Hi all. First post, so please be kind. 
Let me start off by saying that we're building a new home out in the woods where any would-be perp could ply his trade quietly without alarming any neighbors, because we don't have any. Since we don't have the basic security of "prying eyes" of the neighbors, we've got to supply that ourselves. A security camera system is required.
When designing the system, the question quickly turned to "How do I know how much resolution I need, and how long of a lens do I need?" My plan is to cover a perimeter around the house and be able to positively identify anyone approaching either down the driveway or through the woods. The woods coverage will also help for hunting season too.
Here is a rule I'm using to govern my design: A system that can't provide positive identification of someone approaching or breaking in is not a security system, it's a documentation system. A documentation system will only document that someone broke in, not identify who did it.
For my system, I defined a perimeter that I wanted to monitor. The plan is to install enough cameras to cover the perimeter with enough pixels to be able to identify a face at the perimeter distance. If they get through the perimeter, they may or may not get picked up, depending on the path they choose. The idea is to identify them before they know they're being observed.
How many pixels does it take to identify a person's face? This became the single most important question I needed to answer. If my perimeter didn't have enough pixels, my money would be wasted because the perp could walk to an uncovered area before identification could be made. If I put too many pixels on the perimeter, then I wasted $$.
I did a bunch of searching and I think the perfect # of horizontal pixels to have enough detail for positive identity. The magic number is 42 and assuming a width of 12" for a normal face, 42 pixels per horizontal foot would be enough. With this spec in hand, next was to figure what that meant for different resolutions.
1.3MP (1280 horizontal): 1280 / 42 = 30.48'
2MP (1920 horizontal): 1920 / 42 = 45.71'
3MP (2048 horizontal): 2048 / 42 = 48.76'
5MP (2560 horizontal): 2560 / 42 = 60.95'
What this means is that a 2MP camera has enough pixels to identify a person's face up to about 46' away from the camera. If you try to identify someone at a distance further out than that, there are fewer pixels on the face, so the image is grainier and a positive identification is harder to make. Of course, this isn't a hard and fast rule, but rather a guideline. The real shocker is how little is gained with a 3MP camera. Once you look at the actual image size, it's really obvious that the biggest difference between a 2MP and 3MP image is in the height, not the width. This would be most helpful when using a camera from a higher location as the extra height would be helpful in covering more ground due to the taller footprint.
Now that I know how much horizontal distance I can cover and still identify a face, I have to decide how to make that work with my specific situation. I think for most situations, with cameras shooting mainly horizontally, the best coverage for the $$ is a 2MP camera. Moving to a 3MP camera requires a 50% increase in storage space for a 6.67% decrease in cameras required to cover the same horizontal perimeter. Likewise, a 5MP camera requires a 150% increase in storage space for 33.33% decrease in number of cameras required. Of course, there will be times when the expense of more camera installations outweigh the storage requirements, but neglecting that, 2MP cameras look to give the best bang for the buck. For HV value cameras, this means buying the 3MP cameras and running them at 2MP
The last part was to figure out what lenses to select for each location. To keep things simple, I'm looking at HikVision fixed bullets or domes and mainly looking at 2.0mp units. To maximize the foot of perimeter coverage per $$, I want to make sure to minimize overlap and maximize coverage.
I started out with one of the many Angle of View sheets available and converted it to Excel so I could expand it.
https://drive.google.com/file/d/0BxTn1WoReGXdd1BVQ3BsVlM2Mkk/view?usp=sharing
What this shows is how much horizontal area is covered by the camera at a certain distance for a particular lens length. Note: the 1.3MP Angle of View numbers are wrong. For example, a 1.3MP 4mm HV dome is 52 degrees while the 3MP 4mm HV dome is 75.8 degrees. The linked document works HV 3.0MP (selectable from 1.3MP to 3MP) cameras.
To use this spreadsheet, find the intersection between the lens length on the left and the distance at the top. This intersection is how wide of an area is visible. The color of the cell will indicate how much resolution you need. Since I'm trying to stay in the smaller and more affordable HV cameras, the lenses I have available to me are 4, 6 & 12mm. Knowing that for a 2.0mp camera I need to stay under 45.7 feet to guarantee face identification, the 4mm lens is good out to 35', the 6mm is good out to just over 55' and the 12mm is good out past 110'. This means I can have a perimeter 35' from the house if I want to use 4mm lenses, 55' from the house with 6mm cameras and so on. Of course, as the perimeter distance is increased, the amount of perimeter increases too.
Most important note: Regardless of resolution/lens/distance combinations chosen, the number of pixels needed to identify a face doesn't change. This number is the key and it is up to you to change this number to suit your needs and budget. Pixels cost money, so this number is directly related to total system cost. You can save money by covering the perimeter with fewer pixels, but the trade off is going to be that it will be tougher to identify someone at your perimeter if necessary. Ideally, perps would always walk right towards the camera and give plenty of opportunity for identification, but I wouldn't count on it.
This process has worked for me, and I'd be interested in hearing other's system design criteria and process so I can prevent costly mistakes.
Thanks for all of the great information I've found here so far.
RydForLyf
Let me start off by saying that we're building a new home out in the woods where any would-be perp could ply his trade quietly without alarming any neighbors, because we don't have any. Since we don't have the basic security of "prying eyes" of the neighbors, we've got to supply that ourselves. A security camera system is required.
When designing the system, the question quickly turned to "How do I know how much resolution I need, and how long of a lens do I need?" My plan is to cover a perimeter around the house and be able to positively identify anyone approaching either down the driveway or through the woods. The woods coverage will also help for hunting season too.
Here is a rule I'm using to govern my design: A system that can't provide positive identification of someone approaching or breaking in is not a security system, it's a documentation system. A documentation system will only document that someone broke in, not identify who did it.
For my system, I defined a perimeter that I wanted to monitor. The plan is to install enough cameras to cover the perimeter with enough pixels to be able to identify a face at the perimeter distance. If they get through the perimeter, they may or may not get picked up, depending on the path they choose. The idea is to identify them before they know they're being observed.
How many pixels does it take to identify a person's face? This became the single most important question I needed to answer. If my perimeter didn't have enough pixels, my money would be wasted because the perp could walk to an uncovered area before identification could be made. If I put too many pixels on the perimeter, then I wasted $$.
I did a bunch of searching and I think the perfect # of horizontal pixels to have enough detail for positive identity. The magic number is 42 and assuming a width of 12" for a normal face, 42 pixels per horizontal foot would be enough. With this spec in hand, next was to figure what that meant for different resolutions.
1.3MP (1280 horizontal): 1280 / 42 = 30.48'
2MP (1920 horizontal): 1920 / 42 = 45.71'
3MP (2048 horizontal): 2048 / 42 = 48.76'
5MP (2560 horizontal): 2560 / 42 = 60.95'
What this means is that a 2MP camera has enough pixels to identify a person's face up to about 46' away from the camera. If you try to identify someone at a distance further out than that, there are fewer pixels on the face, so the image is grainier and a positive identification is harder to make. Of course, this isn't a hard and fast rule, but rather a guideline. The real shocker is how little is gained with a 3MP camera. Once you look at the actual image size, it's really obvious that the biggest difference between a 2MP and 3MP image is in the height, not the width. This would be most helpful when using a camera from a higher location as the extra height would be helpful in covering more ground due to the taller footprint.
Now that I know how much horizontal distance I can cover and still identify a face, I have to decide how to make that work with my specific situation. I think for most situations, with cameras shooting mainly horizontally, the best coverage for the $$ is a 2MP camera. Moving to a 3MP camera requires a 50% increase in storage space for a 6.67% decrease in cameras required to cover the same horizontal perimeter. Likewise, a 5MP camera requires a 150% increase in storage space for 33.33% decrease in number of cameras required. Of course, there will be times when the expense of more camera installations outweigh the storage requirements, but neglecting that, 2MP cameras look to give the best bang for the buck. For HV value cameras, this means buying the 3MP cameras and running them at 2MP
The last part was to figure out what lenses to select for each location. To keep things simple, I'm looking at HikVision fixed bullets or domes and mainly looking at 2.0mp units. To maximize the foot of perimeter coverage per $$, I want to make sure to minimize overlap and maximize coverage.
I started out with one of the many Angle of View sheets available and converted it to Excel so I could expand it.
https://drive.google.com/file/d/0BxTn1WoReGXdd1BVQ3BsVlM2Mkk/view?usp=sharing
What this shows is how much horizontal area is covered by the camera at a certain distance for a particular lens length. Note: the 1.3MP Angle of View numbers are wrong. For example, a 1.3MP 4mm HV dome is 52 degrees while the 3MP 4mm HV dome is 75.8 degrees. The linked document works HV 3.0MP (selectable from 1.3MP to 3MP) cameras.
To use this spreadsheet, find the intersection between the lens length on the left and the distance at the top. This intersection is how wide of an area is visible. The color of the cell will indicate how much resolution you need. Since I'm trying to stay in the smaller and more affordable HV cameras, the lenses I have available to me are 4, 6 & 12mm. Knowing that for a 2.0mp camera I need to stay under 45.7 feet to guarantee face identification, the 4mm lens is good out to 35', the 6mm is good out to just over 55' and the 12mm is good out past 110'. This means I can have a perimeter 35' from the house if I want to use 4mm lenses, 55' from the house with 6mm cameras and so on. Of course, as the perimeter distance is increased, the amount of perimeter increases too.
Most important note: Regardless of resolution/lens/distance combinations chosen, the number of pixels needed to identify a face doesn't change. This number is the key and it is up to you to change this number to suit your needs and budget. Pixels cost money, so this number is directly related to total system cost. You can save money by covering the perimeter with fewer pixels, but the trade off is going to be that it will be tougher to identify someone at your perimeter if necessary. Ideally, perps would always walk right towards the camera and give plenty of opportunity for identification, but I wouldn't count on it.
This process has worked for me, and I'd be interested in hearing other's system design criteria and process so I can prevent costly mistakes.
Thanks for all of the great information I've found here so far.
RydForLyf
