[Paper Review] 3D Reconstruction of Crime Scenes and Design Considerations for an Interactive Investigation Tool

Erkan Bostancı|arXiv (Cornell University)|Dec 10, 2015

3D Surveying and Cultural Heritage22 references17 citations

TL;DR

This paper proposes an automated 3D reconstruction method for crime scenes using video sequences and computer vision techniques, specifically keyframe extraction and bundle adjustment-based dense reconstruction. The approach generates metric, photo-realistic 3D point clouds from real camera footage, enabling accurate spatial analysis and forming the basis for an interactive forensic investigation tool with measurable, court-admissible results.

ABSTRACT

Crime Scene Investigation (CSI) is a carefully planned systematic process with the purpose of acquiring physical evidences to shed light upon the physical reality of the crime and eventually detect the identity of the criminal. Capturing images and videos of the crime scene is an important part of this process in order to conduct a deeper analysis on the digital evidence for possible hints. This work brings this idea further to use the acquired footage for generating a 3D model of the crime scene. Results show that realistic reconstructions can be obtained using sophisticated computer vision techniques. The paper also discusses a number of important design considerations describing key features that should be present in a powerful interactive CSI analysis tool.

Motivation & Objective

• To develop an automated, cost-effective method for creating accurate 3D reconstructions of crime scenes using standard video footage instead of expensive laser scanners.
• To improve forensic investigation by enabling ex-situ spatial analysis through realistic, metric 3D models derived from real photographic evidence.
• To define essential design features for an interactive 3D analysis tool that supports measurement, annotation, and visualization for investigators.
• To ensure the reconstructed 3D models are suitable for court use by maintaining metric accuracy and traceability to original photographic evidence.
• To explore integration of 3D reconstructions into immersive environments such as virtual or augmented reality for enhanced investigative insight.

Proposed method

• Keyframes are extracted from a video sequence using feature-based detection to ensure sufficient motion and feature correspondence between frames.
• Dense 3D reconstruction is performed using an iterative bundle adjustment technique to estimate camera poses and 3D point positions from selected keyframes.
• The scene is divided into image clusters to manage computational load, and individual point clouds from each cluster are reconstructed separately.
• Misaligned point clouds are realigned using the Iterative Closest Point (ICP) algorithm with at least four corresponding points to achieve global consistency.
• The final 3D point cloud is processed using third-party software for noise reduction, mesh generation, and texture mapping from original images.
• The system supports export of detailed reports, annotations, and measurement data for archival and evidentiary purposes.

Experimental results

Research questions

• RQ1 Can a 3D reconstruction of a crime scene be accurately generated from standard video footage using computer vision techniques without relying on expensive laser scanners?
• RQ2 How can keyframe selection be optimized to maintain sufficient feature correspondence and reduce motion blur artifacts in video-based reconstruction?
• RQ3 What are the essential design features required in an interactive 3D forensic analysis tool to support investigative reasoning and measurement?
• RQ4 To what extent can the reconstructed 3D model maintain metric accuracy and serve as admissible evidence in court?
• RQ5 How can 3D reconstructions be integrated into immersive environments such as VR or AR to enhance investigator understanding of spatial relationships?

Key findings

• The proposed method successfully generates a metric, photo-realistic 3D point cloud of a crime scene from video footage using keyframe extraction and bundle adjustment.
• The final alignment of multiple point clouds achieved a mean error bound of 0.0010 mm using the ICP algorithm, indicating high geometric accuracy.
• The reconstruction process is automated and based on real photographic evidence, which enhances its credibility and potential admissibility in court.
• The system supports critical forensic operations such as 3D distance measurement between objects and victims, with results that can be annotated and reported.
• The integration of 3D models with mesh generation, texture mapping, and annotation tools enables comprehensive ex-situ analysis of crime scenes.
• Future work confirms the feasibility of extending the system to support virtual and augmented reality interfaces for immersive investigation.