Research on Object Detection Algorithms Based on Multimodal Images

Object detection, as a core task within computer vision, holds irreplaceable application value in critical domains such as autonomous driving environmental perception and intelligent security surveillance. Monomodal RGB image detection methods, reliant solely on visible light information, often suffer from insufficient feature extraction and confusion between objects and backgrounds in suboptimal scenarios like low light or complex backgrounds, resulting in inadequate detection robustness.To address this limitation, two RGB-thermal infrared multimodal fusion detection algorithms are proposed: the Cross-modal Attention Enhancement Network employs a joint channel-position attention mechanism to extract deep-level correlated features across modalities, while incorporating a recurrent enhancement strategy to optimise the representation of object edge details;Progressive Feature Fusion Network, which employs a symmetric dual-branch architecture to enhance thermal infrared modality weighting, and adopts a strategy combining coarse-to-fine granularity perception with fine-grained fusion to achieve efficient hierarchical feature aggregation.