This paper presents the ACFR five roundabouts dataset, a collection of over 23,000 vehicle trajectories of real world naturalistic driving at unsignalized intersections. This data was taken using a lidar based perception and tracking system onboard a stationary vehicle near the intersections. The focus of this dataset is single lane roundabouts, as these unstructured intersections have no signals dictating which vehicle has right-of-way, meaning that there is considerable negotiation between vehicles. The method for data capture is presented, as well as a study of the collected data including data preprocessing techniques and data analytics. The data preprocessing techniques are focused around preparing the data for machine learning training. The extensiveness of this dataset makes it essential to validate algorithms for state of the art ADAS and autonomous vehicles. The paper concludes with a focus study of some of the outliers in the data.