Recovery of 3D footwear impressions using a range of different techniques

Three‐dimensional (plastic) footwear impressions are frequently found at, or in the vicinity of a crime scene, and may provide a valuable form of evidence or intelligence. This paper compares the traditional methods of casting and/or two‐dimensional photography with Structure from Motion (SfM) photogrammetry. We focus both on the recovery of class characteristics (sole pattern) and randomly acquired characteristics caused by damage. We examine how different recovery techniques influence visualization of outsole features and discuss what effect this may have on evidential value. Five shoes and their associated three‐dimensional impressions made in both sand and soil were compared using a grid system and tread descriptors commonly used in the UK. We conclude that within the limitations of this study SfM photogrammetry allows superior levels of visualization of both class and randomly acquired characteristics, giving a better definition in detail in some instances. The use of SfM as a complementary approach can therefore lead to a potential increase in evidential value.


| INTRODUC TI ON
Methods for the recovery of 3D footwear impressions at crime scenes have remained relatively unchanged for decades (1,2). Impressions are either cast or simply photographed in situ. There is little or no research into the effectiveness and sources of error associated with casting despite its widespread use. Protocols for mixing the plaster are largely informed by a practitioner's field experience although Bodziak (1) provides some guidance. The research that is available such as the recent paper by Sabolich (3) focuses on the use of fixative sprays or other practical/logistical solutions such as mixing methods and containers (4). None of this work assesses the errors or accuracy of casting. A recent exception is that by Snyder (5) who compared casting and examination quality photography. She states that dental stone casts are often not collected due to costs, mainly in terms of time, and due to a belief, that photography provides a sufficient amount of information.
Snyder shows that casts are superior in most cases capturing randomly acquired characteristics (RACs) that are not visible on photographs.
Recently, the development and increased use of Structure from Motion (SfM) photogrammetry (2) offer an alternative method of capturing 3D impressions. SfM can digitally capture a 3D impression in contrast with a physical cast. The aim of this paper is to provide an initial comparison of the effectiveness of photography, casting, and digital 3D models in capturing class characteristics and RACs in footwear impressions.

| ME THODOLOGY AND ANALYS IS
The aim of the first experiment was to simply compare the visibility of class characteristics across different recovery methods.
Five outsole impressions using five different shoes were made in oven-dry playground sand by the senior author (weight 57 kg). The impressions were made indoors in a controlled environment (i.e., shallow plastic tray). The five shoes used were: 1. Adidas hiking shoes; 2. Adidas fashion trainers; 3. New Look trainers; 4. Nike Running trainers; and 5. Nike fashion high top trainers. They were selected as representative of a cross-section of typical outsole designs currently on the market. Three of those shoes (1. Adidas hiking shoes, 2. Nike running trainers, 3. Nike fashion high top trainers) were worn for a second time by the senior author to make impressions in a natural sand environment. The main differences between the two sand types were sorting, moisture content, and bulk density. The playground sand was well sorted with 90% of particles between 425 and 300 µm, by contrast the beach sand contained only 40% within this grain-size range. Moisture content and bulk density where both higher in the natural sand environment as one would expect.
The above experiment was then repeated replacing sand with purchased topsoil and again placed within a shallow tray in the laboratory. The senior author made all the impressions once again using the same shoes having first cleaned the outsoles. The natural soil had a higher clay content (clay loam) and bulk density than the bought topsoil (sandy loam).
The second part of this study assessed the preservation of individual damage features or randomly acquired characteristics (RACs).
For this study three brand new, identical and unused shoes (Branded Air Tech, Female Sizes 4, 5, and 6) were used to make impressions The three outsoles had artificial damage features added and each shoe had an increasing level of damage severity. The artificial marks were made to replicate the normal damage shoes might experience such as wear, cuts, holes, and abrasions. These shoes are referred to as Shoe A, B, and C, with C being the most severely modified ( Figure 1). Footwear impressions were made in a controlled indoor environment using oven-dry playground sand within a shallow tray.
The shoe wearer simply walked through the tray to leave an impression. Fine dry sand allows for fine detail to be captured in an impression and increases the chances of RACS being transferred from sole to impression.
After all impressions described above were made, they were photographed in line with photogrammetry guidelines (7). This includes taking a minimum of 20 photographs from varying oblique angles and directly above, ensuring all photos overlap. The photographs were then uploaded to freeware DigTrace (www.digtr ace.co.uk), which uses Structure from Motion (SfM) photogrammetry to create 3D point cloud models. The models where scaled and then auto-rectified so that the principal plane was orthogonal to the vertical and then color rendered using a variety of different color ramps (2).
Casts of all footwear impressions were also then made using current and advised methods as set out in the UK National Policing Crime scene quality photographs were also obtained from directly above each impression using a tripod following best practice set out in the UK National Policing Improvement Agency This was repeated a minimum of ten times to ensure repeatability Highlights • Use of SfM Photogrammetry for recovery of footwear impressions.
• SfM recovery compares favorably over other methods when visualizing RACs.
• One key advantage of SfM is the use of depth color renders.
• Digital recovery allows superior visualization, digital file sharing, and searching.

| RE SULTS
The results from the class characteristics experiment are shown in Table 3 and a visual example of the comparison using one shoe can be seen in Figure 3. The percentage of identified descriptors varies both with environment and method. Pair-wise Mann-Whitney tests were used to explore the significance of these differences (  Looking beyond the class descriptors at more specific RACs we see further differences.    Our aim is not to prove that one method is better than the other but simply to raise awareness of alternative methods and approaches. Digital SfM photogrammetry produces accurate, reproducible results (7) and as shown here in certain circumstances give superior visualization of both wear and individualizing footwear characteristics.

ACK N OWLED G EM ENTS
The authors would like to thank Bournemouth University for their continued support and Scott Neville from Bluestar Software for his continued insight and for providing us with pattern descriptor details to aid in our analysis.