User misunderstandings of how video analytics really works has resulted in unrealistic expectations in perimeter detection.
Despite major improvements to VCA algorithms over the years, VCA is still by no means perfect. “The most common misunderstanding regarding VCA and perimeter protection has to do with the fact that most VCA solutions that are built into IP cameras are simply not accurate enough for real world perimeter protection applications,” explained Zvika Ashani, CTO of Agent Video Intelligence (Agent Vi). “There is a very wide range of capabilities and accuracy levels available when talking about VCA. Basic capabilities are available today within many IP cameras which allow defining VCA rules and generating detections but these provide high false alarm rates (FAR) and low probability of detection (POD). A high-end solution provides much better performance in terms of FAR and POD.”
“The most common misunderstanding regarding VCA and perimeter protection has to do with the fact that most VCA solutions that are built into IP cameras are simply not accurate enough for real world perimeter protection applications.”
- Zvika Ashani, CTO, Agent Video Intelligence
Ron Grinfeld, Director of Global Vertical Marketing at DVtel added some of the common misunderstandings and “over-expectations” clients have from video analytics. “Clients believe that perimeter security solutions can achieve 100-percent probability of detection and zero false alarms, that multiple targets can always be tracked individually across different cameras even when crossing each other and/or walking in a group or that visual identification of facial details, license plate numbers, and small objects is guaranteed across the achievable detection distances. Another false belief is that VCA for perimeter security can detect specific scenarios such as a person cutting the fence, holding a weapon, fighting, and so on.”
Jürgen Konetschnig, COO of Austria-based Cogvis pointed out, “Distinguishing cars in size is easier if they cross the line of sight of a camera than if they move towards the camera. However, distinguishing car models, car manufacturers, or detecting persons that carry weapons in real-time using a classical perimeter setup in a reliable way is — up to now — still only possible in Hollywood movies.”
“Over time, improvements in sensors and algorithms extend the range of situations where things can be expected to ‘just work,’ but in the meantime there is an awkward middle ground where things will work in one context and not work in another intuitively similar context.”
- Mike Jamieson, Computer Vision Developer, Aimetis
Over-expectations are a result of a bias humans have comparing VCA with the performance of the human eye, according to Mike Jamieson, Computer Vision Developer at Aimetis. “It is very common for people to discount the complexity of what their visual system is doing while they are taking in a scene. Human vision feels effortless but it uses up a large portion of the brain,” he said. “VCA systems can sometimes be thrown off by changes in a scene that a person suppresses unconsciously. Related to that, and harder to deal with, is the fact that analytic accuracy is affected by so many factors. We’d like to say that if the target is at least X-pixels high, they will always be tracked. In an actual installation, contrast, lighting, weather, image quality, geometry, target motion, and background motion all affect accuracy. A fairly small change in the setting can make the difference between a reliable and unreliable installation. Over time, improvements in sensors and algorithms extend the range of situations where things can be expected to ‘just work,’ but in the meantime there is an awkward middle ground where things will work in one context and not work in another intuitively similar context.”
Besides over-expectations, another problem is users’ commitment to make sure that all the installation recommendations and environmental conditions are met.
“In our experience, perimeter protection is one of the modules where users get most optimal understanding of what to expect from this exact VCA module, so misunderstanding comes only in cases when user wants to save on interval of cameras that allow this module to cover the required area. VCA is not something that can think and/or guess. It works with clear instructions and unable to improvise, therefore it needs to receive exact data, as per the requirements. That is why all the recommendations on light conditions, angle of camera positioning towards expected object, and other details, should be carefully followed,” said Aleksandr Jesikov, Account Manager at Luxriot Europe.