Time-saving preview of surveillance videos
City View
Rather than stare at boring surveillance videos, in which nothing happens 90 percent of the time, Mike Schilli tries the OpenCV image recognition software, which automatically extracts the most exciting action sequences.
In my home city of San Francisco, hardly a day goes by without hundreds of cars, garages, and homes being broken into. Instead of getting upset about this, I tend not to keep anything of value in easily accessible places, and I have also installed security cameras so that I can peruse the video footage of thieves at work for my personal amusement.
Wireless, Even
Of course, installing a security camera is no easy task, because you need to install a cable and route it to the monitor. Although the camera itself often communicates wirelessly with the control panel, it still needs a power supply, and a power supply is not easy to come by in hotspots such as the underground parking lot or the stairwell.
Recently, a company called Arlo started to sell child-fist-sized, battery-powered cameras [1], which amateur detectives can simply hang up using a magnet (Figure 1). These pocket wonders wirelessly send recorded videos to a hub at a distance of up to about 100 feet, which in turn sends the data via the Internet to a server, from which a variety of smartphone apps or a website transfers the data to the user's screen on request.
Conservative Operation
To reduce the load on the four lithium batteries powering the camera so they can last for about one month, the camera is allowed to wake up about a half a dozen times a day if it detects motion in its vicinity and then transfer a one-minute video. You then download the movie from Arlo's website (Figure 2) to see thieves, say, dragging your new bike out of the garage. You typically only see motion at the beginning of a surveillance video, the rest of the one-minute footage normally shows nothing but motionless background (Figure 3).
Fast Forward to the Action
"Cut to the chase," people say when someone fails to come to the point. This probably refers to action films, where the audience does not want to see long-winded, suspense-building scenes but prefers to fast-forward to the car pursuit at the climax of the Hollywood production.
In this sense, it would be nice for the software to scour the video for frames in which a subject actually moves through the scene, so that the viewer knows whether the videos are worth watching and, if so, the location to which to fast forward in the video.
Buy this article as PDF
(incl. VAT)