To compare various surveillance cameras in
3D drawing software we need a set of constants to begin with. Being from Australia, I am working to the
standard AS4806 for the base definitions or resolution criteria. All the examples use here are created using Video CAD Pro but similar results can be obtained with the
JVSG IP Video design tool, which is less than $500.
I use the camera installation height of 2.3 metres and a viewing range of 25 metres long by 2 metres scene height, which is typical of a building mounted external camera covering a road way or car park.
My references in the scene are:
- A typical height man located at distances of 1, 2, 3, 4, 5, 10, 15, 20 & 25 metres
- A vehicle parked at 5, 10, 15, 20 & 25 metres
- A Rotakin target at 5, 10, 15, 20 & 25 metres
For each camera, where the lens is integral to the camera, the lens will be tested at its widest and telephoto angles to show the coverage detail at different test object distances.
Where the camera allows you to inter change the lens a 3.6mm and a 12mm will be used as the reference regardless of the imager size and aspect.
The starting pixel density is calculated on a typical height person (1.7m) occupying 100% of the screen height. This is referred to as
1ICU or approximately 350 pixels / metre. From this we can define our core criteria’s as follows:
Observe
– Generally observe/monitor behaviour within a broad area
At this resolution an operator can observe activity in a scene but not be able to determine what a particular person is doing. Typical applications are crowd control for congestion or disturbances in public areas. This type of coverage is of minimal use outside of live monitored systems where an operator can respond to changes in the scene and would often have a PTZ to zoom in to the scene for a more detailed image.
In the image to the right (click to enlarge), we can see at the back centre of the scene a black man at 36 metres from the camera using a 4CIF camera with a 67 degree lens.
The minimum person size is 29 pixels tall, which at 4CIF is 5% of the scene. This would be a pixel density of 18 pixels / metre.
Detect
– Detect the presence of a person or object in the scene
A person can be detected in the scene but not recognised or identified. Typical applications include tracking a persons movement around a scene as an overview, with some detail to distinguish what they are doing. An example would be an isle in a supermarket.
When combined with other cameras for
identification, the results can be used to track and convict shop lifters.
Our scene would look like this, where the black person is able to be detected and tracked from 18 metres away from the camera using a 4CIF camera with a 67 degree lens.
The minimum person size is 58 pixels tall, which at 4CIF is 10% of the scene. This would be a pixel density of 35 pixels / metre.
Recognise
– The ability to recognise a person known to you
A person in the scene should be able to be recognised if they are already known to you. for example a staff member, regular customer or registered police suspect.
It is common that a surveillance camera will only provide recognition for a portion of the visible scene where the person is close to the camera and the further regions will only provide
detect coverage.
The
pixel density required for recognise is greater than
176 pixels / metre (Vert) which is around 44 pixels for the persons head. On a PAL 4CIF signal the person would occupy at least 50% of the scene height.
Identify
– The ability to identify a person you do not know
This is the requirement of most CCTV surveillance systems but also the one most often not achieved. In any surveillance system intended to cover a general public area, at least one camera should provide a identification image. ideally this is located at a chock point where the public mast pass such as a entry door, corridor or counter.
At a counter it is possible to achieve a good quality identification image and also provide areas of
recognise,
detect and
observe.
The black man to the right is at the maximum this camera could provide ID based on the above settings, using a 4CIF camera with a 67 degree lens. The distance is only 1.1 metres due to the camera angle exceeding 30 degrees.
If I lower the camera by 30cm to 2 metres, I obtain a significantly better result as per the second image.
The pixel density required for identify is greater than 352 pixels / metre (Vert) which is around 88 pixels for the persons head. On a PAL 4CIF signal the person would occupy at least 100% of the scene height.
Number Plates
– The ability to read the characters of a vehicle number plate
Also referred to NPR or LPR (Number or License Plate Recognition) is where the human eye can read each character. This is different to ANPR, where the recognition is automated via character recognition software.
Each character needs to occupy 5% of the scene height based on a PAL 4CIF signal. Because each country has different number plate sizes, this is going to vary depending on where you live but based on an average height of 7cm, the total scene height must be less than 1.4 metres
To read the number plate, we need to tighten the lens angle. This image shows a car at 8 metres with a 17 degree angle of view.
This is a pixel density of 410 pixels / metre or for an Australian number plate, or 29 pixels tall for each character. This is also suitable for person identification but often the camera will be pointed too low to cover faces.
Other factors that should ne considered for NPR applications are the amount of reflected light from the number plate at different times of day and nigh as well as shutter speed to accommodate the speed of the vehicle. Typically a camera with a wide dynamic range is required such as the
Panasonic Super Dynamic range.
There are many other factors that should be considered when designing the surveillance layout but they need to be taken in to account on a site by site basis. For example if a camera is in a shop with windows to the outside, we need to consider the suns position at different times of the day. Will there be sufficient light at night time?
