A first approach for the contactless acquisition and automated detection of toolmarks on pins of locking cylinders using 3D confocal microscopy

Lockpicking forensics is currently a completely manual process, which requires a lot of skill as well as training and is therefore time-consuming as well as expensive. In this paper we make a first move to transfer the most crucial part of this specific forensic process, the contactless aquisition and analysis of traces on locking pins, into the domain of digitized forensics. To do so, we introduce a new five stage processing methodology for semi- or fully-automated lockpicking forensics. Our methodology consists of: trace positioning, acquisition, detection of traces with segmentation (or region of interest determination), determination of the trace type and the determination of the used opening methods. Within this pipeline the last three stages constitute a hierarchy of pattern recognition (PR) problems. In this paper we propose a solution approach for the Trace Positioning, Contacless Acquisition and the first of the three PR problems - the detection of traces with segmentation (or region of interest determination) on which the other two are depending. To implement this segmentation, we use texture recognition with gray-level-co-occurrence matrices to blockwisely describe the texture imposed by toolmarks with adequate features. By that we are able to distinguish between regions including potential traces and regions without relevant traces. With our presented approaches for an automated contactless acquisition and trace detection, we support and improve the classical manual forensic investigation in the field of lockpicking forensics in regards of effort, objectivity and reliability. Additionally, it creates a solid base for future work dealing with trace type determination and opening method classification. We evaluate our approach with a physical test set of 15 lock pins, from locks opened with three different opening methods. On this limited test set our approach achieves True Positive Rates of up to 85% for the detection of potentially trace wielding regions. This first result, although it still leaves room for improvement, constitutes and shows a positive tendency for a seminal first step towards semi- or fully-automated lockpicking forensics.