropriate storage and retrieval mechanisms, completion of the examination is a simple process.
Submitted by Randy Dufault
Director of Solution Development
