Technology can make our lives a lot easier and a lot more complex at the same time.

For example, we never had to worry in the past about how much coverage we could get from a camera. No one used to mention pixels-per-foot (or per meter) ten years ago. Why? Because when we wanted to see up close, we used PTZ (pan-tilt-zoom) cameras to zoom right in on the person’s face or their license plate. With 20X zoom, you had all the detail you needed.

But PTZ cameras need someone driving them to get those close-ups. Megapixel cameras offer the ability to have a fixed camera that can record enough detail. You can zoom in later, or in real time, without moving the camera.

In many cases, that is a lot better. Unfortunately, there just aren’t enough pixels to do all the zooming we do with a PTZ camera. You need 400-600 million pixels to get the same 360 degrees of coverage with 20X zoom that you get with a PTZ camera. That’s 100X more than even the large 5 megapixel cameras sold today.

So, we need to figure out how much coverage we can get with megapixel cameras. Life just got more complex.

John Honovich provided a great service to integrators with the testing he did to show how many pixels per foot you need to get good detection. See his article here:

http://ipvideomarket.info/report/pixels_surveillance_video_test

However, as he points out, you can’t reduce it down to one single number. It depends on many factors.

But here are a few rules to make this simpler.

First, remember this important fact: There is a big difference between general surveillance and trying to recognize a person’s face or their license plate.

Facial and license plate recognition require about 10X – 20X as many pixels per foot, and you need special lighting, plus a number of other factors to keep in mind. Studies have shown that even close up pictures of someone who has changed their hair color or beard is hard to confirm for certain. But it requires less resolution to recognize someone you know. This makes recognition m0re difficult to design for, especially if you want to use the evidence in court cases where jurors never know the people in the videos.

General surveillance, however, where you are only trying to detect if people are in the area, or how many people there are, is much simpler. John’s study shows that 5-12 pixels per foot should be enough. Our testing at VideoIQ shows that our video analytics match this number, which means that our analytics are just about the same in their ability to recognize people in an area as people watching a monitor.

Actually, our analytics might be slightly better, since we recommend that 5-8 pixels per foot for most conditions. This means that if people are too small to see, or the lighting is too dim to see them on a monitor, then video analytics are probably not going to accurately detect them either.

Not all analytics technologies are this accurate. Some require more resolution. More importantly most analytics systems don’t analyze all the pixels. You will find that systems generally scale the video resolution down to minimize processing power for analytics, so be sure to check the actual resolution they are analyzing.

A good second rule of thumb here is: Good video analytics require about the same pixels per foot for accurate detection as humans watching a monitor.

To make life simpler, also remember this: Horizontal pixels determine coverage.

This means that knowing how many pixels are on the horizontal axis, and knowing the pixels per foot for good detection, you can calculate the coverage of a camera. This simplifies estimates. Below are some standard resolutions calculated, based on 5-8 pixels per foot for general surveillance:

CIF (357 horizontal pixels) . . . . . . . . . . . . 45 – 70 feet of coverage

D1 or 4CIF (704 horizontal pixels)  . . . . . 90 – 140 feet of coverage

1.2 MP (1,280 horizontal pixels) . . . . . . . 160 – 250 feet of coverage

1080p (1,920 horizontal pixels) . . . . . . . 240 – 380 feet of coverage

3.1 MP (2,048 horizontal pixels) . . . . . . . 250 – 400 feet of coverage

5 MP (2,592 horizontal pixels) . . . . . . . .  320 – 510 feel of coverage

What is surprising is that the above numbers are true no matter what focal length lens you use!

A 3 mm lens will give you good detection up to about 100 feet away, while a 10 mm lens will work to around 300 feet (for a D1 resolution camera), but in both cases you still have the same horizontal coverage. So, you can copy the above numbers and use them for your planning estimates.

A couple things to keep in mind:

  1. Interlaced video reduces horizontal resolution. The latest studies I have seen suggest a 25% reduction in horizontal resolution. So, be sure to reduce coverage numbers when cameras are using interlaced video (1080i means it is interlaced, while 1080p is progressive, and almost all analog cameras use interlaced video).
  2. You need lenses with the capability of capturing the full resolution of the imager, or you aren’t going to get the full horizontal coverage. This, unfortunately, is a common problem with many megapixel cameras shipping today.

A few interesting observations emerge from the above chart:

  • Even though a 1080p camera only has 2.1 megapixels, it has virtually the same horizontal coverage as a 3.1 MP camera. That’s because of the 16:9 aspect ratio of HD video.
  • A 1080p camera has almost 3X as much coverage as a standard resolution camera (D1 or VGA), but a 5 MP camera only has about 30% more coverage than a 1080p (and you will only get 30% added coverage if you have a true, high quality, 5 MP lens).

You can see why 1080p looks like the resolution that the industry seems to be moving towards.

Even though there seem to be  way too many resolutions of megapixel cameras to choose from, and way too many numbers to keep in mind, things are getting simpler.