A dataset is used to train and define a system that can precisely geolocate persons automatically detected in offline processed images recorded during the SAR mission. The dataset contains data 3D simulations for a few real-world terrains of different configurations and complexity using a custom-made 3D terrain generator and raycaster, along with a person detections with YOLO deep neural network on the real terrain. The collected data is used to define a method for geolocating detected persons based on raycasting, which allows using low-cost commercial drones with a monocular camera in SAR missions.