A consistent methodology for forensic photogrammetry scanning of human remains using a single handheld DSLR camera

Abstract

Abstract   Due to increasingly capable algorithms and more available processing power, photogrammetry is becoming a simple, cheap, and accurate alternative to 3D optical surface scanning. With adequate application, it can be a swift documentation technique for reconstructing the geometry and body surface of deceased persons in autopsies or other forensic medical examinations. Sufficiently easy and swift 3D documentation techniques may allow 3D imaging technologies to become part of the daily routine of any forensic medical examiner or other medical personnel. This paper presents a consistent and systematic photographing methodology (as an alternative to automated or intuitive methods) for photogrammetry scanning of human remains. Although it requires manual photography, the methods presented in this paper offer a swift and easy way to capture an accurate 3D model of human remains under almost any conditions. Four different photographing procedures were tested on four subjects: (i) a systematic circular technique with 100 photos, (ii) a systematic circular technique with 50 photos, (iii) a technique loosely mimicking cameras mounted on a postmortem CT device with 98 photos, and (iv) a technique mimicking cameras mounted on a postmortem CT device with 49 photos. Measurement accuracy was tested with the aid of six adhesive control points placed at approximately the same locations on each subject. Five different distances defined by these control points were measured and compared to the measurements taken by hand. 3D photogrammetry meshes created using these techniques were also compared with point clouds acquired using a 3D laser scanner. We found that a carefully composed, tested, and systematic photographing procedure significantly improved the quality of the photogrammetry models. In terms of relative difference compared to the hand measurements, both Techniques 1 and 2 produced close results, with an average relative difference of 0.160% and 0.197% and a maximum relative difference of 0.481% and 0.481%, respectively, while models reconstructed from images taken using Techniques 3 and 4 seemed to be much less accurate, with an average relative difference of 0.398% and 0.391% and a maximum relative difference as high as 1.233% and 1.139%, respectively. This study highlights the importance of a scientifically tested methodology for obtaining high-quality 3D models in forensic applications. Key points