Automating Crash Diagram Generation Using Vision-Language Models: A Case Study on Multi-Lane Roundabouts
Summary: arXiv:2604.15332v1 Announce Type: cross
Abstract: Crash diagrams are essential tools in transportation safety analysis, yet their manual preparation remains time-consuming and prone to human variability. This study investigates the use of Vision-Language Models (VLMs) to automate crash diagram generation from police crash reports, focusing on multilane roundabouts as a challenging test case. A three-part structured prompt framework was developed to guide model reasoning through interpretation, extraction, and visual synthesis, while a 10-metric evaluation system was designed to assess diagram quality in terms of semantic accuracy, spatial fidelity, and visual clarity.
Three popular models, including GPT-4o, Gemini-1.5-Flash, and Janus-4o, were tested on 79 crash reports. GPT-4o achieved the highest average performance (6.29 out of 10), followed by Gemini-1.5-Flash (5.28) and Janus-4o (3.64). The analysis revealed GPT-4o’s superior spatial reasoning and alignment between extracted and visualized crash data. These results highlight both the promise and current limitations of VLMs in engineering visualization tasks. The study lays the groundwork for integrating generative AI into crash analysis workflows to improve efficiency, consistency, and interpretability.
Introduction
Transportation safety analysis heavily relies on crash diagrams to provide visual representations of incidents, which are instrumental in understanding the dynamics of accidents and formulating safety measures. However, the traditional methods for creating these diagrams are often labor-intensive and subject to inconsistencies due to human error. This study explores the potential of Vision-Language Models (VLMs) to streamline this process.
Methodology
The research employed a structured three-part prompt framework designed to enhance the reasoning capabilities of VLMs in generating crash diagrams. The framework consists of:
- Interpretation: Understanding and contextualizing the details from police crash reports.
- Extraction: Identifying critical data points necessary for diagram creation.
- Visual Synthesis: Generating the final visual representation based on extracted data.
Evaluation Metrics
To measure the effectiveness of the generated diagrams, a comprehensive evaluation system was developed. This system includes ten metrics focused on:
- Semantic Accuracy: Ensuring the data represented in the diagrams accurately reflects the incidents described in the reports.
- Spatial Fidelity: Assessing the accuracy of spatial relationships in the visual representation.
- Visual Clarity: Evaluating the overall clarity and interpretability of the diagrams.
Results
The performance of the tested models indicated a significant variation in their abilities to generate crash diagrams. GPT-4o outperformed the others, achieving a score of 6.29, thanks to its advanced spatial reasoning capabilities. In contrast, Gemini-1.5-Flash and Janus-4o scored 5.28 and 3.64, respectively.
Conclusion
This study underscores the potential of VLMs in revolutionizing crash diagram generation. While GPT-4o demonstrated impressive results, the research also highlighted the ongoing challenges in achieving fully reliable automated visualizations. Future work will focus on refining these models and integrating them into standard crash analysis practices to enhance efficiency, consistency, and interpretability.