计算机视觉
计算机科学
人工智能
目标检测
对象(语法)
模态(人机交互)
噪音(视频)
组分(热力学)
特征(语言学)
钥匙(锁)
工程类
模式(计算机接口)
作者
Yiming Li,Tianyi Liu,Kim–Hui Yap
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-04-21
卷期号:: 8777-8781
标识
DOI:10.1109/icassp55912.2026.11461976
摘要
In-cabin monitoring is vital for intelligent transportation systems, and robust object detection is crucial for safety and intelligence. Yet cabin environments remain challenging due to clutter, occlusions, varying illumination, and subtle object appearances. Basic single-modality solutions struggle with these issues, and even conventional multi-modal fusion methods often yield limited performance without accounting for modality degradations, which may cause misleading fusion. We propose the Degradation-Aware Modality Interaction Network (DeMoNet), which consists of two key modules. The Degradation-Aware Modality Interactor (DMI) enhances the representations of weak modalities via cross-modal interaction guided by modality-specific degradation prompts. The Mixture-of-Modality Experts (MoME) adaptively fuses features at a fine spatial granularity. To facilitate evaluation, we build the Multi-Modal Car Cabin (MC2) benchmark with aligned multi-modal annotations. Experiments on MC2demonstrate that DeMoNet consistently outperforms single-modal baselines and conventional fusion approaches, validating the benefits of our design for robust in-cabin monitoring.
科研通智能强力驱动
Strongly Powered by AbleSci AI