Recent years have seen a massive development in different kinds of video analysis algorithms which have been marketed with lots of optimistic promises. Many have been attempts to exploit algorithms created in a lab or university or research institute, by out-sourcing their development into products for use in video security systems. What looked promising in the lab, was often of only limited use in practice, particularly outdoors. Many suppliers and users have learned to their cost that marketing claims and presentations which raise expectations are not always met in reality.
Most video security systems used today to verify critical situations or alarms need a trigger, a message to the system and the user that something undesirable is happening. This information can come from a third-party system or from the system itself analysing the video streams from cameras.
In video analysis (also known as video content analysis, VCA) there are two different basic processes: motion detection and pattern recognition. Motion detection requires a constant stream of video data to analyse. Depending on the algorithm this can range from very simple (activity detection: ‘something moved!’) to really complex (Loitering: ‘There is someone in the scene who has been moving aimlessly back and forth for xx minutes’). In principle, pattern recognition is also possible with just one picture or photo. In the video security it is used to refine results from motion detection analysis by recognising a sign or text (eg, ANPR) or by comparing pictures with a predetermined pattern (eg facial recognition).
Examples of usage
Nowadays many different algorithms are offered for all sorts of applications, for security as well as business purposes. Nowadays it is not uncommon to find video systems in combined applications. Here are some examples:
Sabotage protection: monitors the contrast levels of the camera (for alerting to cloaking or lighting failure) and the viewing angle (for alerting to tampering).
Perimeter protection: detects moving objects or people in a so-called ‘sterile zone’ within the site (usually secured with a fence or wall).
Object identification: differentiates between people, vehicles and other objects, eg, for monitoring car park entrances (entry only permitted for cars, no pedestrians or cyclists), for toll payment at toll plazas etc.
Direction recognition: analyses the direction of movement of an object (eg, for detecting wrong-way drivers or for generating alarms when rail tracks are crossed in a station).
Detection of abandoned objects eg, for detecting unattended luggage (which could possibly contain explosives) in foyers, corridors or halls, for monitoring the access to evacuation routes and for recognising long-term parking violations in short-term zones (eg, loading and unloading at airports).
Detection of missing objects eg, for securing fine art exhibits or valuable goods shipments in logistic centres.
Number plate recognition: automatic reading of vehicle number plates eg, for access control, for collecting toll charges or for the statistical recording of visitors to an establishment (eg, in a zoo or shopping centre car park).
Facial recognition: automatic comparison with previously recorded photo in connection with access control systems or also with a database of known faces (eg, for identifying people who are banned from the establishment or VIP guests in a casino).
Traffic flow analysis: capturing traffic volumes and the separation between vehicles eg, for controlling entry traffic lights to tunnels.
Visitor flow analysis: the counting of visitors which may be combined with the mapping of their routes in an establishment (eg, for planning the arrangement of goods in a supermarket).
Smoke/fire detection: for reporting fire in open areas or tunnels.
Experience reveals that no single algorithm is suitable for all these tasks. So a field test or a comparable reference has to be an important selection tool. Particularly with outdoor applications there are always difficulties in use because of changing weather conditions, hours of sunshine, the seasons etc.
In selecting a suitable process two performance indicators need to be considered: falsely detected situations (false positives), and not recognised detection situations (false negatives). In the first case, alarms (events, etc,) are reported although there is no real alarm situation (event situation, etc). Although there is no acute security risk, with time, the operator’s trust in the system dwindles. Alarms are no longer taken so seriously and a real alarm can easily be overlooked or ignored. The criticality of the second case is very much dependent on the application. How important is an event which is not recognised as a detection event?
Since these two indicators usually work against each other in opposite directions, the improvement of one indicator is nearly always to the detriment of the other. Therefore, before selecting a system it is imperative to define the protection aims very precisely.
Thermal imaging cameras
It is increasingly common to use thermal imaging cameras in combination with video analysis in outdoor environments. Thermal imaging cameras provide images showing the contrasting temperatures in a scene, rather than the differences in brightness shown by conventional video cameras. With conventional cameras, the video analysis process can be significantly influenced by light reflections and shadows. At night the scene has to be adequately lit, and in bad weather the view can be so severely impeded by snow or fog that the function of the video analysis algorithm is extremely limited.
The use of thermal imaging cameras in this kind of situation can offer distinct advantages since they do not require any additional illumination and yet can still make a clear distinction between light and shade contrasts for ‘normal’ moving objects (people, vehicles, animals, etc,) in an image. Additionally they also offer the advantage that there is no lighting to attract insects and interfere with the detection accuracy. This technology is however not suitable for identifying people or objects as neither facial features nor details are adequately reproduced. And because of this, it is often used together with conventional pan and tilt cameras and appropriate lighting.
When these thermal imaging cameras are used for event detection, the conventional P&T cameras are controlled from the centralised system and moved to the right position for the user to see the event displayed in ‘normal’ format.
3D is not just a consumer electronics trend. In video surveillance technology 3D is also an increasingly frequent discussion topic. Its precise meaning is however very variable depending on the supplier and the components. In the video analysis field the term 3D is often used if the system is able to correct for perspective bias – even though a two-dimensional image is being processed. And the term 3D also comes up in relation to the operation of systems using graphical site plans. ‘Real 3D’ ie, image capture using a camera with two lenses and image sensors is not common in video surveillance. However, it is conceivable that in years to come such cameras will be developed and used, as they could provide additional benefits for video analysis processes.
Combined and other camera systems
One further trend evident in the market at the moment are cameras fitted with several lenses. A variant with up to four image sensors and lenses which provide a 180-degree or 360-degree panoramic view from one location is already widely used. This saves cabling and installation costs. A similar purpose is served by cameras with fish-eye lenses whose images are straightened out for display using special software algorithms. Having only one image sensor and one lens, this makes for a more cost-effective solution but the image resolution and light sensitivity is lower.
One newcomer to the market is a variant, which covers one viewing direction with lenses of different focal lengths to provide both overviews and high resolution detailed images of the same scene. Another option is a single pan and tilt camera system which combines a thermal imaging camera with a conventional camera.
© Technews Publishing (Pty) Ltd | All Rights Reserved