To contribute to the scientific community focused on small object detection and UAV applications, we present a new dataset for human detection in UAV imagery, named Unicamp-UAV. This dataset was created by recording videos using a DJI Phantom 4 UAV at multiple locations across the Barão Geraldo Campus of the University of Campinas (Unicamp), Brazil. The videos were acquired under daylight conditions and favorable weather, considering different camera orientations and individuals in diverse poses to ensure data heterogeneity and to produce a representative dataset for the target area of interest. In total, the dataset comprises 6,500 UAV images, in which 58,555 human instances are annotated. The dataset is an outcome of the doctoral research entitled "Real-time UAV path planning for last-mile delivery operations with a focus on urban safety supported by deep learning-based human detection". The Unicamp-UAV dataset is publicly available through a GitHub repository at https://github.com/SimoesDP/Unicamp-UAV-Dataset .