In our continuing series for reducing perimeter security nuisance alarms we’ve addressed the benefit of image processing and the value of electronic image stabilization. Another requirement for accurate perimeter intrusion detection is for an automated camera to “see” security violations against a background of movement and clutter. Practically speaking, this means the camera must be smart enough to ignore movement from “unimportant” objects. Such movement is also known as non-salient motion.

By definition, non-salient motion comes from objects that always return to the same position within a brief period of time, like branches, trees, foliage, or reflections in water. Alternatively, salient or “important” motion comes from objects that traverse the camera’s field of view, like a person or vehicle.

From a security perspective, salient motion is what matters. It’s activity that likely represents a legitimate intruder entering a secure area which requires investigation. On the other hand, non-salient motion is normal behavior in the outdoors, and must be filtered by the camera or the nuisance alarm rate will reach a level so high they will likely be ignored.

For indoor surveillance applications, such distinctions are irrelevant because background objects are not moving. In the dynamic outdoors the differences between these two types of motion is more important. Think of a typical outdoor scene: a large tree will sway in the breeze within the same general location. The same thing happens with leaves, bushes or reflections in water. Obviously, you want such movement to be ignored, or risk wasting your security efforts chasing after phantom issues.

Filtering such movement is best accomplished with sufficient on-board processing and video memory buffers for making the right determination. This is no easy task, especially considering the amount of data that a camera needs to analyze over a large outdoor scene spanning hundreds of meters. Importantly, when we refer to “video memory buffers” we’re not talking about on-board video storage for future retrieval. In this case, we’re talking about very high-speed memory that’s used for real-time scene analysis.

For example, a large tree takes a period of time to move in one direction and then just as long to return. Cameras that lack sufficient memory will only see the tree moving one way and trigger an alert. On the other hand, if the camera can analyze and store scene information over a longer period of time it can conclude that such background movement returns to the same area and is an “unimportant” object to be ignored.

You can see a real-world example of an intelligent video surveillance camera accurately determining salient from non-salient motion in the following video. Notice how the moving brush and foliage on the left is safely ignored, along with the heat waves coming off the vehicle, while the human targets — which are actually smaller than some of the background objects that are ignored — are accurately detected and alarmed. In fact, this video also shows how intelligent video cameras can “see” what the human eye may have missed.

Without sufficient on-board processing and video analysis memory to filter non-salient motion, such an environment would most likely be overwhelmed with nuisance alerts.

Subscribe to the SightLogix blog and keep updated on this series and other articles related to outdoor security.

 Print |  Email