Face & object recognition (1 2 3)

As we move into the digital era the older CCTV standards need to have an updated version in pixels. A recommendation in Sweden today is that the width of a face (16cm) should be represented by at least 80 pixels. For object identification such as the letters on a license plate, the height of a letter should be represented by 15 pixels. There are also different performance levels of possible face recognition and identification where the following has been recommended.

In order to compare the % criteria used in CCTV installations, use the following conversion table;

Table 1: Previous demands for identification
Performance	Body representation	Face width
Strong identification	120%	40 pixels
Recognition	50%	17 pixels
Detection	10%	3 pixels

The older criteria of 120 % (400 TVL resolution) only converts to 40 pixels and cannot according to the latest demands, be considered sufficient for more than recognition purposes. If it is to be used for identification of faces, environmental circumstances such as light need to be specifically favorable.

Table 2: Latest demands for identification
Performance	Face width	SKL letter chart	Body representation
Strong identification	120 pixels	7-8	N/A
Recognition	80 pixels	5-6	N/A
Detection	40-70 pixels	2-4	120%

The table above gives you the physical demands of image detail. However the ambient luminance in combination with resolution are dependent on each other. Unfortunately, lighting sometimes seem to be one of many uncontrollable parameters. We will come back to criteria around controlling luminance later on.

The SKL chart referred to in the table can be found in Appendix C. This chart offers letters of different sizes and can be used as a guide when installing and calibrating cameras on site.
In order to give yet another guideline; 60 pixels may be sufficient in order to recognize a person due to a specific feature such as a tattoo. However, to positively identify one person (not facing the camera) by minor features such as light skin moles a much higher resolution (100-120 pixels) is needed.

In addition to this, a frame rate above 5 frames / sec is highly recommended.

See further examples using Axis network cameras.

axisFlexSubContent