VLMPed-CoT: A large vision-language model with a chain-of-thought mechanism for pedestrian crossing intention prediction
Yancheng Ling et al.
What the paper says
Pedestrian crossing intention prediction is crucial for autonomous driving. While existing models have achieved high accuracy, their generalization and robustness remain limited, hindering their performance in real-world scenarios. To overcome these limitations, we introduce the LVLMPed-CoT, a large vision language model (LVLM) that incorporates a chain-of-thought (CoT) mechanism to enhance pedestrian crossing intention prediction. It takes multimodal data as input and employs data distillation along with a two-stage fine-tuning strategy to elicit the implicit CoT capability of a lightweight vision-language model for enhanced perception, reasoning, an d prediction. The unified LVLMPed-CoT is trained on a joint open-source dataset (JAAD and PIE) and achieves superior or comparable performance to state-of-the-art models on both large-scale public datasets. The ablation study validates the contribution of the CoT prompt design and the two-stage fine-tuning strategy to the model's performance. Further analysis investigates the impact of input data sequence length and image quality on both accuracy and inference time, as well as the interpretability of the enhanced CoT reasoning ability achieved through fine-tuning.
Evidence weight
Balanced mode · F 0.40 / M 0.15 / V 0.05 / R 0.40
| F · citation impact | 0.50 × 0.4 = 0.20 |
| M · momentum | 0.50 × 0.15 = 0.07 |
| V · venue signal | 0.50 × 0.05 = 0.03 |
| R · text relevance † | 0.50 × 0.4 = 0.20 |
† Text relevance is estimated at 0.50 on the detail page — for your query’s actual relevance score, open this paper from a search result.