UoL CS Notes
Menu

Human Vision

ELEC319 Lectures

Terminology

Field of View
Total angle covered by the camera in a single image.
Field of Regard
Total angle that can be covered by the camera in a series of images.
Focal Length (fl)
Effective distance between lens and focal plane.
Aperture (d)
Effective size of the lens.
Pixel
Individual element of the image.
F-number (fl/d)
Measure of the light allowed through the lens.
Gain-offset Correction
Control of image exposure (gain) and removal of dark current (offset).
Frame Rate
Number of frames per second.
Dioptre
Optical power of a lens (same as $\frac 1 {\text{fl}}$)

Human Eye

The human eye has two optical components:

  • The cornea has an optical strength of 31 dioptres.
  • The lens is only 19 dioptres.

Rods & Cones

Rods:

  • Sensitive to all types of light.
  • About 100 times more sensitive to light than the cones.

Cones:

  • Concentrated around the centre of the field of view.
  • Come in three types:
    • L (Long)
    • M (Medium)
    • S (Short)

Other animal often have limited colour vision due to a differing number of rods and cones.

Fovea & Saccades

The fovea is responsible for what we normally think of a human vision.

  • The high resolution region of human colour vision is only 2 degrees.

Stereo Vision

Binocular vision relies on the two eyes having different views of the same scene.

The distance between the two cameras is called the baseline.

The fact that the two cameras can see different views is called parallax.

Empirical Condition:

\[\frac{\theta_A-\theta_B}{\theta_\text{FOV}}\times N_\text{Pixels}>1\]

Colour

Colour images are composed of three bands: R,G,B

Often it is better to use HSI (Hue, Saturation, Intensity) instead of RGB, as the three colour bands overlap. This causes a change in one band to effect the others.

It is possible to convert from one colour space to another (CMYK to RGB conversion). A formula should be provided if this is required.

Camera Resolution

Resolution is limited by pixels and the field of view.

A field of view of $\theta$ rads with $N$ pixels gives an angular resolution of:

\[\Delta\theta = \frac\theta N\]

Johnson Criteria

This is an empirical experiment that determines the ability to correctly detect and recognise objects.

  • Detection (with 50% probability) = resolve 1 bar.
  • Recognition (with 50% probability) = resolve 3 bar.
  • Identification (with 50% probability) = resolve 5 bar.

1 bar is a minimum of 3 pixels.

Optical Effects

Vignetting
The shadowing of the corners of an image from the optical aperture (the effect of the finite focal length).
Optical Distortion
This is the effect of imperfect lenses. Barrel distortion expands towards the centre. Chromatic distortion presents itself as a prism effect.
Image Noise
Non-uniform response to stimulus across the focal plane.
Blooming
Where high intensity pixels leak into neighbours.