Efficient sensing on resource-limited platforms is a hot research topic in the field of information processing. Different from conventional array image acquisition, single-pixel data recording and compressed sensing-based image reconstruction effectively reduce the bandwidth, but the reconstructed images generally contain many data irrelevant for high-level vision tasks. Single-pixel sensing is an emerging technique that directly infers high-level semantics from one-dimensional encoded measurements without multidimensional image reconstruction. Compared with the conventional first-reconstruction-then-perception scheme, the sensing efficiency is greatly improved. It has broad applications in many fields, such as remote sensing, intelligent transportation, biomedicine, and the national defense military. This overview focuses on the history and development of single-pixel sensing and introduces its technical architecture and research progress in computer vision applications. Finally, we outlook the development trends, hoping to provide some highlights for future studies in this direction.