Performance Comparison of Multi-Modal Fusion Techniques in Tissue Perfusion Analysis Using Homography Calibration

This study explores the impact of fusion techniques and camera calibration on tissue perfusion analysis using multi-modal imaging. Building on the existing TTPD dataset collected with a prototype hardware platform, we focus on optimizing data alignment and fusion strategies to improve classification accuracy. The imaging system integrates infrared (IR), thermal, and RGB cameras, capturing complementary information about tissue perfusion. To improve modality alignment, we apply Homography-based calibration, reducing spatial discrepancies between different imaging sources. Furthermore, we evaluated early and late fusion approaches using deep learning models (ResNet50 and ResNet101) to determine the most effective integration strategy. Experimental results demonstrate that late fusion, particularly the combination of thermal and RGB modalities, achieves the highest classification performance and that Homography-based alignment improves results. These findings highlight the importance of precise calibration and modality selection in the development of robust and non-invasive tissue perfusion monitoring systems.