A deep learning approach to real-time video analytics for people and passenger counting

Recent advances in computer vision and deep learning are driving unprecedented interest in the use of video analytics as a key component in IoT solutions for intelligent transport systems and smart city infrastructure. In particular, the development of real-time video analytics algorithms to support the analysis of passenger counting and patronage analysis, deployable “on the edge” and capable of supporting robust and accurate performance under varying environmental conditions, is emerging as a key area of focus. This chapter presents a state-of-the-art review of real-time video analytics for people and passenger counting, highlighting the latest developments and future directions for this research. Focus is given to the emergence and recent developments in deep learning techniques for the analysis of video, and in particular those designed for deployment on the edge. Further, case studies are presented showing how this research is being applied in the context of real-world, public transport passenger analytics problems, specifically to support demand prediction for and analysis of rail replacement bus patronage, and for acquiring real-time public space patronage statistics. These case studies demonstrate how robust, fast and accurate video analytics are being achieved on low-cost edge computing hardware to support real-time patronage analytics. The chapter elucidates key methodological and practical challenges to consider when deploying and evaluating such systems in the field.