One of the most important challenges when deciding on a security camera is determining the camera specs to suit the scene. You don’t want to have a small 2MP pixel overlooking a large parking lot where faces or number plates won’t be discernible. This is where DORI comes into play, and in this article, we’ll talk about DORI and how you can optimize it to pick out the most optimal security cameras for your needs.
How To Use DORI When Selecting Security Cameras
DORI is a very important spec in helping users figure out how accurate and at what distance they want the security camera to detect important information. Although these accuracy levels are somewhat subjective, they have been categorized using an absolute metric called ppm/ppf. (Pixels Per Meter/Feet.)
When selecting security cameras for your home or workplace, it’s important to set the stage. This involves figuring out where you will be installing the camera and the amount of information you are hoping to glean from it.
- First, determine what accuracy level you want to prioritize in your security camera. Do you want to “recognize” faces and number plates, or is it enough to just “observe” the scene?
- After that, consider the location you want to cover, is it a small backyard or a wide and open parking lot? Try to get a sense of the field of view (FOV) you want to cover in terms of horizontal distance. (meters or feet)
- Finally, calculate the camera resolution that will accommodate both the ppm/ppf and the measured FOV. (We will cover how to do this, in the final section titled: Pixel Density.) Also, you can do this the other way around and instead calculate the FOV your camera should be able to cover.
How to Use DORI When Designing a CCTV/Security Camera System
DORI stands for Detection, Observation, Recognition, and Identification. This defines the four different levels of accuracy that a security camera can reproduce from a real scene ranging from “Detection”, being the least accurate, and “Identification” (as the name suggests) being the most accurate.
Although this is a spec that is related to security cameras entirely, there are some other things we can do to improve the footage of the entire system starting with improving the scene lighting.
Depending on the range you are specifying for accuracy, you can opt to set up external lighting sources to help improve the camera footage, especially at night. Depending on which low-light system your camera will be using the most (either Infrared or full-color night vision) it will be best to install external IR illuminators or visible white light sources to help the cameras capture better footage.
On top of that, depending on your circumstances, it might also be better if you can use multiple cameras for different purposes. For example: in the case of a parking lot with a view of the street, you might have to use a static camera, with a lower resolution, covering a wider FOV while you use another PTZ (Pan Tilt Zoom) camera with a narrow FOV to focus on specific objects or persons. Therefore, with the help of DORI, you can pick out the cameras which will be perfect for these specific scenarios.
“Detection” is the first level in the DORI system, and it is also the least accurate. In terms of ppm/ppm, the acceptable value for detection is 25 ppm or 10 ppf.
Since this is the least accurate level, it cannot detect faces or number plates. Also, in some extreme cases, it might not even be able to distinguish between a person or a vehicle at all.
Unfortunately, in the case of image processing and computer vision, this level of accuracy is not enough to improve the clarity of the image via post-processing or artificial intelligence but it can recognize motion in the image. Therefore, the only information that anybody (computers and humans alike) can gather from a “Detection” accuracy level is that something is moving around in the scene.
This is the second accuracy level with a ppm of 63. (20 ppf) In this stage of accuracy, it is safe to say that the distinction between humans, animals, and vehicles is apparent. On top of that, at this accuracy level, it is also possible to get a basic sense of the activities which are going around. For example: at this level, we can identify if a person is walking, running, sitting, or lying down. Also, depending on the circumstances, it is possible to observe rudimentary features such as colors and shapes.
Therefore, if you could only see movement in the previous accuracy level, in the observation stage you can see some characteristic details such as clothing, type of vehicle, what someone is doing, etc…
Unfortunately, at this level, you cannot recognize faces or read license number plates but humans and computer vision systems alike will be able to segment whether this is a person, animal, or vehicle.
“Recognition” is the third accuracy level and it usually sits around 125 ppm. (or 40 ppf.) In this stage, the shapes, colors of the person along with other characteristics such as clothing, make and model of vehicles, patterns, etc.. can be recognized very easily.
Depending on the quality of the camera, and lighting levels, it is very easy to distinguish a person or vehicle that you’ve seen before. (either in the scene or in real-life) Also, in this stage, faces are somewhat recognizable, especially if it’s someone you recognize.
For license plates, the letters and numbers can be somewhat recognizable, but depending on the angle and light level, these can be a bit blurry and difficult to pinpoint exactly. Computer vision systems will have no problem differentiating between people, animals, and vehicles. However, exact facial recognition and number plate recognition might be somewhat difficult without the help of post-processing techniques.
“Identification” is the final and most accurate stage. Anything that’s 250 ppm (80ppf) and up is considered highly accurate and the subjects within this range are considered to have the highest amount of information possible by the camera. Faces can be identified very clearly (even by computers) and number plates are very well detailed. However, depending on the focal length of the camera that is used, this accuracy level requires an average distance of no less than (approx) 20m or 65 feet.
What is Pixel Density/PPM/PPF?
Pixel density is used for categorizing the different accuracy levels in DORI. From the name itself (pixels per meter/feet) you can get an idea that it represents the number of pixels dedicated to one meter/ft in the camera’s output resolution. Therefore, the higher the ppm, of an object in the frame, the larger the object is and the clearer it becomes.
PPM is calculated by dividing the image width of the camera (in pixels) by the field of view (FOV) that the camera is covering. Therefore, if you want a camera with a DORI level of “Recognition” (125ppm/40ppf) you’d want a camera with a higher resolution (more megapixels) and lower FOV.
DORI is a subjective scale for the level of accuracy or amount of information that can be gathered from camera footage. The closer an object is to the camera, the more information can be gathered. (assuming all the lighting conditions are met.) Although having high accuracy all the time is ideal, it can be difficult to accomplish. (In some cases, it might also be unnecessary) This is why, when setting up a security camera, users have to prioritize what they want out of the footage and use DORI to pick out the most optimal camera specs for each location.