Order picking is one of the most important tasks in modern warehouses, while augmented reality (AR) is a friendly alternative by conveying manual picking information into visual guidance to improve the picking performance. Nevertheless, there is still a lack of scalable and systematic consensus to deploy the pick-by-AR method associated with actual warehouse workplaces. This article describes a comprehensive methodology for the use of wearable AR in the order picking process. To enable a robust AR assisted manual order picking system in a warehouse floor, the marker-based global map about the actual shelves is established automatically, empowering a scalable, low-cost and long-term tracking capacity for pick-by-AR within the warehouse floor and allowing the operator to move freely in the actual workplace. The system is evaluated in the automobile assembly line, and experimental results illustrate that the proposed work can provide a systematic solution for pick-by-AR applications in actual warehouse fields.