What does 50 percent probability of target detection signify on Sarix TI thermal series of cameras Specification sheets?

Issue

Unclear or unintuitive statement on the Sarix TI datasheet.

Product Line

Pelco Cameras

Environment

ESTI635, 650, and 6100 Series - all versions.

ESTI314, 335, and 3100 Series - all versions.

TI614, 635, and 6100 Series - all versions.

TI314, 335, and 3100 Series - all versions.

TI206 Series - all versions.

Cause

In the late 1950s, Johnson's criteria was originally established by the US military to develop a standard method to evaluate the effectiveness of camera systems to detect, recognize, and identify objects.

Resolution

In the mid- and late-1950's the military was developing electro-optical image intensifiers which provided enhanced visual surveillance capabilities under conditions of limited visibility. The complexity of these intensifiers and associated target acquisitions systems required a methodology for evaluating performance characteristics.

Johnson's criteria defines the relationship between a person's visual perception of an image and the quantitative specifications that can be measured in the laboratory. The "50% probability" is defined as the maximum range which a person with average visual acuity can detect/ recognize/ identify a vehicle or person 50% of the time.

 Through a series of experiments using trained observers, Johnson was able to develop a method relating the "decision response" by normalizing resolved line pairs for a critical target dimension. Johnson found that the minimum resolution required for a particular "decision response" was nearly constant for a group of nine military targets.

 Decision Response Chart

The data shows that the minimum resolution required for a particular decision activity is a constant for nine military targets within a maximum error excursion of +/-25%. These target transformations were found to be independent of contrast and scene signal to noise ratio as long as the contrast in the resolution chart was the same as the contrast in the complex target. These results indicated that complex military targets may be considered equivalent in a visual sense to repetitive resolution patterns of appropriate spatial frequencies for each decision level. The results are general, at least for the limited group considered, and are independent of distance. They simplify considerably the determinations of decision level activity in any imaging system, since it is only necessary to determine the angular resolution characteristic as a function of a few parameters. These transformations, which provided target discrimination criteria based upon resolution, gained widespread acceptance within the industry and became the accepted criteria for performance measurement of optical systems. These criteria were referred to as the "Johnson Criteria".

Other definitions of this terminology are used, but most conform closely to those in the Table above. For example, the Air Force defines target detection as "an object is detected although no further target information can be determined"; target orientation as "target symmetry and dimensional shape are noted"; target recognition as "the target can be placed (e.g. the target is a house, a truck, or a tank)"; target identification as "the target can be described to the limit of the observer's knowledge (e.g. the target is a motel, a pickup truck, or an enemy tank)".

  1. When a target is not quickly found, searchers tend to 'over search' (repeatedly search) likely areas and completely avoid areas dismissed as either unsuitable or as suitable but not containing the target. Frequently targets in contextually unlikely places are not found for minutes even though of adequate size, resolution, and contrast for quick recognition when examined.
  2. Despite instructions and training, few observers systematically search a scene until after initial rapid scene-appropriate search fails to find a target. Clearly, search is neither purely systematic nor purely random.
  3. Observers sometimes forget which areas have been searched and assume that they have searched an area when they have not. This leads to large time scores when the target is there.
  4. Other things being equal, target objects closer to the center of the picture tend to be found quicker.
  5. Numerous moving image studies show that subjects under high pressure do hurry to find targets much quicker than those under little or no pressure.
  6. Some observers quickly find targets that others with equal training find only after extended search time or do not find at all. Chance factors, such as looking at the right place early in search, are clearly important. However, some subjects are consistently as much as two to three times faster than others over dozens of targets and scenes, and across studies.
  7. Averaged across many subjects, identically-appearing target images vary drastically in the time required to detect and to recognize them in different backgrounds (scenes). In other words, there is a strong target background interaction.
  8. When briefing target pictures are rotated relative to the target in the scene, or are of a different size or lightness, target detection and recognition are slower.

 
Some sixteen years after publishing his original paper, Johnson collaborated with Walter Lawson on another paper in 1974 which modified the original work and extended it to cover infrared systems. They added another decision response term, identified as target "classification", which they defined as "the visual act corresponding to perception of the general class of military targets e.g. tracked versus wheeled vehicles". Much of the nomenclature in Johnson's original paper was changed in the second paper. What was referred to as "decision response" is now referred to "discrimination levels".

The later paper emphasizes that the values for the various discrimination levels are "representative values, essentially average values required for 50% probability, and must not be construed as rigid values or optimum values for specific targets and target aspects". They thus recognized the less than precise nature of the empirical data. Another change in nomenclature is the change to "cycles for 50% probability" from "resolution per minimum dimension". The later paper changes the average values shown in Table 1 to cycles for 50% probability, as shown in the following table.

The 1974 paper also introduced the use of minimum resolvable temperature (MRT) for thermal viewers. MRT is used to determine the maximum subjectively resolvable frequency for the effective target temperature difference. The procedure developed by Johnson and Lawson for thermal viewers involved five steps, as shown below:

  • The first step requires determination of the effective target temperature difference and determination of the minimum dimension. The temperature difference associated with the target is the area weighted mean temperature difference calculated from actual target signatures; atmospheric properties are then used together with this temperature difference to establish the effective target temperature (apparent temperature) difference at the observer station.
  • The second step requires calculation of the device minimum resolvable temperature (as a function of spatial frequency)
  • Determination of the number of resolvable cycles across the target minimum dimension.
  • Determination of the recognition probability (or other discrimination level) from the number of resolvable cycles.
  • Construction of the recognition probability versus range function.

To determine the number of resolvable cycles across a target (N), Johnson developed the relationship:

 

He also developed a function he referred to as the Target Transform Probability Function (TTPF). This function was derived from laboratory psychophysical experiments in which the ability of observers to discern the nature of tactical targets as a function measured. The sensors used for these experiments were low light thermal viewers. The TTPF is a target detection and recognition probability function and is derived from laboratory and field experiments. The below chart shows plots of data for target detection and recognition.

The two plots to the left are derived from laboratory experiments and the plot to the right is derived from field data. Johnson considered the TTPF to be of fundamental importance in the prediction process. The TTPF replaced the earlier optical image transformations became the new "Johnson Criteria". The new criteria had broader applicability since it was valid for ' both optical and thermal viewers.

Johnson's 1974 paper also addressed the issue of background clutter. He emphasized that the detection function is applicable under conditions which require some degree of target shape discrimination in order to detect the target, i.e. where significant background clutter is present. He also stated that the number of cycles required to attain a particular detection probability can vary significantly depending upon the nature of the background clutter. Both amplitude and spatial measures appear to be required to predict observed trends.

   

 

Based upon this definition, clutter could be categorized as high, moderate and low, where high clutter exhibited a signal to clutter ratio (SCR) of less than one, moderate clutter an SCR of 1 to 10, and low clutter an SCR greater than 10.

From this data it can be seen that for a detection probability of 0.50, the number of cycles (LP/TGT) varies between 0.5 for low clutter and 2.5 for high clutter, with moderate clutter at 1.0. Using the moderate clutter case as a reference, the low clutter condition requires 50% fewer cycles and the high clutter requires 250% more cycles for the same detection probability of 0.50.

It was concluded that "detection performance is a strong function of clutter as well as resolution. Detection range performance prediction models must therefore include clutter effects. Since the number of line pairs per target sub tense necessary for detection is inversely proportional to detection range, changes in SCR can be expected to significantly alter range performance."

Johnson's criteria were developed based upon well controlled laboratory tests using models, but were not verified during field tests against actual targets. As any observer can testify, lab conditions and "real world" conditions can be vastly different.