[Paper Review] A Solution for Crime Scene Reconstruction using Time-of-Flight Cameras

Silvio Giancola, Daniele Piron+2 more|arXiv (Cornell University)|Aug 7, 2017

Robotics and Sensor-Based Localization15 references3 citations

TL;DR

This paper presents a real-time 3D crime scene reconstruction system using a Kinect V2 Time-of-Flight (TOF) RGB-D camera and a SLAM pipeline based on 3D keypoints. By combining colorimetric and geometric features for keypoint description and matching, the method achieves sub-centimeter accuracy (mean error ~1 cm) in both indoor and outdoor environments, offering a portable, low-cost alternative to laser scanners with robust performance under poor lighting.

ABSTRACT

In this work, we propose a method for three-dimensional (3D) reconstruction of wide crime scene, based on a Simultaneous Localization and Mapping (SLAM) approach. We used a Kinect V2 Time-of-Flight (TOF) RGB-D camera to provide colored dense point clouds at a 30 Hz frequency. This device is moved freely (6 degrees of freedom) during the scene exploration. The implemented SLAM solution aligns successive point clouds using an 3D keypoints description and matching approach. This type of approach exploits both colorimetric and geometrical information, and permits reconstruction under poor illumination conditions. Our solution has been tested for indoor crime scene and outdoor archaeological site reconstruction, returning a mean error around one centimeter. It is less precise than environmental laser scanner solution, but more practical and portable as well as less cumbersome. Also, the hardware is definitively cheaper.

Motivation & Objective

• To develop a practical, portable, and cost-effective solution for 3D reconstruction of crime scenes and archaeological sites.
• To address the limitations of traditional 2D sketches and bi-dimensional measurements in preserving scene integrity and enabling dynamic analysis.
• To improve reconstruction accuracy and robustness under poor lighting and in outdoor conditions compared to earlier RGB-D SLAM systems.
• To enable dynamic scene exploration and virtual navigation for forensic analysis, such as trajectory reconstruction and blood pattern analysis.
• To provide a real-time, on-the-fly reconstruction method that supports loop closure and minimizes drift in SLAM pipelines.

Proposed method

• Utilizes the Kinect V2, a TOF-based RGB-D camera, to capture dense, colored point clouds at 30 Hz with 512×424 resolution.
• Employs a SLAM pipeline that processes successive point clouds using 3D keypoints detected on depth frames.
• Describes keypoints using a hybrid C-SHOT descriptor combining colorimetric and geometric features to enhance robustness in low-light and texture-poor environments.
• Aligns point clouds via 3D keypoint matching using RANSAC and ICP for iterative refinement, enabling real-time on-line reconstruction.
• Applies loop closure detection to correct cumulative drift, though the method identifies limitations when bad matches propagate errors.
• Validates results against a high-precision environmental laser scanner (Stonex X300) as ground truth for geometric consistency.

Experimental results

Research questions

• RQ1 Can a TOF-based RGB-D camera like the Kinect V2 achieve sufficient accuracy and robustness for forensic 3D crime scene reconstruction in real-world conditions?
• RQ2 How does the use of 3D keypoints with combined colorimetric and geometric features improve SLAM performance compared to purely color-based or 2D keypoint methods?
• RQ3 To what extent does the proposed SLAM pipeline mitigate drift in reconstruction accuracy during freehand scanning of complex scenes?
• RQ4 How does the system perform in terms of geometric consistency (e.g., planarity, perpendicularity) compared to high-precision laser scanning?
• RQ5 What trade-offs exist between cost, portability, and accuracy when using consumer-grade TOF cameras versus professional laser scanners in forensic applications?

Key findings

• The proposed system achieves a mean reconstruction error of approximately 1 cm in both indoor crime scene and outdoor archaeological site scenarios.
• The system demonstrates robustness in low-light conditions due to the use of colorimetric and geometric features in the C-SHOT descriptor, reducing reliance on texture.
• The Kinect V2 TOF camera enables reliable depth sensing in outdoor environments, unlike earlier structured-light sensors such as Kinect V1.
• Geometric consistency evaluation shows a mean error of 10.80 mm for the SLAM reconstruction versus 3.04 mm for the laser scanner ground truth, with up to 3° angular deviation in wall perpendicularity.
• Despite drift issues from poor keypoint matching, loop closure helps reduce global error, though it can also propagate distortions if initial matches are incorrect.
• The solution is significantly more portable and cost-effective than environmental laser scanners, making it practical for field deployment in forensic investigations.