# Virtual Plaster Cast: Digital 3D Modelling of Lion Paws and Tracks Using Close-Range Photogrammetry

> Can a camera and a ruler replace a bag of plaster? Researchers in KwaZulu-Natal, South Africa photographed lion paws and tracks from many angles and turned the photos into 3D models that matched calliper measurements to within a fraction of a centimetre. Seven or eight photographs were enough, taking less than a minute per track.

Summary by Kersey Lawrence, Original Wisdom. Updated 9 November 2026. Page: https://www.originalwisdom.com/tracking-literature/2016-marchal-lejeune-virtual-plaster-cast-digital/

## Citation and access

- **Authors:** Antoine F. J. Marchal, P. Lejeune and P. J. N. de Bruyn
- **Year published:** 2016
- **Journal or source:** Journal of Zoology
- **Full citation:** Marchal, A. F. J., Lejeune, P. and de Bruyn, P. J. N. 2016. Virtual plaster cast: digital 3D modelling of lion paws and tracks using close-range photogrammetry. Journal of Zoology 300(2): 111–119.
- **DOI:** 10.1111/jzo.12342
- **Access status:** Hosted (freely available elsewhere)
- **PDF:** https://www.originalwisdom.com/wp-content/uploads/2026/11/2016_Marchal_and_Lejeune_etal_virtual-plaster-cast.pdf

## Tags

- **Species:** Mammals > Lion
- **Region:** Africa > South Africa
- **Tracking:** Track measurements; Track photographs; Tracking; Tracks; Tracks and signs identification
- **Methods and technology:** Camera traps; Direct observation; Mann–Whitney U test; Wilcoxon matched pairs test
- **Data type:** New field data
- **Observer qualification:** Not reported
- **Discipline:** Wildlife tracking
- **Year published:** 2016

## Detailed keywords

- **Species:** Studied: lion (Panthera leo), 20 paws of captured lions and 20 tracks. Mentioned from earlier track studies: leopard (Panthera pardus), tiger (Panthera tigris), mountain lion (Puma concolor), black rhinoceros (Diceros bicornis) and white rhinoceros (Ceratotherium simum); dinosaur tracks recorded with photogrammetry in ichnology.
- **Location and habitat:** KwaZulu-Natal, eastern South Africa. Paws: Tembe Elephant Park (about 300 km², in Maputaland on the border with Mozambique; sandy plains with ancient coastal dunes, mean annual rainfall 700 mm; about 40 lions in July 2015). Tracks: Hluhluwe-iMfolozi Park (about 900 km², hilly, 40 to 560 m above sea level, mean annual rainfall 650 to 985 mm; about 120 lions in July 2015). Tracks were found in dry riverbeds in Hluhluwe-iMfolozi and on sandy roads in Tembe.
- **Topics:** three-dimensional (3D) recording of paws and tracks; close-range photogrammetry; structure-from-motion; virtual plaster casts; track measurement accuracy; masking (cutting out the background), camera calibration and optimization; re-projection error; correction factors; minimum number of photographs; contour lines to outline a track; processing time; manipulator (measurer) bias; limits of two-dimensional (2D) track recording (tracings, acetate drawings, photographs); pugmark census of tigers in India; future individual identification of lions from tracks. Inconsistencies in the paper: in Figure 4 the middle panel (b) is labelled “Pad-masked paws” inside the graph, but the caption and Table 2 show it is pad-masked tracks. Table 2 gives 194 measurements for unmasked tracks against 200 for the other two models, without explaining the difference. The abstract says auto-calibration gave better reconstruction quality, but calibration made no significant difference for paws (P = 0.822). Optimization is reported as significant, although the mean re-projection errors differ by only 0.01 pixels (1.03 versus 1.04 for paws; 0.47 versus 0.48 for tracks).
- **Statistics or models:** Mann–Whitney U test (paws versus tracks; number of photographs needed); Wilcoxon signed ranks test (effect of reconstruction settings); Spearman’s rank-order correlation (direct versus digital measurements); mean geometric deviation and correction factors (Direct = Digital minus Digital × Correction Factor); mean predicted error with 95% confidence intervals; significance at P ≤ 0.05; analyses in R. Software: Agisoft PhotoScan Professional 1.1.4, Agisoft Lens, Adobe Photoshop, CloudCompare (Principal Component Analysis (PCA) fit).
- **Data type:** New field data: 20 lion paws photographed during night-time captures in Tembe Elephant Park and 20 lion tracks in Hluhluwe-iMfolozi Park. Dates of data collection are not given (lion numbers are given for July 2015; the paper was received in October 2015).

## Synopsis

### 1. What they did

Track measurements have long been used to identify species and even individual animals, but they are usually recorded in two dimensions (2D): measured directly, traced on acetate sheets or photographed. The authors point out that such methods are controversial, because tracings and measurements can vary with the person doing them and with the substrate. They tested whether close-range photogrammetry, building a three-dimensional (3D) model from many ordinary photographs, could serve as a quick, cheap and reliable field method for recording lion paws and tracks: a “virtual plaster cast”.

- Sampling. They photographed 20 lion paws during night-time captures in Tembe Elephant Park, KwaZulu-Natal, South Africa (the captures were part of park management, not this study), with the paw held still on a stand. They photographed 20 clear lion tracks in Hluhluwe-iMfolozi Park. Each paw or track was photographed 10 to 15 times from different angles and distances, with a ruler beside it for scale. The length and width of the main pad and the four toes were also measured with a vernier calliper. One person, the first author, did all the sampling.

- Software settings. Using Agisoft PhotoScan, they tested three settings that affect how well the photographs line up: masking (cutting away everything except the pad and toes), camera calibration (automatic or done beforehand) and an optimization step.

- Accuracy. They compared the calliper measurements with the same measurements taken from the 3D models. For tracks they also tried leaving the photographs unmasked and outlining each toe and pad with 0.5 mm contour lines instead of cutting them out by hand.

- Number of photographs. They rebuilt models from random subsets of 5 to 10 photographs to find the fewest needed.

### 2. What they found

Software settings

- Masking clearly improved the alignment of both paw and track photographs.

- Automatic calibration was slightly better than pre-calibration for tracks, but made no significant difference for paws.

- Optimization gave a small but statistically significant improvement for both.

Accuracy

- Without correction, the 3D models overestimated pad and toe sizes by an average of 0.23 cm for masked paws and 0.50 cm for hand-masked tracks. Unmasked tracks outlined with contour lines were the closest, underestimating by 0.06 cm on average.

- A correction factor brought the average error down to 0.03 cm for paws, 0.04 cm for masked tracks and 0.01 cm for unmasked tracks.

- Direct and digital measurements were closely related (Spearman’s r of 0.98 for paws, 0.96 for masked tracks and 0.93 for unmasked tracks).

- The authors suggest that the paw overestimate may partly come from the calliper squeezing the soft, elastic pads, so that the direct measurement is too small.

Number of photographs

- Error rose and model volume shrank as fewer photographs were used. The minimum for a reliable model was seven photographs for paws and eight for tracks. With only five photographs, model volume was about 68% (paws) and 85% (tracks) of that with ten.

- About 7% of photographs were discarded as blurred or of poor quality.

Time

- Taking the photographs took less than one minute per paw or track.

- Cutting out the background by hand took about 1.5 minutes per photograph for tracks, and semi-automatic masking of paws in Photoshop took about 3.4 minutes on average.

- Processing ten photographs took about 12 to 53 minutes for a paw and 2 to 6 minutes for a track, depending on the computer. Unmasked track photographs took much longer to process (about 46 minutes against about 2 minutes for masked ones).

### 3. What they concluded

- To the authors’ knowledge, this is the first use of close-range photogrammetry to record the paws and tracks of living animal species in 3D. It gives objective and reliable 3D models of both.

- The field kit is only a digital camera and a ruler, and photographing takes less than a minute, which keeps handling of immobilised animals short and suits work near dangerous animals.

- They recommend taking about twice the theoretical minimum (14 to 16 photographs), masking paws but not tracks, and using automatic calibration and optimization.

- 3D recording should be less affected by the person doing it than 2D tracings or single photographs, and should allow more rigorous, objective and repeatable use of tracks. They expect it to allow more individuals to be identified, on more kinds of substrate, from fewer tracks per animal.

- The authors were working on applying the method to identify individual lions from their paws and tracks.

### 4. How tracking was or was not used

The study is about recording tracks, not following them. The 20 lion tracks were chosen as clear tracks and identified as lion either after the lion was seen making them, from a camera trap in front of the track, or by identifying them with a tracking field guide (Liebenberg 1990; Gutteridge and Liebenberg 2013). The pad and toe measurements of each track were then compared with measurements from its 3D model. No trailing is reported. The introduction notes that San trackers in Namibia were 96% accurate in reading species, age, sex and individual from tracks (Stander et al. 1997).

### 5. Tracker or observer certification

Not reported. The first author did all the sampling; the paper does not say who identified the tracks, and no tracking qualification or test is given. Most tracks were confirmed by direct observation or camera trap, but some were identified with a field guide only.

### 6. Other technologies, skills or innovations

- Close-range photogrammetry with Agisoft PhotoScan Professional, combining structure-from-motion and dense multi-view 3D reconstruction, to build dense point clouds and 3D surface meshes.

- Nikon D7100 and Nikon D80 digital cameras, a ruler for scale and two markers on the photographs.

- Semi-automatic masking in Adobe Photoshop for paws and manual masking in PhotoScan for tracks.

- CloudCompare to fit a box around each pad and toe and extract length and width automatically, and to draw 0.5 mm contour lines that outline toes and pads in unmasked tracks without hand cutting.

- A camera trap (Cuddeback Attack) to confirm some tracks.

### 7. Implications for conservation (from the paper)

- Monitoring threatened species provides the baseline for conservation, research and management, and tracks are a non-invasive, cost- and time-effective way to study animals that are hard to see.

- Identifying individuals from their tracks would have major implications for behavioural ecology, conservation biology and wildlife management.

- The authors note that tracking is less invasive than camera trapping, needs less investment and logistics, and is not prone to hardware failure and theft.

### 8. Implications for tracking (Original Wisdom's observations)

_These are Original Wisdom's observations, not the authors'._

- A track is a 3D impression, and this method keeps its depth and shape rather than flattening it into a tracing or a single photograph. A phone or camera and a ruler may be enough to make a lasting digital record of a track for later checking or measurement.

- The accuracy of the method was tested against calliper measurements of clear tracks, not against how well people identify tracks; it is a recording tool, and identifying whose track it is still depends on the tracker.

## Cite this summary

Lawrence, K. 2026. Summary of: Virtual Plaster Cast: Digital 3D Modelling of Lion Paws and Tracks Using Close-Range Photogrammetry (Marchal and Lejeune et al., 2016). Original Wisdom, Tracking Literature. https://www.originalwisdom.com/tracking-literature/2016-marchal-lejeune-virtual-plaster-cast-digital/