Imagery Analysis Techniques

CCTV and Imagery Analysis Techniques and Terms CCTV, Anthropometry, Facial and body mapping, analysis techniques, stereoscopy

CCTV systems We live in a world of CCTV - imagery has become crucial in both the detection and prosecution of crime. CCTV imagery received by FVS varies in quality from very poor to excellent (with the majority being

at the lower end of this scale). CCTV can be filmed in colour, black-and-white, infrared (IR), and can be

under daylight, night, lit and unlit conditions. The way the light is captured, processed and stored can be

lossless or lossy (displaying compression artifacts, blur or pixellation). Our analysts are conversant with all these

concepts and can understand the different achievable results, and as such are trained and capable of providing the

best results with the available footage.

Video imagery is subject to many factors that will affect the quality. Such factors are: Colour Balance: Poor colour balance can adversely affect the way a system perceives an

object’s colour. It may, for example, record a white object as orange under some lighting

conditions (e.g. sodium street lighting).

Distance: The distance that an object is from a camera is inversely proportional to the level of

quality captured i.e the closer the subject is to the camera and its field of focus, the better the

captured image quality will be. It is not possible to add extra detail back into an image of a

distant object by enlarging it; when a digital image is enlarged, it will pixelate resulting in a

perceived further loss of detail.

Sensor Type: Some cameras are classified as day/night capable. Many of today’s security

cameras use sensors which are sensitive to both visible wavelengths of light (3-7 microns)

and the near-infrared (IR) spectrum (7-11 microns), which is emitted from sources such as

sunlight, moonlight, halogen fixtures, etc. In order to produce accurate colours in daytime

most IR light needs to be filtered out, using an IR Cut Filter, which sits in front of the sensor

in the same way one might use a pair of sunglasses to block UV light from the eyes. When

the light drops below a certain threshold the filter is removed to allow more visible and IR light

to reach the sensor. This improves low-light performance however colour performance is

affected. Most implementations remove colour information, yielding a black-and-white image

which is more usable and without chroma noise (graininess caused by the system trying to

process colour in conditions where light is insufficient).

By using IR Light Emitting Diodes (LEDs) the cameras provide ‘black light’ illumination, which

is invisible to the human eye but not the IR sensor. CCTV systems usually automatically

switch into the IR mode when the light level falls. Various types of IR emitters are available,

some of the lower-cost variants operate at wavelengths which encroaches the visible

spectrum, giving a red/orange glow which is visible to the human eye. Other IR LEDs operate

further along the IR spectrum - these do not emit any visible light however require a much

more sensitive camera and as such these systems are more expensive.

Compression: Digital recording systems’ encoders use compression algorithms to

transfer information from the camera sensor and store it on the hard disc drive. Compressing

the information means it takes less disk space enabling the device to store more information.

An algorithm is a specific set of mathematical instructions and rules determining what

information the encoder keeps and what it discards. Typically differences are retained more

robustly, so while a minor difference in brightness across a surface as light hits it may be

discarded, the difference in colour between two surfaces would be retained, as would distinct

motion. Compression can lead to imperfections in the image, known as artifacts. These can take the

form of off-colour rings, blobs or grains around edges of sharply contrasting colour (such as

black objects on a white background), degradation of colour or some blockiness or pixellation

in the image, depending on the algorithm used and how aggressively it is applied.

It must also be noted that excessive noise and graininess (such as in poor light) can be

falsely interpreted by the encoder as movement. It will therefore seek to retain this

information resulting in higher bandwidth consumption. As a result, a greater level of

compression will be applied in order to conserve disk space. Noise and compression

therefore both serve to reduce the quality of CCTV footage, particularly in low light.

Lighting Conditions: At the simplest level, good lighting offers a greater level of definition

and detail. Dusk, night time and bad weather can result in poor lighting for the camera. In

addition, objects can appear differently

under certain lighting conditions due to

refraction effects - for example, when

standing directly under street lighting

flat, straight hair can appear lighter in

colour than if stood in more even light.

IR lighting presents its own set of

challenges based on how an item of

clothing absorbs and reflects light. A

‘dark-toned’ item of clothing recorded

on a daylight system may glow vividly

white under IR.

Movement: This can result in blur. Noise: This is unwanted electronic

interference to the video and can manifest itself as tiny dots (sometimes referred to as ‘snow’)

or patterns on the screen.

Jitter: Video cameras acquire images by sampling "pixels" (an acronym for "picture

elements") at periodic time intervals. Most cameras utilize self-contained crystal-controlled

oscillators for timing the acquisition of pixels, resulting in very stable pixel sampling rates.

Corruption of synchronisation signals or electromagnetic interference during transmission can

randomly displace the horizontal lines of the video image, causing jitter - a jumpy image or a

ghosting effect trailing any movement.

Frame rate: The standard PAL frame rate (i.e., that used in most UK television/film

productions) is 25 frames per second. This frame rate produces a ‘natural’ progression of

images over time and the brain cannot perceive two individual

frames. Many CCTV systems record at much lower frame rates,

capturing only 4-5 frames per second, in which time the subject may

move out of frame or visibly jerk along.

Analysis Techniques Facial mapping: The scientific identification and analysis of the

features that make up the face, their respective relationships with

each other and the morphological differences between individual

parts of the face.

Elements of interpretation: Type, style, shape, size, shadow,

tone, morphological descriptors and associated features.

Morphology: The study of differences, more specifically the

differences between common features (e.g. features which, by their

presence, absence or expression can differentiate two of the same

thing, such as a nose or lips). Photogrammetry: The science of taking accurate measurements

from imagery with a known scale, either through use of an object of

known size and scale within the frame, comparison to a map of

known scale, and by comparison of focal length and distance. It is

crucial to be aware of the strengths and weaknesses of each

technique when conducting photogrammetric analysis.

Methodology Elements of interpretation: Type, style, shape, size, shadow, tone, morphological

descriptors and associated features.

Morphology: The study of differences, more specifically the differences between common

features (e.g. features which, by their presence, absence or expression can differentiate two

of the same thing, such as a nose or lips). Superimposition: Two images from the same angle can be compared by overlaying one

over the other. An image of the offender at the time of the offence superimposed over the

defendant can allow features to be accurately compared.

Stereoscopy: The ‘apparent displacement of an object when viewed from two different

viewpoints. By viewing two images of the same subject taken offset it is possible to view it as

a 3D image through use of a stereoscope device.

Anatomically, there are 3 levels of binocular vision required to view stereo images:

1. Simultaneous perception 2. Fusion (binocular 'single' vision) 3. Stereopsis The perception of depth is created in the brain and not all persons are able to achieve this: either through eye problems, no stereo acuity (12% of the population are unable to view in 3D), 30% of the population have weak stereoscopic vision or they can only see perfectly through one eye. The analyst should wherever possible strive to use stereoscopy in the analysis as it adds another dimension to the interpretation and often proves very useful and can assist in height evaluation, shape and depth of objects and where an object lies in the distance (parallax errors). Reverse stereoscopy: The reversal of the perception – height appears as depth, hollows as

mounds. This occurs when the images are reversed – left for right eye and right for the left

eye. Odd and even field stereoscopy: This is the same

process described above but using the movement

contained within a frame i.e. capturing the two fields and

separating them out into left eye and right eye.

Reverse stereoscopy: It is possible to obtain stereo

images when the object moves rather than the camera

position. As long as the object has not moved too much,

acceptable levels of stereoscopy can be achieved. It is

always best to experiment with adjacent frames.

Anaglyphs: These are stereograms created using photo packages such as Photoshop and

when viewed using red and green glasses a three dimensional depth can be appreciated.