Evaluating Augmented Reality and Computer Vision for Crime Scene Investigation

Sammanfattning: When investigating a crime, investigators often need to revisit the crime scene to get a better understanding of the crime. However, all traces of crime and objects of interest are quickly removed from the scene. Using augmented reality, one can allow investigators to visit crime scenes in their original state readily. Importantly, as the solution is designed to be used within the judicial system, accuracy is crucial. The thesis evaluates different positioning methods for virtual objects by comparing against ground truth positions as well as the perceived movement of the objects. The findings show that in specific scenarios, anchored positioning methods are preferred. Moreover, reconstructing a crime scene is a very tedious process. To simplify the process the paper presents a novel method for raising a 2D bounding box to 3D. The results show a poor estimation of 3D bounding boxes, with mean IoU values for a cup and bottle being 0:3 and 0:1 respectively. However, it is noted that the system is adequate to assist in reconstruction a crime scene, but not replace it.