Few terms generate more interest in the world of video surveillance and perimeter security than Artificial Intelligence, and few are as misinterpreted. Artificial Intelligence and machine learning have the power to truly identify or classify objects and virtually eliminate false alarms for camera-triggered alarms. This technology is not only difficult to do well but requires a massive amount of computing power, which most platforms cannot support.
To understand the hype — and confusion — surrounding Artificial Intelligence in video surveillance, it’s important to understand the differences between motion detection and the classification of objects.
Motion Detection
The most common method of detecting a perimeter threat is motion detection. Security cameras set at regular intervals monitor an area and record sequential images of the light and dark pixels in the scene. If the pixel pattern changes between frames, the camera’s software determines that something has moved and sends an alert.
The problem with pure motion detection is that there are many things that move within a scene that are not a threat, such as bugs, wildlife, organic movement, wind, and weather. Even a distant flash of lightning or the headlights from a car can trigger an alert, resulting in a high number of false alarms.
Evolon Edge and Evolon Enterprise video surveillance solutions also detect motion at the camera level but apply patented algorithms to filter out all the environmental video noise and other distractions. Using advanced video analytics, Evolon Edge detects an object as moving pixel clusters up to one mile away and instantaneously filters out all other anomalies—even in harsh weather conditions.
Object Detection
Object detection is a technology that uses algorithms to locate objects within a video stream. Typically identifying interest with “bounding boxes” drawn around detected objects.
Object detection is best suited for situations requiring alerts on objects of interest that are triggered in milliseconds and work regardless of the number of objects in a scene. In detecting objects, Evolon’s technology again uses algorithms and analytics to filter out environmental video noise, only targeting “objects of interest” and not the endless anomalies throughout the scene.
It’s important to note that object detection has its place in finding movement within a scene, as long as the technology ignores the typical anomalies. As a built-in camera software-driven solution, Evolon Edge has perfected this technology with deployments worldwide for commercial and government facilities. With Evolon Enterprise, Evolon expands its universe of supported cameras by being able to detect using virtually any IP, HD analog, and analog camera.
Object Classification
Object classification is the holy grail of video surveillance. By leveraging AI and Deep Learning to recognize a person and/or vehicle as a threat-based upon pre-determined criteria and set threshold levels, object classification greatly reduces false alerts. Depending on the solution, it may not be as fast as object detection due to the complexities of classification algorithms. A huge benefit of Deep Learning is its ability to improve over time, as the system learns (or is trained) by analyzing past scenes to determine what is and is not a threat.
Evolon’s object classification solution is Evolon Verify, a complete AI-based video analytics and Deep Learning system native to Immix central monitoring stations that quickly and proactively validates all inbound video events before they arrive in the central station or command center, eliminating up to 90% of false alarms. Verify is designed for easy implementation and use, enabling customers to cost-effectively improve operator engagement by allowing personnel to respond to genuine threats quicker with fewer resources.
