The Evolution of Video Forensics: What’s Next?
Almost every organization today uses CCTV in one form or another. Some of the most important and even tragic events of our time have been captured on video. Video recordings are rightly the cornerstone of many successful post incident investigations. But in situations where time is of the essence, getting to the crucial segments of video that reveal clues cannot wait.
Previous generations of forensic video solutions have attempted to address this challenge, but with limited results.
Let’s say your security operations center received a report that a person dressed in black was seen entering your large facility complex carrying a suspicious item. An earlier generation of forensic video solution called pixel-based motion detection offered the following approach: Search through video from each camera that may have seen the suspicious person Highlight a section in the field of view and ask the system: ‘Tell me all the times that something moved in this area during the past hour’ Repeat this process manually for each camera until you find your guy
The search could turn up thousands of hits, depending on the flow of people and the number of cameras involved. Imagine having to sift through thousands of images. It would be like searching for the proverbial needle in the haystack. Eventually you would find the suspect, export a snapshot image from the video and share it around hoping that someone would recognize him. But of course, by then, it might be too late to have stopped him from acting on any mal-intent.
The next generation of analytics, content-based forensics, improved on this somewhat. Instead of simply detecting motion, it could to some degree distinguish the source of the motion (e.g. a person vs. a small animal, especially useful for outside perimeters). It could also distinguish primary sources of color (e.g. the person was dressed predominantly in black). This had the effect of reducing the number of video images that you’d have to manually scan, but it still required you to repeat the process for each camera, each with its own unique perspective-distorting camera angle, lighting and white balance which throws off the color. And because of the time it took to do that, content-based forensics was still considered mainly a post-event investigation tool.
In both of the above cases, it would be pretty unlikely that you could find your suspect fast enough to stop something bad from happening. It is clear that in order to be useful for real-time forensics, video analytics must evolve into something more. So what’s next? Find out in our next blog post.