Traffic research requires detailed empirical knowledge grounded in observations from real traffic environments. Such naturalistic data enable studies of traffic flow, road user behaviour, and interactions as they occur under real-world conditions, unaffected by experimental setups or the data collection process itself. This type of empirical information is central to traffic analysis and to the development of methods for modelling and simulation of the traffic system, which in turn can inform traffic planning and infrastructure design. Video-based observation — from unmanned aerial vehicles (UAVs) or fixed cameras mounted on the side of the road — makes it possible to observe traffic without influencing road user behaviour. UAVs offer broad spatial coverage, enabling analysis of processes that unfold over longer road stretches, such as overtaking manoeuvres, platoon formation, and lane-changing behaviour. Fixed cameras are more limited in spatial coverage but provide higher positional precision. Using object detection and tracking algorithms, the position of visible road users can be followed over time, enabling extraction of detailed trajectories from which traffic analytical measures can be derived. This application concerns the automated detection and tracking of road users, including pedestrians, cyclists and other micromobility modes, motorcycles, cars, buses, and trucks, from high-resolution video footage collected at both urban and rural traffic environments in Sweden. The work supports several ongoing research projects at VTI aimed at improving the understanding of road user behaviour and traffic flow through empirical data collection and analysis. Video data is collected from two sources: drones (UAVs) and fixed cameras, both recording at resolutions up to 4K. Recording sessions range from approximately 1 to 3 hours of traffic per location per day, and in some sessions multiple cameras operate simultaneously. Data collection is planned primarily between August and October 2026, covering a variety of urban and rural locations with drones, and a few fixed-camera deployments. In total, the dataset across ongoing projects comprises approximately 80 hours of raw traffic video, with an estimated total storage of around 2,500–3,200 GiB. All video material is collected from public spaces in accordance with applicable Swedish regulations. The processing pipeline is based on YOLO (You Only Look Once) deep learning models for object detection, followed by multi-object tracking to produce trajectories for each detected road user. The pipeline also involves georeferenced output, i.e., mapping detections to real-world coordinates, which adds additional computational steps. While specific implementations vary across projects, the overall processing pipeline is consistent across all datasets. The resulting trajectory data supports ongoing research on several topics across multiple modes and environments. Topics include, but are not limited to, queue formation and platoon dynamics in bicycle traffic at signalised intersections, following and overtaking behaviour of cyclists on segregated bike paths, lane-changing behaviour of car drivers at motorway weaving sections, and pedestrian flow and congestion at train stations.