David Brent is a technical information engineer for IT systems at Bosch Security Systems, Inc.
Gil Grissom and his team from CSI: Crime Scene Investigation showcase amazing forensic work, and shows like this have affected expectations for video surveillance systems.
Photo credit: CSI publicity photo/fair use
"Greg, can you magnify the pixels in this video? I believe we may be able to ID the killer from that reflection in the victim's eyes."
That was Gil Grissom to Greg Sanders on CBS' CSI: Crime Scene Investigation.
The "CSI Effect," a very real phenomenon, is how the exaggerated portrayal of forensic science adversely affects jurors, criminals, and forensic science itself. Nested somewhere within forensic science now lies video surveillance, perceived as the magic bullet that solves hundreds of cases within minutes. By way of either mystical pixel magnification, or the likes of CBS' Criminal Minds' Penelope Garcia accessing any video system in the world and gathering the needed video within seconds, the recent portrayal of video surveillance has led to, in many cases, a very misinformed customer base with unrealistic expectations.
The CSI effect has also, in many ways, driven a change in our industry, as we see more movement from standard definition (SD) cameras, to megapixel (MP) and high definition (HD) cameras. The argument now rages on about what equipment to deploy so that you or your customer can be ready for that CSI moment.
In this article, we are going to look at a few of the factors in the deployment of video surveillance systems that affect the quality of video and the ability of the user to find the video or evidence they need. These fundamentals apply across the board no matter what video technology you are deploying.
Field of View
A camera's field of view (FOV) is determined by the angle of view from the lens to the scene and can be measured horizontally or vertically. The lens size (in millimeters), the type of lens, the size of the sensor inside the camera, and the distance from the camera to the target area are all factors that affect the field of view. Most manufacturers have online or handheld field of view calculators to assist with lens selection. Lenses play a key role in any application -- especially in HD and megapixel cameras. The basic equation for lens selection is as follows:
Lens (mm) = Distance (ft) / field of view (ft) x imager format (mm)
To determine the field of view (FOV) needed for an application, reference Johnson's Criteria. Johnson's Criteria was developed in 1958 and is currently used by the Army Night Vision & Electronic Sensors Directorate. It defines four levels of surveillance: Detection, Classification, Recognition and Identification.
This rule simply states the area of a scene, in percentage, that must be occupied by a person or object in order to be Detected (5% FOV), Classified (15% FOV), Recognized (50% FOV), and Identified (65%+ FOV). Although these numbers are based on night vision standards, the concept applies to all video surveillance applications.
For example, to read a license plate during daylight conditions, the rule is as follows: A 1 foot license plate must take up 10 to 15% of the horizontal field of view, depending on compression and resolution used (4CIF, H.263 or H.264).
Frames per Second (FPS) and Resolution
You pay for three things when you install a video surveillance system: inputs, storage and bandwidth. The most expensive portion of the system is the storage, and as I covered in an earlier article "Video and Data Are Not Created Equal," storage and bandwidth are joined at the hip. Frames per second settings are a pivotal part of any video surveillance system, and settings will vary from installation to installation based on the specifications and the type of facility. The first mistake usually made in specifications for a video surveillance system is reducing the frames per second and/or the resolution in an attempt to lower storage expenses.
Running at low frames per second at night coupled with higher frame rates on alarm can be beneficial if you are up against a "small storage, long retention time" scenario. But low frame rates become a problem when you are running at 1 to 10 frames per second all the time. In a high motion area with people and objects moving fast, you can miss important information (evidence) if you are only recording at 1 to 10 frames per second.
I have been to several sites where they have attempted a tradeoff of sorts, reducing the resolution so they could increase the frames per second. This will lead to more issues if the field of view calculations were done with resolution for identification in mind. Reducing resolution from 4CIF to CIF on standard definition cameras reduces your ability to identify targets at a distance.
Standard Definition Cameras
This industry staple has come a long way over the last decade. A good standard definition camera today comes with H.264 compression, a 20-bit dual exposure CCD, and day/night capabilities for deployment with infrared illumination, as well as back light compensation, white balance, and auto black features. If deployed correctly, standard definition cameras can do a terrific job.
What's the catch? With the advent of large HD monitors deployed in operations centers, customers are disheartened with the image that they once were satisfied with, saying the images are pixilated. It is the same 4:3/704x480/345,000 pixel image they once loved, but it has been stretched across a 60 inch/16:9/1920x1080/2,073,600 pixel canvas. You would look a little pixilated too.
Over the last few years, megapixel cameras have been all the rage. The lure is that you can buy fewer cameras and cover more area while still being ready for that CSI moment. But as always, there is a catch. Megapixel cameras are not subject to any standards except for the number of pixels involved. There is no imager compliance, no specific compression compliance, aspect ratios can vary (4:3, 5:4, 16:9), and megapixel cameras have often been limited to lower frame rates (1 to 15 frames per second).
Lenses play a much larger than advertised role in megapixel cameras. Modular transfer function is the measurement of light before and after the optics of a camera. This measurement tells you how well a lens can resolve a single spot (pixel) on the camera's imager. It is represented in line pairs per millimeter (LP/mm). Typical lenses for standard definition cameras have a 30 LP/mm rating, while megapixel cameras must have a 60 LP/mm rating or better.
Choosing the correct lens for a megapixel camera can be difficult, as most lens manufacturers do not disclose the LP/mm rating of their lenses, and the LP/mm needed can vary depending on the size of the sensor in the camera and the size of the pixels on the sensor. Here is the calculation:
0.5 x 1/pixel size (mm) = LP/mm to resolve
Let's look at an example of how this works: You have two 2.1-megapixel cameras with the same resolution of 1920x1080. One camera has a 1/2-inch sensor and the other has a 2/3-inch sensor. The 1/2-inch sensor has a pixel size of 4.5um and the 2/3-inch sensor has a 6.5um pixel. We would need two different lenses to get the most out of the cameras, the 6.5um pixel needs a 77 LP/mm lens, while the 4.5 um pixel needs a 111 LP/mm lens.
Finally, the bandwidth/storage consumption of a megapixel camera is very large, especially if the camera's full capacities are being used. As an example, let's take the "Brand X" 3-megapixel camera that I have in my lab that records in MJPEG format. If I record 10fps for 15 days with an average bit rate between 40Mb/s and 56Mb/s, I would need about 9TB of storage. If this was the only camera I deployed to monitor a parking lot, I would also have a single point of failure. Using this same storage limitation and standard definition cameras, I could deploy roughly 40 cameras at higher frame rates.
High Definition Cameras
HD cameras are megapixel cameras (2.1MP max) with rules. In other words, while all HD cameras are megapixel, not all megapixel cameras are HD. High definition is a global video standard defined by the Federal Communications Commission, Advanced Television Systems Committee and the Digital Video Broadcasting Project.
Where megapixel cameras only adhere to a pixel count, HD cameras are progressive scan devices and have defined resolutions of 1280x720 and 1920x1080. They have a defined aspect ratio of 16:9 and frame rate standards of 30 and 60 frames per second. HD cameras fill the void between standard definition and megapixel cameras; this is the same void created by the discrepancy between reality and the futuristic world of CSI.
HD cameras have the same lens requirements as discussed earlier, but the bandwidth and storage consumption is more reasonable than that of their megapixel brethren. The megapixel camera discussed earlier had an average bit rate of between 40 and 56Mb/s. A HD camera using H.264 compression running at 30fps would have an average bit rate of between 5 and 10 Mb/s depending on the scene. The storage consumption is higher than standard definition cameras but is about one-third of the megapixel cameras. Plus, the HD cameras will look good on your 60-inch plasma in the operations center.
The number one problem I've seen over the years with video surveillance systems is a lack of recorded video for one reason or another. Anything from time synchronization to network outages can affect your system.
Several years ago, I went on a service call to pull video evidence from a location where a murder had taken place. Upon arrival, I found that the video system in question had not recorded a frame of video in more than four months due to internal hard drive failures. That means no one had looked at or checked the video system in more than four months.
Video systems are 24/7 systems that are continuously writing massive amounts of data, and they need periodic system checks. Checks on a weekly basis should include:
- Live playback: Check video feeds, camera power, and network status if working with IP cameras
- Short-term playback: Check the functionality of network video recorders, DVRs, or iSCSI recording services
- Long-term playback: Check the functionality of any DAS SCSI or NAS iSCSI subsystems to ensure retention time requirements are being met.
"I need all the video from this incident in the next hour." It's a request you might hear on the CSI: Crime Scene Investigation television show, and this should be a reasonable request when dealing with small events in a specific period. Post motion detection has been a common feature in video surveillance systems for years, and with video content analysis readily available in better systems, searching metadata after the fact is now a reality if the features are enabled beforehand.
Misconceptions create problems when you are dealing with large systems and looking for the needle in the haystack. In 2001, I worked for a manufacturer that at the time had its CCTV system in the Washington Dulles International Airport and the Pentagon. After the 9/11 attacks, I was part of a team that had the laborious task of reviewing all the video from the airport with several federal agents looking over our shoulders. Did you notice I said all the video? That's every frame from over 300 cameras with 30 days of retention time. The task took three weeks of 15-hour days.
So, what is the answer?
Education, education, education. Educating your customers goes a long way, and an educated customer will likely settle on a nice blend of standard definition and high definition cameras with a sprinkling of megapixel where needed, as long as it suits their storage budget and gets the job done the right way. Educating your customer on proper system checks assists law enforcement with the ability to retrieve video when there is an incident.
The Federal Bureau of Investigation (FBI) has been leading the charge with an educational campaign that it put in place several years ago. In 2010, it released a training video, "FBI: Caught on Camera," which can be viewed on the FBI's website. This video covers camera placement, compression and native formats, among other topics. Don't hesitate to share this with your customers.
And, remember, a nonfunctional or poorly configured video surveillance system is as useful as no system at all. It won't help anyone, not even Criminal Minds' tech-whiz Penelope Garcia.
About the author: David Brent is a technical information engineer for IT systems at Bosch Security Systems, Inc. He has extensive knowledge of video surveillance systems and holds a number of IT and networking certifications. He can be reached at firstname.lastname@example.org.