Testing Bandwidth vs Low Light 2014
The median was 'just' 275% pulled up by a few cameras with over 800% spikes.
Did maybe mean median or maybe meant mean?
Median shalln't be budged when pulled up/down by spikes (stalagmites/tites).
Median is the middle guy with 5 cameras less than and 5 cameras greater than, no matter by how little/lot. (if remembering my homework right)
Ethan,
The tests confirm what I've been saying for years: while AGC may be necessary to obtain usable images in low light, it has extreme downsides - mostly in the form of bandwidth requirements and noise.
The other side of the coin regarding AGC is that it often "muddies" the images, especially of brighter objects in darker scenes. For that reason, we try to either avoid it altogether by turning it off or minimize its effect by lowering AGC level substantially during camera setup.
I strongly believe that manufacturers rely on AGC way too much to inrease their cameras' low light capabilities. I've often found that limiting AGC to 10db or 15db is a good compromise. The image may be slightly dark if the light level is very low but most of the noise, muddiness and excessive bit rates disappear.
Lowering gain levels is a bad idea in low light (especially sub 2 lux which is a typical challenge in outdoor or no light indoor surveillance).
This is a good demonstration of what happens:
OK. Why not try the same test at more moderately low light levels, say 5-10 lux? And try it in higher-contrast situations, where areas of the FOV may have 10x the lighting of other areas. That would be more in line with what I often run into.
Obviously, ultra-low light levels are a different situation, yet the fact remains: AGC can cause as many problems as it solves. Obviously, what works in one situation may not work in another. It's very easy to confuse AGC, especially in my application.
Thank you for clarifying about your 5-10 lux levels.
That's a completely different situation since you have 10 / 20 / 30x the amount of light as a true night / dark scenario. At that point, whether to adjust gain levels is a personal preference issue.
However, at sub 1 lux, AGC delivers far more gains than problems, simply because without it, you'd have no image at all.
Not "a personal preference issue" - a necessity if the image is to be usable.
I've been to many casinos who don't understand the disadvantages of AGC. I've even noted cases where manual irises are somewhat-to-largely closed with AGC struggling to compensate.
Tedor,
Excellent example. I note that high AGC levels show just as poor or poorer picture as very low AGC levels, with 9db to 18db appearing to be the sweet spot. That pretty much jibes with my findings in our application. I've found that in general, manufacturers tend to use higher-than-necessary AGC levels when AGC is set to "Auto".
Of note: AGC compensates for the overall illumination, often sacrificing image quality in brightly lit areas to compensate for less well lit areas.
Ethan,
Interesting you recommend VBR with cap to provide 'spike protection' in fact in most current implementations it does not provide any protection from big data spikes, however it can and should shorten the duration of the spike, as it simply responds to spikes by increasing compression ratio after the event. VBR-with-cap implementations that provide effective 'spike protection', have to drop frames to limit the data size, which is the same as pushing 12mbps over a 2mbps pipe, you lose the data.
The main benefit of VBR-with-cap is reducing storage requirement. It has minimal benefit on network usage/planning as the average 100BaseTX network will easily be saturated with very few cameras when peaks occur, which means packet/data loss. For site to site applications or where bandwidth is truely restricted (Wi-Fi, Mesh, DSL links) VBR with cap is only useful if you are happy drop frames whenever there is significant change or significant activity in the image, unfortunately it is often the time when you need every frame for evidence.
If you have a 100BaseTX network how many cameras can you hang on it with bit rate set to 4mbps? In theory allowing for 20% protocol overhead, and some wriggle room you can say there is approximately 60mbps capacity which at 4mbps per camera would support 15 cameras. This is true with a good CBR encoder, but with most VBR encoders peaking at 16-30mbps (depending on quality setting, you limit yourself to 2-4 cameras on the network to avoid data loss. With VBR-with-cap limit set to 4mbps, since the peaks are not sustained you could argue that more cameras can be supported, but how do you plan for them? How many would you use? 7 - 8 cameras? This is not a big impact on a small system but as the system scales up to a few hundred cameras, it quickly becomes a network planning issue, not just on the camera to VMS link but also for monitoring side.
By their nature, all encoding systems have their limitations, I think VBR-with-cap is mis-understood and often hyped as a cure-all, but it too has limitations.
A, there are a number of errors in your recommendations / analysis.
First,
"with most VBR encoders peaking at 16-30mbps"
That's not correct, assuming you are using H.264, which most everyone does, and a default / average compression level, which almost everyone does. Even at 1080p / 30fps, VBR encoders will rarely exceed 10 Mb/s, even in high complexity / traffic scenes. We've tested this in numerous shootouts.
"[VBR with a cap] it simply responds to spikes by increasing compression ratio after the event"
We are recommending VBR with a cap specifically for low light / high noise scenarios. Since low light / high noise continues for hours (or as long as it is dark / night), it does not occur 'after' any event.
The most important benefit of VBR with a cap in low light / high noise scenarios is that the increased quantization / higher compression levels that occur do not result in reduced quality in those scenes.
So we agree with you that VBR with a cap on a bandwidth 'spiking' camera in low light / high noise will increase the quantization level used. However, this will not cause visible video quality to decrease. We have done such tests on various cameras. Feel free to do that yourself and report back your findings.
Hello,
I have noticed this discussion and few questions came to my mind. 1) If you are recommending VBR with a cap specifically for low light scenes - how to deal with day/night changes? Is there a possibility to set different settings/schedule for day and night?
2) Is it true that VBR does not reach a cap by dropping files/packages as it is claimed by the Undisclosed Manufacturer #1?
Good questions.
1. It is sometimes possible to set different settings for day and night, but usually the cap is active for both. In most cases, nighttime bitrates will be higher than day due to increased gain which increases digital noise. However, that is not always the case, as some scenes may have high movement during the day or repetitive patterns which increase bitrate, but lower activity at night.
So, while it is generally true that the cap is used to prevent spikes at night, I'd recommend both day and night bitrates be measured, to make sure you aren't unknowingly capping bandwidth too low during the day.
2. There are two ways of handling what happens when the cap is reached: dropping frames or increasing compression. Some allow you to choose, but others will simply increase compression. It's technically possible that the cap may still be so low that compression cannot be increased enough to reduce the stream size, but that is rare and likely means the cap is simply set too low.
Thank you for explanation. So am I safe to say that if cap is reasonable, compression should deal with spikes and do not lose packages?
Thanks for the report Ethan. I am curious if you enabled DNR on the Hikvision cameras? By default, Hikvision has the DNR turned off. We turn that configuration item ON and set the Noise Reduction Level to 50 (the default).
In addition I am curious whether True Wide Dynamic Range (WDR) might also have a reduction effect on the nighttime bandwidth. We also enable that feature on the Hikvision cameras with a Wide Dynamic Level setting of 50 (the default).
We use VBR with a cap on these cameras. With configure the stream for 30 fps and 1080p resolution with a 4096 kbps cap and very rarely see any of the cameras even come close to that bandwidth cap. Note that we are also using the H.264+ compression provided by the Hikvision cameras.
Note also all of our Hikvision cameras have integrated IR and most of them have the Smart IR as well so that as you identified that may help with nighttime bandwidth especially indoors.
With all of those camera features enabled, I just don't see that explosion of bandwidth at night on our Hikvision cameras. We might see a 50% increase but not orders of magnitude.
You have piqued my curiosity so the next time I am on site I'll see if I can get some real-world hard numbers on nighttime vs. daytime bandwidth. I struggle with establishing an appropriate VBR cap value. Generally I use 4000 kpbs for 1080p and 8000 kbps for 4K camera resolution. These are just some educated guesses but I don't know what the end effect is on stream quality at night when it might approach some of these cap limits. Another good metric you could collect, if possible, is the quality degradation when cameras hit their bandwidth caps. If the default quality is set to 60 on Hikvision cameras, how much is the quality reduced when it approaches the bandwidth caps (50, 40, ...)?
We have found that in some areas, like outdoor where there are brick pavers for the surface, or lots of grass, we will see a big difference between a 4Mbps and 8Mbps cap, even on a 4MP camera. We prefer to allow them to spike to 8Mbps to get enough details. We generally only use 7-10 FPS though, so using 30 FPS would be even higher.
Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.