<p>With the proliferation of intelligent systems, multi-sensor information fusion has become a key enabler for enhancing perception and reliability. This paper presents a structured review of fusion technologies, spanning theoretical foundations from classical probabilistic models to modern deep learning and hybrid approaches. We systematically classify fusion strategies into data, feature, and decision levels, comparing them in terms of integrity, robustness, and computational cost. Representative applications in autonomous driving, smart healthcare, and predictive maintenance demonstrate how sensor complementarity overcomes the limitations of single-source perception. A core contribution of this work is the identification of five key reference indicators including data homogeneity, dimensionality, environmental complexity, real-time requirements, and data sparsity to guide the selection of optimal fusion models for specific engineering scenarios. Unlike previous reviews, this study connects theoretical insights with practical challenges, addressing issues such as data standardization and model interpretability. It serves as a technical guide for researchers aiming to design adaptive, resource-efficient, and trustworthy multi-sensor fusion systems.</p>