Crime Scene Reconstruction with RGB-D Sensors

Abstract

Photographic surveying, a fundamental procedure in crime investigation, is typically performed using 2D cameras. Although useful, such cameras remain limited due to the lack of depth information. In this work, we propose a 3D reconstruction solution that leverages the advantages of cheap RGB-D sensors to create a 3D model of the crime scene and to provide the investigator with an interactive crime scenario simulation environment. A structure from motion approach is proposed in order to align the captured point clouds on each other using 3D key points. An iterative refinement and a global optimization algorithm are later adapted for the optimization of the registered 3D model, which is then triangulated before the underlying surface is reconstructed. The resulting model is used for interactive crime investigation and object dynamics simulation. The obtained results show the elrectiveness of our solution with a visually appealing rendering, an accurate simulation and a quantitative error of less than 18cm for the $4m \times 4m$ indoor scene. An accompanying video is provided in order to illustrate the processing pipeline1.