Video analytics: Turning video into insight
Video cameras are quickly becoming ubiquitous in public spaces. Every time you drive through an intersection or into a parking garage, or walk into a bank, grocery store or shopping mall, there’s probably a video camera recording your activity. While video cameras and recorders are powerful technology providing tremendous advantages to law enforcement, corporations and even private citizens, the sheer volume of cameras makes it impossible to actively monitor all of this video.
There are simply not enough people or eyes available to monitor all the cameras all the time. Searching for video recordings that capture activity of interest, while not impossible, is a highly labor-intensive task that can be exacerbated by the average person’s 22-minute attention span when watching video.
Enter video analytics. Generally defined as the application of algorithms to detect activity on video, early video analytics was able to detect motion and the location of this motion on video feeds. With this capability, operators did not have to monitor cameras, and forensic investigators did not have to search recordings in which nothing was happening. While this approach was a significant improvement, end users quickly discovered that not everything that moves is interesting.
As a result, the term false positive—or false alert—quickly became a buzzword in the industry. You do not want to be told “something’s moving on the front entrance camera” every time it happens. Video analytics had become the The Boy Who Cried Wolf.
Researchers and experts then went back to the drawing board, and a new generation of video analytics was born. Modern analytics tells us what is happening on a camera in much greater detail. Size, color, track, speed, object type and more can now be determined by analytics. Instead of something’s moving you now get a red car is moving eastbound on 33rd Street or a bald man with eyeglasses wearing a red shirt is walking down the hallway.
Armed with this new level of information, alerts and forensic searches can be much more precise, and false alerts are reduced to a level where people can start paying real attention to alerts. Video analytics is now providing end users with situational awareness instead of just awareness of pure activity.
Emerging technology such as convolutional neural networks—aka deep learning—and facial recognition and vehicle make and model recognition are improving both precision and accuracy in the field, paving the way for advanced capabilities and new use cases. This technology is expected to add significant value in the industry, but already a vast amount of untapped data exists that can be used to gain new insights.
Enter big data analytics. We’ve all heard the cliché, “a picture is worth a thousand words,” and video typically produces 15 to 30 frames/second. Without video analytics, these frames were just pictures; with video analytics, we now have structured data that can be rich fodder for big data analytics. A single car passing a camera is probably uninteresting unless it was the getaway car in a bank robbery. A thousand cars passing the same camera at different times and the knowledge of where and when these cars passed by can give us an understanding of traffic patterns that can be used to reduce congestion. Retailers can better understand customer behavior, and banks and airports can understand queue waiting times. The possibilities are virtually endless.
Someday, in the not too distant future, if you’re standing in a line at your local grocery store and a new cashier magically appears to open a new checkout line, you may have a combination of video analytics and big data analytics to thank for your reduced wait time. For now, find out more about how IBM Intelligent Video Analytics and big data analytics are changing the way we see and understand video.