by Oliver Vellacott
Analytics can detect "suspicious movements" of people walking on the street; it can detect terrorists walking on a hillside more than 1,500 meters away. It can distinguish a criminal from a multitude of faces. These are some of the most widespread errors in relation to current analytics.
Has there ever been a technology like this, that promises so many things and then can't deliver on them?
No. The truth is that analytics is still practically in its infancy. The point is that you must be realistic with the user regarding the results that can be obtained. The fundamental problem is that humans do these types of tasks without realizing it; for example, we read license plates and recognize faces unconsciously; skills that took us years of learning during childhood, but that we now take for granted.
However, computers do not have even a minimum of visual intelligence. There are some video analytics features that computers can perform correctly, but it's often only possible by greatly limiting the application.
Current analytics
What is possible today? The identification of license plates has been done for a long time and its effectiveness is proven although it is not 100% accurate. Also, making a reliable face recognition is very difficult and cheating using costumes is very simple, and a good photograph of the subject is needed to work with a minimum of precision.
Today there are some basic analytical functions that can be carried out quite accurately. The simplest and most elementary form of analytics is motion detection, which is offered by various manufacturers, but very few systems have a false alarm rate low enough to be useful. A system that generates more than 20 false alarms per night is ineffective, as alarms will quickly begin to be ignored or the motion detection function will be disconnected, in order to prevent the operator from drowning in a sea of alarms.
From the basic detection of movement emerged the detection of congestions. When the density of people or cars reaches a certain point, an alarm goes off. The countercurrent motion function detects objects moving in an "anti-current" direction and is useful in certain applications, such as airport security checks. The virtual trigger cable feature is also an advancement of motion detection, as it sets off an alarm when someone or something crosses a line that has been drawn in the image. This is very useful in applications such as large areas with forbidden zones such as factories. In this way people can move quietly through the areas of free access, but if any person leaves these areas the alarms will sound.
Camera position, lens choice and lighting are absolutely critical to all of these analytics applications. Simply changing the position of the camera can improve the results of the analysis to a great extent. For example, in the detection of countercurrent movements the algorithm is much simpler if the camera is placed pointing downwards to focus on the area as if it were a plane since, if the camera sees an image with perspective, in which people cover each other, the process of tracking people becomes much more complicated and, once again, expectations can become unrealistic.
A busy market
There are literally hundreds of companies that offer analytics software. On average, one analytics provider per month contacts IndigoVision, as manufacturers of end-to-end IP video solutions, to ask us to integrate their product into our IP video management platform. Why so many? because any small software company can develop a set of video analytics, if you buy an image capturer and a powerful computer and design the software. The software is then sold separately, as a stand-alone system that is placed together with the main CCTV system. The video is then separated from the matrix and sent to the analytics system. This has the disadvantage that it is not truly integrated into the operation of the ccTV main system and therefore offers few benefits. Analog systems can't really use built-in analytics. DVRs (digital video recorders) can use it more, but in reality they remain "isolated", that is, like analog TV closed circuits they are not truly integrated.
Integrated analytics – the IP video solution
IP-based video management systems offer the ideal platform to fully integrate excellent analytics into the system and these would be a fundamental and integral part of its operation. The most outstanding IP video solutions are capable of executing analytics in two fundamental modes: live (to detect events as they occur) and after processing (to test various scenarios in the recorded images).
As is evident, the ideal place to place live analytics is the camera, since this is the only solution truly capable of modernizing and also does not consume bandwidth. Real-time central processing will eventually lose momentum, while each camera may have its specific processing. That is, a camera that has its own comprehensive analytics can monitor the activity of the image and transmit only specific events, for example, when a person advances in the opposite direction at the security control of an airport. This reduces unnecessary video traffic from the network, thus decreasing bandwidth needs, which is not possible to do with a traditional analog system.
The ideal place to place the analytics after processing is, of course, a central server, in order to be able to check the recorded video on several occasions according to different parameters. One of the biggest losses of operator time used to be the advance and rewind of the tape of the video devices, which improved with dvrs, but most are still basically an advanced or digital rewind. Analytics offers the possibility of continuing to transform this essential task, since possible events can be searched in large quantities of recordings and the operator only has to validate them. So computers do what they know how to do well, identify possible events and people do their part, verify those events.
Conclusion
It is essential to establish and define what is possible, so it is important not to believe all the ridiculous virtues and nonsense related to the results of analytics. Probably all this will be possible in 30 years, but today it is about making sure that what the possibilities are and carrying them out extremely well.
* Oliver Vellacott founded IndigoVision in 1994. He was previously a production manager with product experience for smart cameras. Oliver studied piano at the Guildhall School of Music before obtaining a Ba in Software Engineering (Imperial College, London) and a PhD in Electrical Engineering (University of Edinburgh).
Leave your comment