This is a research I started through a collaboration with some colleagues as part of the "RPASInAir" PON project.
Drone vision
Unmanned Aerial Vehicles (UAVs), commonly known as drones, are increasingly used in many domains, from fast delivery to video surveillance and aerial photography. Their growing popularity is mainly due to the commercial availability of a large variety of drones, even at very low prices. Some of these drones come with inexpensive but powerful embedded cameras and GPUs, making them excellent platforms for decision-making tools. However, while these perspectives are fascinating, there are also some drawbacks to be aware of. On the one hand, computer vision algorithms applied to aerial images are burdened with further difficulties because the scale and perspective issues are taken to the extreme. On the other hand, the methods commonly used in this field, which are sophisticated and computationally intensive, must meet the strict computational requirements imposed by the UAV. This makes any computer vision task applied to images captured by drones (e.g., crowd detection tasks) a real challenge.
Some publications
Castellano, G., De Marinis, P., & Vessio, G. Applying Knowledge Distillation to Improve Weed Mapping With Drones. FedCSIS (2023)
Castellano, G., De Marinis, P., & Vessio, G. Weed mapping in multispectral drone imagery using lightweight vision transformers. Neurocomputing (2023)
Castellano, G., Cotardo, E., Mencar, C., & Vessio, G. Density-based clustering with fully-convolutional networks for crowd flow detection from drones. Neurocomputing (2023)
Caputo, S., Castellano, G., Greco, F., Mencar, C., Petti, N. & Vessio, G. Human Detection in Drone Images Using YOLO for Search-and-Rescue Operations. AIxIA (2021)
Castellano, G., Castiello, C., Cianciotta, M., Mencar, C. & Vessio, G. Multi-view Convolutional Network for Crowd Counting in Drone-Captured Images. ECCV Workshop (2020)
Castellano, G., Castiello, C., Mencar, C. & Vessio, G. Crowd Detection in Aerial Images Using Spatial Graphs and Fully-Convolutional Neural Networks. IEEE Access (2020)