Detekce chodců pomocí dronů

Abstract

The main focus of this study was pedestrian detection using drones and convolutional neural networks. 2 detection networks were used - YOLOv5 and Retinanet. The performance was compared based on precision and speed of detection and the demands on training process. Impact of certian training parameters on results was also observed. For training and testing Stanford Drone Dataset was used, containing video recordings captured by drones at Stanford University campus.