2018
Heike Hofmann
Charleston, WV
Automatic matching algorithms are only as good as the data that they are trained on. Unfortunately most of the data that is available for training algorithms based on 3D scans are available through NIST’s Ballistics Toolmark Research Database. This means that most automatic methods are only training on the ten barrels from Hamby’s famous study. So the questions remains; how well are these algorithms performing when they are applied to bullets that are not part of the Hamby set, have different calibers, were shot from different brand and makes of handguns.
Last year, at AFTE in Denver, the Center for Statistics and Applications in Forensic Evidence (CSAFE) presented a method for automatic matching of bullet striation marks. We asked practitioners to help with validating the automated matching algorithm. Over the last year we got a chance to work with forensic centers and police departments all over the country. This has given us much insight into how well the matching algorithm is working and what kind of limitations we are encountering.
Our findings are encouraging and seem to match experiences from practitioners; some brands of handguns are more challenging than others to match. Coated bullets are not really suited for matching and polygonal rifling is, at the moment, too challenging for the algorithm because of the lack of well- defined lands. What is important is that all these findings are backed up by scores quantifying the strength of a match (or lack there of).”