Crime Scene Reconstruction Using 3D Convolutional Neural Networks (3D-CNNs)

In this study, the role of 3D Convolutional Neural Networks (3D-CNNs) is investigated for CS reconstruction considering the improvement of forensic investigation. The classical approaches to crime scene analysis and visualization can merely involve an analysis of images and require a lot of time for the preparation of a reconstructed 3D model. In contrast, 3D-CNNs are able to consider volumetric data, like 3D-scan data or multiple view images, and capture a vast amount of information of the crime scene to construct an exact 3D model from multiple views. It comprises of the collection of high-resolution 3D data, followed by preprocessing it so as to feed it into the requested network, and training the 3D-CNN through the supervised learning. The findings have shown the effectiveness of the proposed 3D-CNN approach as compared with the conventional methods that were used for subsequent steps of the scene reconstruction with the better accuracy and time effective in reconstruction of the walls and objects and evidence markers. The research also defines the computational complexity of its method and shows that it can be applied in real-time investigations of forensic cases. Therefore the findings indicate that 3D-CNNs present a leap forward the application of CSF crime scenes reconstruction presenting a more objective and accurate tool to forensic expert findings for the enhancement of crime scene analysis in future investigations.