Analysis of data from UAVs for surveillance and threat identification in maritime areas

• Main Authors: Lampropoulos, Fotis, Daramouskas, Ioannis, Perikos, Isidoros, Paraskevas, Michael
• Format: Conference Proceeding
• Language: English
• Subjects: computer vision; Data science; Deep learning; Image capture; Image recognition; Object recognition; Surveillance; UAVs; YOLO
• ISSN: 2835-4419
• DOI: 10.1109/BCD57833.2023.10466306

In this paper, we present a methodology for drones for recognizing different types of objects in maritime areas. The concept and the aim is to assist the national maritime surveillance authorities in the identification of treats and the recognition of the exact type of the treat. The methodology relies on the use of deep learning networks and YOLO framework as well as regression models. The YOLO model detects where each object is and which label should be applied. In this way, object detection is subject to the analysis of Machine Learning-based approaches and Deep Learning-based approaches providing more information about the video or an image than traditional approaches to recognition. These approaches are used to identify groups of pixels that may individually belong to an object. This then feeds a regression model with the help of a convolutional neural network and more specifically, R-CNN, Fast R-CNN, Faster R-CNN and after that object detection algorithm approximates the location of the object and gives its label at the same time. The purpose of the application is to detect objects with an emphasis on identifying one or more targets of interest from data of a video or an image capture. An experimental study was conducted in real-world conditions and revealed quite interesting findings.