The vast majority of video surveillance cameras feature noise reduction technology. Noise can be a significant issue with regard to video performance, and addressing it not only increases quality but makes the cost-of-ownership more palatable. Benchmark assesses noise reduction to see whether it delivers the required functionality.
[dropcap]W[/dropcap]hen the team at Benchmark first started testing video surveillance cameras many years ago, our specified low light criteria showed that most devices had a sensitivity of around 8-10 lux. This wasn’t surprising as most professional manufacturers quoted such figures.
As time passed, the specified figures dropped, and today it is not unusual to see camera specifications with a sensitivity figure that is below 1 lux. Indeed, some cameras’ specifications are into three decimal places, such as 0.001 lux!
You would be forgiven for thinking that a drop in specifications from 10 lux to 0.001 lux is a sign of great innovation, of rapid developments in technology, and therefore a cause for celebration in the surveillance sector. However, that’s not the case.
The main reason for these dramatic reductions is not because cameras are significantly better with regard to low light performance, but because manufacturers now measure sensitivity in different ways. Whilst there have been advances with regard to processing, they are nowhere near as dramatic as the figures indicate. For the installer and integrator, it makes sensitivity specifications pointless when trying to ascertain the likely performance in the field.
As image resolutions increase, so working in low light becomes more of a challenge. The higher number of pixels on an HD or megapixel chipset means that each image element is considerably smaller than on a standard resolution sensor. Getting an appropriate amount of light to fall onto significantly smaller pixels creates difficulties. It’s physics, and all the marketing spiel in the world can’t change that!
Most cameras’ DSPs use gain to amplify the video signal. Whilst this does amplify video information, it also amplifies other parts of the signal too, and this results in noise.
Video noise can be annoying. It doesn’t look good, and whilst it is typically the result of a camera being made to work beyond its capabilities, it gives the user an impression that their investment has been wasted on an inferior system. Expectations are damaged!
With systems using image compression, noise can have another negative impact. Because it is seen by compression engines as change within the image, noise can increase bandwidth and storage needs, and that equates to a system with a higher cost-of-ownership.
It is a rare camera which does not include noise reduction, and whilst manufacturers have tweaked their methods, the bottom line is that the reduction of noise is dependent upon a number of criteria. It is simple to set up a camera in a sympathetic environment to show clean images in low light. Translate that into the real-world, and problems can surface.
Noise reduction is achieved through processing of the video image, and as such the actual reduction itself can have a detrimental effect on quality. Where a camera requires other processing such as WDR, highlight compression or BLC, there will be a need for balance between all processing elements. In many Benchmark camera tests, we have seen image degradation caused by certain combinations of video processing. Because HD and megapixel cameras can suffer in low light, noise reduction is often one of the problem functions!
For the purpose of this test we used cameras from Axis Communications, Samsung Techwin and Panasonic (see Equipment panel for further information). The cameras were installed at a working site, and the end user allowed us to take control of the system over a weekend to carry out the various assessments.
It should be noted that the cameras were configured in different ways, as per the needs of the user. Because the aim was to assess the impact of noise, rather than to pitch the cameras against each other, this wasn’t an issue. Indeed, if anything it allowed a greater variety of real-world assessments to be carried out.
For the purpose of testing, the cameras were set to stream with a variable bit-rate. Whilst we would never recommend the use of variable bit-rate in applications, for the purpose of this test we wanted to see how the data flow was impacted by noise and noise reduction processing. Constant bit-rate is the preferred choice as it allows the installer or integrator to allocate bandwidth and storage levels to each video stream in accordance with their needs.
It is worth remembering that when a constant bit-rate stream hits its limits, something will inevitably be lost. Some cameras will drop frames, others will lower quality, some will show high levels of pixelation. In short, any increase in the stream size will impact on the final image quality!
Before we started the test we carried out a straw poll of opinions about the impact we expected noise to have on overall bit-rates. The general consensus was that bit-rates would rise by up to 100 per cent as a result of the compression engine over-working.
With the test site properly illuminated, ambient light levels were around 300 lux. We used this light level to measure the base settings for the cameras. Initially we turned off all the processing features that we could (some cameras have automatic management which runs a level of processing for certain features regardless of settings).
Whilst the cameras had different streaming configurations, the results were pretty much as expected. All cameras were configured for real-time HD streams using H.264, with an emphasis on image quality.
Gain was set to a middle point, ensuring some amplification without over-doing things. The Axis camera was optimised for minimal motion blur: why go for 50fps and accept blur? The Panasonic camera had its DNR set at Low. Finally, the Samsung unit had SSNR turned off.
At HD720p with a frame-rate of 50fps, the Q1614 had an average bit-rate of 9Mbps with occasional motion. The WV-SP509 delivered HD1080p streams at 25fps with an average bit-rate of 11Mbps, while the SNB-6004P streamed HD1080p video at 50fps with an average bit-rate of 12Mbps.
These figures do not represent a minimum for the delivery of good quality video. We opted to pretty much give the cameras free reign!
Defining what constitutes a level for acceptable noise will always be subjective, so instead we reduced illumination in the protected area. We opted to lower the illumination level to 2 lux.
This was well within the quoted sensitivity levels of each camera, and allowed a colour image to be captured, albeit with a degree of noise. Also, in recent tests of HD cameras, the average switching point for day/night operation – as defined by the manufacturers as a factory default – was just below 2 lux!
What was interesting, and admittedly slightly unexpected, was the degree to which bandwidth needs increased as a result. Noise was visible in all of the camera streams, and to be fair it was at a pretty similar level across all three feeds.
The Q1614’s bandwidth went from 9Mbps to an average of 30Mbps during periods of motion. That represents an increase in transmission and storage needs in excess of 310 per cent! The WV-SP509 increased from 11Mbps to an average of 26Mbps: the camera would not permit the DNR to be switched off so it was set at ‘Low’. Finally, the SNB-6004P saw an increase from 12Mbps to 34Mbps!
Whilst the figures quoted here are averages, there were some peaks which occurred during the test, including a rise of 700 per cent in bit-rate during motion-heavy activity which was typical for a commercial site.
It is also worth noting that all devices were optimised for image quality, and so used light levels of compression. If compression was to be used at higher levels, the impact would obviously be greater.
The next stage was to assess what impact noise reduction had. There are two considerations with regard to this. The first is how the bit-rate is affected, but the second – how the image looks – is arguably more important. As with anything, balance is key. The reduction of noise is only acceptable if the processing doesn’t introduce other degradation. Replacing noise problems with motion blur or artefacting doesn’t do anyone any favours!
The Q1614 uses a simple slider which balances between noise and motion blur. The degree to which the latter is acceptable will always depend upon the application, and the viewed scene. For example, if a protected area includes a lot of high speed motion, even a low level of blur can be detrimental to the overall level of surveillance.
By setting the level at around 25 per cent bias towards low blur, the visible noise levels don’t really fall, but bit-rate eases a bit. With the level balanced at 50 per cent low noise and 50 per cent low motion blur, bit-rate falls to around 20Mbps. However, motion does seem a little less fluid, and there is slight blur. Also, latency is more pronounced.
Moving to a 75 per cent bias toward low noise does show a visibly cleaner image, and bit-rate falls further to around 17Mbps. Motion blur is obvious with fast moving objects, and there are signs of slight stutter. Moving to a total bias towards low noise sees bit-rate drop marginally to 16Mbps, and motion blur is an obvious issue.
With the WV-SP509, noise reduction cannot be disabled, and so our starting figures were for footage with a Low level of DNR applied. The only option is to increase this to High. Visually it doesn’t have much of an impact, and you really need to be looking carefully to see change. The bit-rate only reduces by a little, by around 20 per cent, and as such the noise remains a problem.
The SNB-6004P includes SSNR, which was arguably the feature that the Samsung brand was first noticed for. SSNR (Samsung Super Noise Reduction) has been through various developments over the years, and is claimed to reduce bandwidth and storage needs.
The implementation requires a careful touch. At the high levels, the image does clean up considerably, and bit-rate fell by almost 50 per cent. However, motion blur was obvious, and even a person walking at a normal pace showed signs of ghosting.
In our test, a setting of higher than 16 created too many issues with motion blur, whilst a setting of lower than 10 saw noise problems remaining obvious.
Whilst it is often said that video noise can impact on the proficiency of compression, the degree to which this happens is quite significant, and often misunderstood. Obviously, the heavier the level of compression applied, the more significant the impact will be. Whilst most applications will use constant bit-rates, thus avoiding network flooding, the degradation of the image will be detrimental to the overall performance of the entire system.
Installers and integrators should realise that noise reduction technologies do what they state, with the emphasis on reduction. They are not noise removers, nor do they have the ability to perform miracles! Take any ‘reduction’ too far, and other issues with image quality will result.
The test illustrates that noise is a bigger problem than many think it is, and it is also a costly issue with regard to the drain on resources. It is staggering how differently cameras behave in adequate light and low light, and the real performance-killer is noise.
Noise reduction can help, but the best way to ensure the highest level of performance is to design systems that do not create noisy images in the first place!
Axis Communications – Q1614 The Q1614 is a box-type HD720p camera. It delivers HD720p streams at frame rates of up to 50ips. The main stream algorithm is H.264, and Motion-JPEG is also supported. Sensitivity is quoted as 0.2 lux for a 50ips stream, or 0.1 lux for a 25ips stream. For the test, the camera was configured for 50fps streaming. The Q1614 uses an exposure menu to adjust gain, and the priority for processing is set via a slider. This allows a bias towards ‘Low Motion Blur’ or ‘Low Noise’.
Panasonic – WV-SP509 The WV-SP509 is a3 megapixel box-type camera which can stream HD1080p video. It incorporates a number of smart options, including digital noise reduction. The camera streams at rates of up to 25ips, using H.264 and Motion-JPEG processing. The camera uses a 1/3 inch MOS sensor, and has a claimed sensitivity of 0.5 lux. DNR is automatically applied, although the level can be changed via the GUI. Options are for High or Low.
Samsung Techwin – SNB-6004P The SNB-6004P is an HD1080p box-type camera which features the WiseNet III processing engine. This incorporates a host of features including SSNR, Samsung’s proprietary noise reduction system. The camera streams at rates of up to 50ips, using H.264 and Motion-JPEG processing. The camera uses a 1/2.8 inch CMOS sensor, and has a claimed sensitivity of 0.1 lux for a 50IRE image. SSNR can be set as active or inactive, and it has 32 levels to allow a good degree of fine tuning.