Height Estimation from CCTV Footage

The reference-object method is the most commonly applied approach in casework because it requires no scene survey and can be applied to archived footage when the scene is no longer accessible. The logic is straightforward: if a door of known height H occupies h pixels, and the suspect occupies s pixels at the same distance, then the suspect's height is approximately H times (s divided by h).

The key phrase is 'at the same distance.' The moment the subject and reference object stand at different distances from the camera, the simple ratio breaks down. A person standing one metre closer to the camera appears taller in the frame than the same person standing behind the reference object, even if both are the same height. The correction factor requires knowing the geometry, which takes the method toward the more rigorous vanishing-point approach.

1. Identify a reference object of confirmed known height visible in the same frame or at the same distance from the camera as the subject.
2. Measure the pixel heights of both the reference and the subject's standing position, from sole of foot to top of head (with hair included, as stature is measured to the top of natural hair height in forensic practice).
3. If positions differ, apply a perspective correction using the camera geometry or an approximation from scene measurements.
4. Calculate height and propagate uncertainty from each of the above steps.

Where the scene provides sufficient parallel lines, the vanishing-point method can estimate height without relying on any specific reference object at the subject's exact position. The scene's vertical vanishing point (directly above or below the zenith) and two horizontal vanishing points define the camera's orientation. With at least one horizontal reference length (a tile width, a road marking length), the analyst can construct a metric vertical scale anywhere on the image ground plane.

The vanishing-point method handles camera tilt explicitly because the vertical vanishing point's location encodes the camera's tilt angle. A strongly tilted surveillance camera pointed steeply downward will have a vertical vanishing point far outside the image frame, which the analyst locates by extrapolating the converging verticals in the scene. Once located, it provides the correct perspective correction for vertical extents anywhere in the image.

Standard CCTV installations mount cameras on ceilings or high on walls and tilt them downward to cover the scene. A camera tilted 30 degrees below horizontal introduces a perspective distortion that means a person standing one metre closer to the camera appears measurably taller in the frame than a person of identical height standing two metres further away.

The correct treatment is to model the camera tilt explicitly as part of the calibration. When a ground survey is possible, the camera height, tilt angle, and horizontal distance to reference points can all be measured directly. When the scene is no longer accessible, the tilt angle can be estimated from the vertical vanishing point position in the image, or from measuring the pixel heights of horizontal lines at known floor-level distances.

When no reference object is visible and no scene geometry is available, analysts sometimes resort to body proportion estimation: counting how many 'head heights' the standing figure occupies, or using head-to-body ratios drawn from population data. This method has a long history in anthropometry but a limited accuracy record for forensic metric purposes.

• The adult head-height to body-height ratio varies substantially across populations, ages, and sexes. The often-cited '7.5 head-heights per body' is a mean value with a standard deviation of roughly half a head-height, equivalent to plus or minus 5-7 cm on a typical adult.
• Clothing, hair, posture, and shoe thickness all affect the apparent head-to-body ratio in an image. Thick-soled shoes alone can add 3-4 cm to apparent stature.
• Combined, these sources produce uncertainty ranges of plus or minus 10-15 cm at 95% confidence, compared with plus or minus 3-5 cm achievable by geometric methods under good conditions.

The UK Forensic Science Regulator's guidance on CCTV measurement, echoed by SWGDE guidelines in the US and similar documents in Australia and Canada, is unambiguous: height estimates must include a stated uncertainty range at a specified confidence level. The sources of uncertainty that must be considered are:

• Pixel localisation: selecting the exact pixel corresponding to the top of the head and the bottom of the foot is typically accurate to plus or minus 1-2 pixels, translating to a scene distance depending on the camera geometry.
• Reference dimension: the known height of the reference object carries its own uncertainty. A door measured on-site to the nearest millimetre contributes less error than a door assumed to be standard height.
• Camera model: any error in the calibration or tilt estimate propagates to a systematic error in every measurement.
• Subject posture: a slightly slouched posture, or a person standing on their toes, shifts the apparent height. The analyst should select frames where the subject is standing most naturally upright.

A report conforming to FSR guidance would state something like: based on the reference-object method with camera tilt correction and a physical survey of the scene, the subject's height is estimated at 1.78 to 1.86 metres at 95% confidence. The dominant source of uncertainty is pixel localisation of the crown position in a low-frame-rate recording, contributing approximately 3 cm to each bound. This is a forensic measurement. A statement that says the subject is approximately 1.82 metres is not.