Detection vs. Classification in Video Surveillance
Few terms generate more interest in the world of video surveillance and perimeter security than Artificial Intelligence, and few are as misinterpreted. AI 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 AI in video surveillance, it’s important to understand the differences between motion detection and the classification of objects.
In the scene on the left, movement on the street has been masked to avoid false alarms, while the movement of the person has been detected as motion. In the scene on the right, Evolon’s Verify uses AI algorithms and video analytics to validate the motion as a person and sends an alert to a central station or command center.
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 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 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 handles object classification through two innovative solutions. The first 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.
Our second solution is Evolon Enterprise 2.0, a groundbreaking solution that fuses detection and classification on a near-edge appliance with robust artificial intelligence (AI) capabilities built-in. Evolon Enterprise 2.0 helps lower total cost of ownership (TCO) because organizations can effortlessly add advanced AI and analytics to any legacy analog, HD, or IP camera. And like Evolon Verify, Evolon Enterprise 2.0 ensures alerting on only valid threats, with clients reporting up to 98% fewer false alarms through better detection and analytics. Today, Evolon Enterprise 2.0 is the most accurate and affordable on-premise video event verification system available.
The Evolon Advantage
Founded by former engineers and scientists from the world-renowned Los Alamos National Laboratory, Evolon’s award-winning technology has been evaluated by installations requiring the highest levels of perimeter security and met or exceeded the most rigorous standards of intrusion detection.
From the Edge to the Cloud, Evolon’s solutions are fully scalable for commercial and government organizations that require real-time situational intelligence to enhance security effectiveness. To achieve the best possible results and eliminate virtually all false alarms, combine object detection and object classification by deploying Edge or Enterprise along the perimeter while processing events through Verify at the command center, allowing your security personnel to avoid time-consuming distractions and focus exclusively on real events.