Driver Intention Prediction Using Text Prompts with In-Cabin and Out-Cabin Cameras

Human-machine collaborative driving is poised to become a cornerstone of intelligent transportation systems, where accurate recognition of driver behavior and intent is crucial for vehicle decision-making and real-time traffic management. Existing methods for driver maneuver recognition, primarily based on 3D convolutional neural networks (3D-CNNs) and Transformers, encounter challenges such as spatiotemporal redundancy, limited spatial perception, and inadequate capture of detailed object motion. To overcome these limitations, we propose a novel text-driven approach utilizing Large Language Models (LLMs) for driver maneuver recognition. By integrating textual semantics from both in-car and external environments, the Contrastive Language–Image Pre-training (CLIP) text encoder enhances the model's comprehension of complex video scenarios, improving both training and inference efficiency while enabling more precise identification of motion dynamics. Experimental results demonstrate that our approach outperforms state-of-the-art methods in driver intention prediction.