Know Your Bandwidth - Network Use and Planning

How much bandwidth will this all take? What's the load on my network? How can I intelligently manage things?

Many things can affect the bandwidth required to move/record/store video surveillance data. Everyone knows all about resolution and frame rate, and at least a little bit about compression levels. What most people don’t know is that none of those matter as much as the most basic variable of all: how much motion there is in field of view of the camera.

Virtually all systems use event based recording, there’s no point recording when there is nothing going on. At Cloudastructure, the first cloud based video surveillance company in the world, we coined Event Based Transmission (tm) – we only transmit recordings offsite when something is going on. The savings on the customer’s upload broadband are immense.

What’s an event? Events could be analytics driven (object in frame, e.g. person, vehicle, animal; object left in room; object taken from room; someone entering through an exit; tailgate/piggyback etc.), but the most basic analytic is motion detection. If there is no motion in the field of vision, there’s almost always no reason to bother recording it. Certainly, there is no reason to use upload broadband speed to transmit it.

So, what matters most?

• Resolution?
• Frame Rate?
• Encoding/compression?

Take a look at this table.

• Resolution goes from 640×480 to megapixels. That’s a factor of about 16.
• Frames per second can go from 1 to 30. That’s a factor of 30.
• Compression can go from 1 to 50.
• Motion detection, though, varies from 0.1% (e.g. Fire Exit) to 100% (e.g. busy intersection in a major city). That’s a factor of 1,000.

As you can see, motion matters most. So, what can you do about it?

The best answer is the simplest answer: mount the cameras so that they are not looking at continuous motion scenes. For example, mount your front door camera so it is looking at people going through the door, not at the cars driving by on the street behind it. If you want to find an event, it’s easy if every event is someone coming through your door. It gets a lot harder if, for every time someone comes through your door, there are 1,000 cars driving by on the street — you’ll get bored watching cars really fast. Plus your upload broadband connection will thank you for not transmitting all that video you specifically don’t want to watch.

Can we record 24/7? Of course, if you’ve got the broadband and a need for continuous recordings, let us know. After your IT department looks at your upload speed, however, you might decide to go with Event Based Transmission (tm) after all.

Cloudastructure, at least, doesn’t charge based on resolution, frame rate, compression or motion. We’re both on the same side when it comes to good camera placement, and we’re happy to help you figure out the best mounting methods for your cameras. You’d like to save broadband and irrelevant recordings as much as we would.

What about compression? Two notes on compression:

• Cameras are really bad at it. The processors in the cameras can barely keep up serving multiple streams. Almost all video surveillance compression algorithms are lossy, meaning they lose data. You can’t uncompress them and get all the pixels back. To make matters worse, that over tasked tiny processor that came free inside your network camera is going to have to throw away a lot of data just to keep up with the bit rate. That is data that you’ll literally never see again.
• A common trick of NVR vendors is to turn the already lossy compression way, way up — making it more and more lossy. This allows them to advertise longer recording time etc. It’s better to keep compression way, way down. When something happens, you want the best frame-by-frame resolution you can get. Those compression algorithms (esp. h.264) are made to give the appearance of smooth video playback when you’re watching a DVD, not the best frame-by-frame view of a suspect’s face.
• Cloudastructure, btw, does all the compression in the x86 class CPU’s in the CVR -- sometimes we ship 32-cores in a single CVR), at very high fidelity (NOT lossy) settings.

Frame rate? You’re much better off with 5 good frames per second than 30 blurry ones. We think 5-10 frames per second is a great setting. 24-30 works too. It’s only a factor of 6, so set it wherever you want. Personally, I keep mine at home at 10 (where I have fewer cameras) and the office at 5 (where there are more cameras).

Pixels? Why pay for a high resolution multi-megapixel camera if you’re not going to use it? Max that sucker out. The forensic value of the data will go way up when you can zoom in on a face too.

TL;DR: record at your camera’s maximum resolution, set the frame rate in the middle. Keep your compression settings turned way down (large file size), if you have access to them. The smart move is to mount your cameras so that you’re seeing the events you want and not the events you don’t.

‍