Detection, Recognition, and Identification – Thermal vs. Optical IP Camera

By | Published on

Resolution required for Recognition, Detection, Identification depends on the type of camera

By Bob Mesnik

 

Thermal camera identification There is some confusion in the industry about how much camera resolution is required to detect an object, recognize the type of object, or identify exactly what or who it is.  The criteria are different between thermal and optical cameras.  Resolution for thermal cameras and optical IP cameras are measured differently.

 

For example, when defining the performance of a thermal camera we use the Johnson Criteria of “detection”, “recognition” and “identification” (DRI).  On the other hand, IP camera resolution performance is usually defined by the number of pixels in the sensor, and we are usually interested in the ability to identify a person.

 

How much resolution do you need?  This article compares how resolution is defined using thermal and optical technologies.

 

Thermal Cameras

In military circles, “identification” is used along with “detection, and “recognition” as part of the DRI criteria established by John Johnson in 1958.   This standard was established to define the performance of thermal imaging cameras rather than optical IP cameras. Thermal cameras do not provide the real-world color. They can provide color based on the heat from the object they are viewing, but it does not help to define the object.

 

The generals expect “identification” to mean that they can determine if a person is a man or women, or if they are wearing a hat, or maybe caring a gun.  Unfortunately, the Johnson criteria is sometimes misused for optical systems.   As a review, here is the criteria for DRI.

 

The Johnson Criterion defines “Detection”, “Recognition” and “Identification”. This is the ability  of an observer to detect the criteria 50% of the time.

 

  • Detection:                 Ability to distinguish an object from the background
  • Recognition:            Ability to classify the object class (animal, human, vehicle, boat …)
  • Identification:         Ability to describe the object in details (a man with a hat, a deer, a Jeep …)

 

The following pictures illustrate these definitions:

 

detection, recognition, identification

    Left image: Detection – At several kms, 2 targets are detected out of the background 

    Center image: Recognition – a human is walking along the fence 

    Right image: Identification – 2 males with trousers and jackets are identified – one is smoking.

 

Optical IP Cameras

Optical IP cameras provide color images and this can help identify or recognize the object. They will switch to monochrome or black-white when it gets very dark.  We use the terms detection, recognition and identification differently in surveillance applications.  Sometimes we have different meanings depending on the application.  Here are some examples:

 

  • Detection: can mean a number of things.   We sometimes detect or observe something moving in the distance, but in many cases it refers to alarm conditions.  Alarms can include such things as “motion detection”, object removed, object left, line crossing, and other intelligent automated alarm conditions that help provide security.  We require different resolution for each of these applications, so we have to define what we mean before calculating the resolution

 

  • Recognition: Sometimes we use this term when we want to recognize the object as a man or a woman, but in some other cases we talk about situational awareness.  In this case we are interested in how people or vehicles are moving, rather than the details (identity) of the objects.

 

  • Identification: In surveillance applications we usually mean identifying who a person is, or the license plate number, or maybe the numbers on the tail of a plane.

 

Comparison of Thermal and Optical Technology

Thermal cameras and optical cameras use totally different technology.  The thermal cameras provide images based on the heat energy (Infra-Red) emitted by the object, while the optical camera provides an image from the light reflecting off the object.

 

The electromagnetic spectrum consists of radiation that includes visible light (to our eye) and invisible frequencies. Optical cameras can “see” the visible portion of the spectrum and some of the near infrared spectrum.  The infra-red spectrum includes Near infrared, Mid and Far infrared frequencies.  Thermal cameras operate in the Mid to Far infrared frequencies.

 

The-Electromagnetic-Spectrum

 

There are “Uncooled” and “Cooled” types of thermal imaging cameras.  The uncooled thermal cameras operate in Long-Wave Infrared (LWIR) band from 7-14UM (7000nm-14000nm, while the cooled thermal cameras operate in Mid-Wave Infrared (MWIR) and typically use wavelengths of 3-5UM (3000nm ~5000nm).  The cooled thermal cameras are more sensitive to heat differences than the uncooled variety, but cost more.

 

The optical IP cameras provide better detail than thermal cameras because they provide color video rather than the false color provided by monochrome thermal cameras.  Color provides much more information and makes it easier to identify things.   Since, the thermal cameras use heat energy from the target, they operate in total darkness, while optical cameras require some light source.

 

To “see” at night with an optical IP camera, we use standard white light or IR light that reflects off the target.  Optical cameras are sensitive to infra-red (IR) light, but at a different frequency than the thermal cameras.   IR illuminators that operate at 850 nm or 940 nm can be used with IP cameras for night vision.     For more information about the right camera to use when it’s dark, take a look our article, Seeing in the Dark.

 

We usually measure thermal resolution in line pairs, and we measure IP camera resolution in pixels.

 

Resolution Criteria for Thermal Cameras

The Johnson criteria was based on experiments using observers.   It estimates the number of line pairs required across a target, and indicates that there is at least a 50% chance that an observer can accurately achieve the criteria.  This criterion was the first attempt at predicting performance, it has many limitations and should be used with care.  The Sandia Report, History and Evolution of the Johnson Criteria, describes some of the problems with this criterion.  For example, thermal cameras can sometimes detect objects at a further distance than optical cameras because a hot object (relative to its surroundings) can give off a lot of energy and bloom in the video.  This complicates the definitions because in some cases you can “see” a small hot object further away, but if it’s the same temperature as the surroundings, may be impossible to see with a thermal camera.

 

Here’s a summary of the criteria for resolution.  Note that the Criteria was defined in cycles on target (or line pairs across a target).

 

Table 1: Summary of Johnson Criteria

Discrimination Level              Cycles on Target                         Description

Detection                                  1.0±0.25                                         Object is of military significance

Orientation                               1.4±0.35                                         Object aspect

Recognition                              4.0±0.8                                           Class of object (Jeep, tank, etc.)

Identification                           6.4±1.5                                           Member of class

 

Resolution Criteria for Optical Cameras

IP cameras use different criteria for detection, recognition and identification.  Instead of line pairs, we use the pixel count to measure resolution.  This provides a more defined way of determining what detail we can see.  The resolution of IP Cameras is defined by how many megapixels they provide (1.2 megapixels, 2.0 megapixels, etc.).  This is further defined by the horizontal and vertical pixel count (i.e. 1280 x 720, or 1920 x 1080, etc.).   We also use pixels /ft. (or pixels/m) rather than pixels on target to determine the resolution we need for specific applications.  In some cases, we convert the pixels on target to pixels /ft. for consistency.  This criterion is based on a much higher level of certainty of actually meeting the requirement.

 

Here is a summary of the criteria, with a description of what we can discern at a certain resolution (defined in pixels/ft.).  These are estimates from various manufacturers, and documents published by Homeland Security.

 

Table 2: Summary of IP Camera resolution terms

 

Discrimination Level Pixels/ft. Description
Detection: 1.6 pixels/ft. Assumes that we require at least 9 pixels across a target, for example, a 5’ 7” person

 

Recognition 2.7 pixels/ft. This assumes we need at least 15 pixels across the target (that’s 5’ 7”), and allows us to determine if the target is a man or women (if they are wearing different clothes).

 

Identification 80 pixels/ft. Usually refers to being able to identify a person you know, and requires about 40 pixels across the face.

 

 

To learn more about how much resolution is required to identify an object using an optical IP camera take a look at our article, What IP Camera Resolution Do You Need?  To learn more about detection, recognition, identification (DRI) take a look at our article How Far Can We See with the PTZ IP Camera.

 

Summary

Thermal camera resolution is measured differently than Optical IP camera resolution.  They use different technology.  Thermal cameras are measured using line pairs and the DRI criteria, while optical cameras generally use pixels/ft. (or pixels/m).


If you need help in determining the best camera for your application, please contact us at 800-431-1658 (in the USA), or 914-944-3425 (everywhere else), or use our contact form.