<p>The growth of digital technologies, sensors, and interconnected devices has led to an explosion of data from multiple sources and perspectives, giving rise to multi-view data. Multi-view data refer to information about the same underlying phenomena or entities collected from different sources, perspectives, or modalities. These data provide a more comprehensive and nuanced understanding of complex systems or situations. For instance, modern security surveillance systems utilize heterogeneous cameras and sensors. These devices collaboratively capture video, analyze data, and share knowledge to provide effective surveillance services. Effective coordination and control in these systems are highly challenging and are currently receiving substantial attention for research and relevance to accomplish different security surveillance activities. This study is a survey paper that seeks to present current datasets on the topic of multi-view learning in a comprehensive way by categorizing them according to their applications. We first aim to establish a new taxonomy to classify multi-view data, including different sources, data types, and modalities. Moreover, this survey presents up-to-date multi-view learning datasets, points out their shortcomings, as well as directs experts in selecting the best datasets for algorithm benchmarking, and makes recommendations for future research. To deepen the understanding of existing datasets, this survey conducts a critical analysis of design trade-offs, cross-modal annotations, and cross-domain comparisons, offering researchers a strategic roadmap for dataset selection and future benchmark development. Another aspect of the study involves identifying specific unresolved problems that could potentially advance multi-view machine learning research. The datasets, corresponding links, and supporting documentation associated with this survey are available in the companion repository: <a href="https://github.com/cv-mines/multi-view-datasets-survey">Multi-View Datasets Survey</a>.</p>