Automating Crash Diagram Generation Using Vision-Language Models: A Case Study on Multi-Lane Roundabouts
Organizations: Smart Mobility and Infrastructure Lab, College of Engineering University of Georgia Athens
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.
Figures & tables
| Code | Description |
|---|---|
| 1 | Left Front Corner |
| 2 | Front Center |
| 3 | Right Front Corner |
| 4 | Right Front Fender |
| 5 | Right Side Door |
| 6 | Right Rear Fender |
| Metric | Description |
|---|---|
| Collision Type Extraction | Correct collision type (e.g., rear-end, angle) is accurately extracted from the report and depicted in the diagram. |
| Vehicle Labeling–V1 | Vehicle 1 is clearly labeled as “V1" in the crash diagram. |
| Vehicle Labeling–V2 | Vehicle 2 is clearly labeled as “V2" in the crash diagram. |
| Collision Location | The crash is depicted in the correct quadrant or at the correct entry/exit point of the roundabout. |
| Collision Point Accuracy | The exact spot of vehicle impact is accurately marked in the diagram. |
| V1 Damage Code–Extraction Accuracy | The damage code for Vehicle 1 is accurately extracted from the crash report. |
| Metric | GPT-4o | Gemini-1.5-Flash | Janus-4o |
|---|---|---|---|
| Collision Type Extraction | 0.95 | 0.70 | 0.33 |
| Vehicle Labeling–V1 | 0.98 | 0.93 | 1.00 |
| Vehicle Labeling–V2 | 0.91 | 0.98 | 1.00 |
| Collision Location | 0.42 | 0.44 | 0.09 |
| Collision Point Accuracy | 0.16 | 0.02 | 0.00 |
| V1 Damage Code–Extraction Accuracy | 0.47 | 0.19 | 0.07 |
Vision-Language Models
Vision-language models connect visual inputs with text for retrieval, question answering, generation, and grounded multimodal reasoning.
See all 736 papers