A Year-Long Dataset of Pedestrian Trajectories with Synchronized Climate Observations

Pedestrian trajectories provide fundamental data for a wide range of disciplines, including urban analytics, crowd management, and transportation systems. However, existing trajectory datasets are typically limited to short-duration benchmark datasets or lack contextual environmental information. Here, we present the first publicly available year-long dataset comprising 10,768,177 pedestrian trajectories and 4,365 synchronized hourly climate observations. The data are derived from videos recorded daily between 06:00 and 18:00 over a full year (16 February 2025 to 15 February 2026) by a fixed camera facing a commercial street in Shinjuku, Tokyo, Japan. Hourly climate observations are extracted from the embedded weather panel using a vision-language model (VLM), while pedestrian trajectories are extracted through fine-tuned YOLOv12 detection, BoT-SORT tracking, perspective transformation, and trajectory post-processing. Both climate observations and pedestrian trajectories are independently validated to ensure data reliability. This dataset enables fine-grained, longitudinal analyses of pedestrian behavior under varying climate conditions, providing a valuable foundation for future data-driven urban research.