Abnormal phenomena on urban roads, including uneven surfaces, garbage, traffic congestion, floods, fallen trees, fires, and traffic accidents, present significant risks to public safety and infrastructure, necessitating real-time monitoring and early warning systems. This study develops Urban Road Anomaly Visual Large Language Models (URA-VLMs), a generative AI-based framework designed for the monitoring of diverse urban road anomalies. The InternVL was selected as a foundational model due to its adaptability for this monitoring purpose. The URA-VLMs framework features dedicated modules for anomaly detection, flood depth estimation, and safety level assessment, utilizing multi-step prompting and retrieval-augmented generation (RAG) for precise and adaptive analysis. A comprehensive dataset of 3034 annotated images depicting various urban road scenarios was developed to evaluate the models. Experimental results demonstrate the system’s effectiveness, achieving an overall anomaly detection accuracy of 93.20%, outperforming state-of-the-art models such as InternVL2.5 and ResNet34. By facilitating early detection and real-time decision-making, this generative AI approach offers a scalable and robust solution that contributes to a smarter, safer road environment.

Persistent multimodal conversational memory requires unified access to textual and visual evidence across sessions, yet differences between image content and textual queries make relevant sources harder to locate. We propose DualBridgeMem, a structured visual semantic consolidation method that generates long and short visual descriptions at memory-writing time using a nine-aspect observation checklist, preserving detailed facts and compact discriminative information, respectively. The descriptions, source text, and original-media references are packaged into MemoryNotes for evidence retrieval through a shared semantic and lexical scoring interface, with access to original images and source traceability. We evaluate the method on LoCoMo and Memory-QA through retrieval-input ablations, comparisons with original-image embedding baselines, and paired statistical analysis. Relative to source text alone, the configuration combining source text with long and short visual descriptions has a higher point estimate for each of R@1, R@10, and answer F1 on both datasets. The respective differences are 7.14, 7.30, and 1.56 percentage points on LoCoMo, and 1.53, 0.45, and 1.14 percentage points on Memory-QA. Comparisons with original-image embedding baselines, including image-only CLIP, further support the effectiveness of this retrieval path for evidence localization. These findings provide empirical support for structured visual descriptions as an interpretable and traceable retrieval interface for persistent multimodal memory.