Camera optics: DORI and occlusion

A camera either resolves the detail you need at the distance you need it, or it does not. Everything on this page exists to answer that before the camera is on a pole.

DORI, and why pixel density is the unit

IEC 62676-4 defines four operational categories by the pixel density on target, in pixels per metre:

Category	Pixel density	What it supports
Monitor	12.5 px/m	Presence and movement in a scene
Detect	25 px/m	Something is there
Observe	62.5 px/m	Characteristic details, clothing
Recognise	125 px/m	A known individual is identifiable
Identify	250 px/m	An unknown individual can be identified

Each step is a doubling. That is the number worth carrying into a specification conversation, because it means the distance at which a camera identifies is roughly half the distance at which it recognises — and a customer asking for “identify across the yard” is often asking for four times the camera they have budgeted.

Targets are set per zone rather than globally. A loading dock that needs Identify at the door and Observe across the apron is two requirements, and the design is scored against each in its own polygon rather than against a single site-wide number.

Projection: four models, not one cone

A rectilinear lens, a fisheye and a 360° panoramic do not distribute pixels the same way, so they are not modelled the same way. The platform implements four projections — rectilinear frustum, equidistant fisheye, panoramic and equirectangular — and the difference is not cosmetic.

On a fisheye, pixel density falls off toward the edge of the image as a function of the angle from centre. A tool that draws a fisheye as a wide cone with uniform density will tell you a corner is covered when it is not. The equidistant model reflects that falloff, which is why fisheye coverage looks smaller here than in tools that draw a circle.

Varifocal lenses are interpolated between their published wide and telephoto endpoints, so a camera specified mid-zoom is evaluated at the field of view it will actually be set to.

Occlusion: what actually blocks the view

Line of sight is evaluated in three dimensions and is height-aware. A ray that passes above a low wall is not blocked by it. Walls without a recorded height are treated as full height, which is the conservative assumption.

Warehouse racking is the case most often missed. Racking areas are decomposed into their solid rack runs and treated as opaque structure, because a thirty-foot pallet rack in the middle of a field of view is a wall regardless of what the drawing calls it. This is the single biggest source of difference between a computed coverage map and a cone drawn on a plan.

Glass deserves a precise statement. Window elements are see-through to cameras — a camera can cover the space beyond a window. A wall whose material is glass, however, blocks camera line of sight exactly as concrete would, because occlusion is decided by element type and height rather than material. Material and thickness are what drive radio-frequency attenuation, not optical line of sight. This distinction is real and it is stated here rather than glossed, because “we model glass” means two different things across the two disciplines.

Night

Infrared reach comes from the datasheet’s rated IR range. Coverage can be evaluated in night mode, which is how a doorway that identifies a face at noon and loses it at midnight shows up at design time.

The cone drawn in the 3D scene is illustrative — the range is the number to work from.