Enhancing Multimodal Large Language Models for Safety-Critical Driving Video Analysis
About
Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in general visual understanding. However, their application to safety-critical driving scenarios remains limited by an inability to accurately perceive and reason about rare high-stakes dynamic events, such as collisions or near-collisions. To address this, we introduce a pipeline that enhances MLLM perception by fusing downsampled video frames with synchronized high-frequency telematics data (IMU and GPS) and semantic insights from specialized computer vision models. Our pipeline generates high-quality pseudo-labels, including descriptive captions and question-answer pairs, specifically designed to train MLLMs to identify and describe Safety-Critical Events (SCEs) in real-world driving footage. We show the effectiveness of our approach fine-tuning the open-source QwenVL-2.5 model via DoRA adapters: our experiments demonstrate significant improvements in identifying and explaining safety-critical events, with fewer than 50M trainable parameters and limited computational budget.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| 3-class Scene Classification (SCE) | Private dataset video + IMU | Accuracy87.7 | 8 | |
| Captioning | Private dataset video + IMU | ROUGE-L0.44 | 8 | |
| Binary Safety-Critical Event (SCE) identification | Private dataset video + IMU | Precision (Positive Class)93.3 | 8 | |
| Binary Safety-Critical Event (SCE) identification | Nexar video only | Precision (y=1)85.6 | 8 | |
| Closed QA | Private dataset video + IMU | Accuracy92.7 | 8 | |
| Open-ended Question Answering | LingoQA | ROUGE-L28 | 8 |