Radar2Text: Generation of Linguistic Summary from mmWave Radar Signatures Using Fine-Tuned Multimodal Language Models

Millimeter-wave (mmWave) sensing technology enables the perception and measurement of the physical world in a non-invasive and privacy-preserving manner. However, radar signals are inherently non-intuitive, requiring specialized expertise to process datasets and interpret system observations. To address these challenges, this paper explores the feasibility of leveraging fine-tuned Multimodal Large Language Models (MLLMs) to extract critical and accurate textual descriptions from radar-generated range history images. By fine-tuning a multimodal foundation model, the study establishes a connection between radar signatures captured when a target moves away and toward a radar and their corresponding textual descriptions. Experimental results demonstrate the effectiveness of the proposed approach, highlighting the potential of mmWave radar technology for natural language-based multimodal target tracking.