When doing traffic surveillance, what should be the first and most important point is often overlooked: Where should the camera be placed? For this new paper, we have captured a unique traffic dataset in which the same scene has been captured by both a drone-mounted camera and a camera mounted on an existing pole near the intersection. This is extremely useful for assessing the differences between the two viewpoints.
Furthermore, we have painstakingly marked (annotated) every single cyclist, car, truck and bus in the data. We have not only marked where the objects are, but have annotated them to pixel-level precision. This means that we know the exact shape of every object. This data can be used by other researchers to evaluate how well their detection and classification systems work on pictures from different vantage points.
To learn more, see the full paper Presenting the Multi-view Traffic Intersection Dataset (MTID): A Detailed Traffic-Surveillance Dataset, by Morten B. Jensen, Andreas Møgelmose and Thomas B. Moeslund, published at this year’s Intelligent Traffic Systems Conference.