Enhancing Crime Scene Investigations: From technology to evidence

Crime scene investigation is a foundational element of the criminal justice process, yet it remains largely dependent on manual methods such as photography, sketching, and written descriptions. These techniques are constrained by human perception, time pressure, and the inability to revisit a scene once it has been released. At the same time, advanced technologies including three-dimensional (3D) reconstruction, multi-modal sensing, and artificial intelligence (AI) have matured considerably in recent years. Their integration into investigative practice, however, is hindered by high costs, insufficient training, institutional resistance, and a persistent gap between public expectations and actual capabilities.

This dissertation investigates how these emerging technologies can be responsibly and affordably integrated into crime scene investigation, ensuring that they add evidential value without compromising accuracy, interpretability, or trust. The research is structured around five interlinked objectives, progressing from conceptual foundations through technical evaluation and applied analysis to ethical and societal reflection.

Conceptual and technological foundations

The dissertation begins by establishing the conceptual landscape of modern crime scene investigation. Through a series of peer-reviewed publications, the role of 3D reconstruction is examined not merely as a technical tool but as a socio-technical system that transforms how evidence is documented, analysed, and communicated.

This analysis extends to advanced sensing technologies including depth cameras, LiDAR, thermal imaging, hyperspectral imaging, ground-penetrating radars, electronic noses, and unmanned aerial vehicles, which enrich scene documentation with data layers beyond human visual perception.

A further contribution addresses the CSI effect, the phenomenon whereby fictional media portrayals inflate public expectations of forensic technology. Through narrative-driven, interactive workshops with adult learners, this research demonstrates that targeted educational interventions can reduce exaggerated expectations and foster realistic technological literacy, thereby contributing to more informed science communication in the forensic domain.

Technical foundations of 3D reconstruction

The technical core of the dissertation focuses on 3D reconstruction, particularly Neural Radiance Fields (NeRF) and Gaussian Splatting. A comprehensive review first documents how environmental factors such as shadows, specular reflections, exposure imbalance, and colour-temperature shifts degrade photometric consistency across both classical geometry-based pipelines and modern neural-rendering approaches.

Controlled experiments then identify optimal camera settings—an ISO of 200, a shutter speed of 1/60 s, and an aperture of f/3.5—as the best compromise between noise suppression and faithful detail reproduction. Further experiments establish that filming in landscape mode at a slow walking speed of approximately 0.09 m/s, capturing three vertical layers, and dividing the environment into three spatial zones before processing maximises splat density and minimises artefacts.

A direct side-by-side comparison of NeRF and Gaussian Splatting across indoor and outdoor environments demonstrates that Gaussian Splatting outperforms NeRF on all evaluated criteria, exhibiting lower reconstruction noise, higher detail preservation, and substantially shorter processing time.

A complementary engineering contribution delivers a CPU-based Gaussian Splatting thumbnail generator that achieves a 156-fold speed-up, enabling forensic investigators to obtain immediate visual previews without specialised hardware.

Artificial intelligence as a decision-support layer

Beyond reconstruction, the dissertation examines how AI can support crime scene investigation at the level of interpretation and decision support. One contribution demonstrates how AI-supported 3D reconstructions enable spatial evidence contextualisation, improved collaboration through remote sharing, and enhanced analytical tasks such as trajectory reasoning, with a particular focus on firearms trafficking investigations.

A second contribution addresses AI systems for generating plausible crime scene scenarios, going beyond measurement and classification into the domain of narrative construction. These scenario-generation systems are conceptualised as tools for facilitating investigative reasoning, synthesising available evidence and constraints to represent different scenarios for analysis, comparison, and review, while remaining grounded in the evidence and conveying uncertainty rather than supplanting human judgment.

Ethical, operational, and societal dimensions

A systematic ethical analysis of AI-generated crime scene scenarios addresses concerns including bias, accountability, explainability, privacy, and narrative framing, and proposes an ethics-by-design framework with human-in-the-loop control, transparency, auditability, and participatory system design as core elements.

A parallel investigation examines the ethical constraints specific to 3D reconstruction, addressing issues of realism bias, persuasive effects in courtroom presentation, data ownership, and the admissibility of reconstructed environments as evidence. These analyses are situated within the broader paradigm of data-driven and evidence-based policing, exploring how principles of transparency, traceability, proportionality, and continuous evaluation can govern the responsible integration of AI-enabled crime scene technologies.

Main findings and the way forward

Three cross-cutting findings emerge. First, accuracy in 3D reconstruction is conditional, not absolute: it depends on environmental conditions, camera settings, capture strategy, and algorithmic choices, making a reconstruction’s evidential reliability inseparable from the transparency of the process that produced it.

Second, AI-driven systems can generate alternative hypotheses, flag overlooked evidence, and structure complex datasets, but they are interpretive mediators that encode assumptions about relevance, causality, and narrative structure; uncritical acceptance risks automation bias and narrative reinforcement.

Third, trust in technologically mediated evidence requires institutional scaffolding, including transparency, interpretive literacy, organisational governance, and alignment with professional and legal standards, rather than technical performance alone.

The transition from technological capability to evidential integrity is therefore a socio-technical challenge requiring simultaneous attention to accuracy, interpretation, and trust. If technological systems are treated as aids to reasoning rather than substitutes for it, they can counter tunnel vision and support the transparent, evidence-based practice that the criminal justice system requires.