The Reliability of Forensic Body-Shape Identification

Neerja Thakkar, Georgios Pavlakos, Hany Farid; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2022, pp. 44-52


Photo-based forensic identification can be critical in the prosecution of, and defense against, criminal charges. Identification techniques range from the specific biometric-based to the more generic, based on height, weight, gender, and race. Although fairly basic, accurate height and weight estimation remains challenging due to physiological factors, concealing clothing, body pose, and the scale ambiguity inherent to the photographic process. We describe an extension to 3D body-pose estimation that more accurately estimates body shape across a broader range of body sizes. We evaluate the reliability of this technique in making metric estimates of height and weight, and in making non-metric categorization of people based on a scale-agnostic measure of body shape. Although this approach improves on previous efforts, we find that accurate body-shape identification from a single, reference-free image remains challenging.

Related Material

@InProceedings{Thakkar_2022_CVPR, author = {Thakkar, Neerja and Pavlakos, Georgios and Farid, Hany}, title = {The Reliability of Forensic Body-Shape Identification}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2022}, pages = {44-52} }