June 7 (Wednesday) 4:00 p.m., room D401
Presentation of the current NATO SPS project #G6032 “UAV Mosquito Fleet for Smart Swarm Operations (UAVM4SSO)”
Abstract
The Edge Intelligence concept of ultra-small and ultra-cheap Unmanned Aerial Vehicles (UAVs) with correlated behavior, UAV Mosquito Fleet (UAFMF), is proposed for implementation of Unmanned Drone Swarm Operations (UDSOs) in monitoring, rescue, reconnaissance, … operations. The main idea is to create UAFMF management platform, improve the state-of-the-art (SOTA) deep neural networks (DNNs) for UDSOs, and adopt DNNs for UAVs with the limited computing resources and power supply.
Review of the Current Object Detection Methods Used in NTUU KPI (Kyiv, Ukraine)
2.1. YOLOv4 - Context-Aware Data Augmentation for Efficient Object Detection by UAV Surveillance.
Abstract
The problem of object detection by YOLOv4 deep neural network (DNN) is considered on drone dataset with object classes (pedestrians, bicyclists, cars, skateboarders, golf carts, and buses) collected by Unmanned Aerial Vehicle (UAV) video surveillance. Some frames (images) with labels were extracted from videos of this dataset and structured in the open-access SDD frames (SDDF) version (https://www.kaggle.com/yoctoman/stanford-drone-dataset-frames). The context-aware data augmentation (CADA) was proposed to change bounding box (BB) sizes by some percentage of its width and height. Several CADA-sequences were analyzed and the best strategy consists in first-IN-then-OUT CADA procedures, where the extent of decrease and increase of BBs width and height can be different for various applications and datasets
2.2.Example 2: YOLOv5 family - Object Detection for Rescue Operations by High-altitude Infrared Thermal Imaging Collected by Unmanned Aerial Vehicles.
Abstract
The analysis of the object detection deep learning model YOLOv5, which was trained on High-altitude Infrared Thermal (HIT) imaging, captured by Unmanned Aerial Vehicles (UAV) is presented. The performance of the several architectures of the YOLOv5 model, specifically ’n’, ’s’, ’m’, ’l’, and ’x’, that were trained with the same hyperparameters and data is analyzed. The dependence of some characteristics, like average precision, inference time, and latency time, on different sizes of deep learning models, is investigated and compared for infrared HIT-UAV and standard COCO datasets. According to the findings, the significance and value of the research consist in comparing the performance of the various models on the datasets COCO and HIT-UAV, infrared photos are more effective at capturing the real-world characteristics needed to conduct better object detection.
2.3.Example 3: The Other Current Object Detection Methods and Future Plans Proposed by NTUU KPI (Kyiv, Ukraine)
Abstract
The real-world (from Russia’s war on Ukraine) UAV experience collected by the authors and end users will be reviewed in the context of possible application of advanced optical systems and AI-ML-DL-based methods.
3. Speakers: Dr. Olexandr Rokovyi, Oleg Alienin
Review of the Current UAV Navigation Methods Used in NTUU KPI (Kyiv, Ukraine)
Abstract
The current UAV operation experience collected by the authors and end users will be reviewed in the context of possible application of advanced optical systems and UAV navigation methods under field conditions in the dangerous operation zones.
Demo Conditions: The further details will be provided later, but at the moment we assume that any “playground” with the sizes (100 m * 100 m * 50 m) which is free of any objects will be sufficient to demonstrate the basic concepts and ideas. It could be like an empty football field or any kind of grass field, lawn, etc. This “playground” is assumed to be used for the further experiments at LRC.