Image-based Automated Framework for Detecting and Classifying Unmanned Aerial Vehicles

UAVs are expected to be extensively used in many applications related to modern transportation systems, e.g., traffic monitoring, flying police-eye, and flying roadside units. However, it is important to note that UAVs can also be maliciously utilized as a threat to public safety and pose a significant risk to the stability and execution of smart city applications. Detection of flying intruders can be performed using various sensors such as cameras, RADAR, and LIDAR. In this paper, we propose an automated framework for UAV detection and classification using ground cameras. First, we use YOLOv8 for detecting UAVs, and then we use an unsupervised clustering approach to classify the detected objects according to their different categories. The clustering process is performed by extracting the Histogram of Oriented Gradients (HOG) features of the detected UAVs. Afterward, the features are embedded by mapping them into a two-dimensional space where the separation of the classes is possible. To this end, we use the t-distributed stochastic neighbor embedding (t-SNE) approach. Our framework is tested on an anti-intrusion UAV unlabeled dataset where we identify all categories of UAVs within that dataset. Our approach has shown remarkable clustering performance compared to existing machine-learning methods.