The key to saving energy in commercial buildings is to deliver building services only when and where they are needed, in the amount that they are needed. Given that building services are usually employed to provide occupants with satisfactory indoor conditions, it is therefore important to accurately detect the occupancy of building spaces in real time. This paper starts with some discussion on building occupancy resolution and accuracy as well as a brief introduction to traditional explicit occupancy detection approaches. The focus of this paper is on the review and classification of emerging, potentially low-cost approaches to leveraging existing data streams that may be related to occupancy, sometimes referred to as implicit / ambient / soft sensing approaches. About 40 related projects / systems are reviewed and compared in terms of occupancy sensing type, occupancy resolution, accuracy, ground truth data collection method, demonstration scale, data fusion and control strategies. It also briefly discusses technology trends, research challenges, and future directions.