Enhancing Crime Scene Investigations through Virtual Reality and Deep Learning Techniques
Antonino Zappalà,Luca Guarnera,Vincenzo Rinaldi,Salvatore Livatino,Sebastiano Battiato
2024-09-27
Abstract:The analysis of a crime scene is a pivotal activity in forensic investigations. Crime Scene Investigators and forensic science practitioners rely on best practices, standard operating procedures, and critical thinking, to produce rigorous scientific reports to document the scenes of interest and meet the quality standards expected in the courts. However, crime scene examination is a complex and multifaceted task often performed in environments susceptible to deterioration, contamination, and alteration, despite the use of contact-free and non-destructive methods of analysis. In this context, the documentation of the sites, and the identification and isolation of traces of evidential value remain challenging endeavours. In this paper, we propose a photogrammetric reconstruction of the crime scene for inspection in virtual reality (VR) and focus on fully automatic object recognition with deep learning (DL) algorithms through a client-server architecture. A pre-trained Faster-RCNN model was chosen as the best method that can best categorize relevant objects at the scene, selected by experts in the VR environment. These operations can considerably improve and accelerate crime scene analysis and help the forensic expert in extracting measurements and analysing in detail the objects under analysis. Experimental results on a simulated crime scene have shown that the proposed method can be effective in finding and recognizing objects with potential evidentiary value, enabling timely analyses of crime scenes, particularly those with health and safety risks (e.g. fires, explosions, chemicals, etc.), while minimizing subjective bias and contamination of the scene.
Computer Vision and Pattern Recognition