Development of a Deep-Learning Based People Counting Algorithm Using Drone Capturing Images

This study proposes an algorithm that can be utilized in various drone-related fields. The algorithm builds training data based on drone images and applies deep learning to detect and count the objects. The real-time detection algorithm YOLO was used for training, and the results obtained were used to validate object detection performance. To address identification errors, the labeling area was refined, and additional training data containing shadows were collected, whereby an accuracy of more than 90% was achieved. Furthermore, by developing a people counting algorithm, the proposed method can be applied to various fields, such as crowd crush prevention, tracking people’s movements, and access management. In addition, if diverse environments and types of training data are secured, this algorithm can be widely applied to disaster management fields such as searching for missing persons