We live in a world that is constantly in motion.
Even in secure and enclosed places where we’d expect a lack of motion, motion is still often happening - at least in the eyes of a security camera.
Take your home for example, when the doors are locked nobody is home. Your curtains might be blowing in the wind. The light you left on might be flickering. Shadows cast by the sun are dancing about. Your puppy is playfully running around the house and being up to no good. Tiny insects might also be buzzing about.
To a camera, these are all examples of motion. If the camera was situated outside of the home, it is exposed to even more motion.
Traditionally, motion detection has been used to determine if a segment of footage from a security camera is noteworthy or just noise. The assumption that there should be no motion when no one is around was quickly discovered to be false. To combat this, other techniques were employed in attempt to improve the results of motion detection. Some examples include:
Approach: Filter out more noise by only treating “large movements” as motion.
Using a motion sensitivity setting still poses many problems. Firstly, the end-user would have to preselect the sensitivity threshold, often either a “one to five” or “one to ten” scale, or a “low/medium/high” setting. It is impossible to know which is the right threshold to use immediately so a trial-and-error approach is required. Once a threshold is set, the user would still have to make a big trade-off: set a low sensitivity and get too many false negatives, or set a high sensitivity and get too many false positives.
Using “large movements” as a filter for interesting events also poses other problems. A person that is farther away from the camera lens will obviously be smaller in the perspective of the frame. Movement caused by this person would probably be interesting, but would not be triggered by motion detection. An object that is moving slowly might also be missed by motion detection. Conversely, lots a large shadow moving in frame would be easily picked up by motion detection.
Approach: Filter out more noise by ignoring motion that occurs in certain parts of the frame.
Detection zones can be useful for situations where there are certain portions of the frame that are you absolutely not interested in and there is no other feasible way to angle the camera. A camera surveying a driveway might invariably also cover the public footpath or a camera surveying a backyard from a high vantage point might also include portions of neighbouring backyards.
There are however limitations to the feasibility of detection zones. A zone that is used to cut out a noisy section of the background will also cut out legitimate movement in that area. Even well defined zones will be subject to motion sensitivity issues described above.
Solution: use intelligence
The only way to properly determine if an event is important or merely noise is to intelligently analyse the footage in order to make a decision. The naive approach would be to have a person monitor the camera feed continuously. This traditional approach of having a security guard in a control room is not only expensive but relies on the alertness of the guard.
Image Intelligence provides best in class person detection using the power of Artificial Intelligence. Our deep neural networks are specifically trained to provide intelligent analysis of security footage.
For situations where the utmost accuracy is required, our unique system pairs AI with human operators to produce unparalleled accuracy.