Drone Detection Using Dynamic-DBSCAN and Deep Learning in an Indoor Environment

Drones have found extensive utility in both public and personal places. Consequently, the accurate detection and tracking of drones have emerged as pivotal endeavors in terms of ensuring their optimal performance. This research paper introduces a novel application for discerning the movements of humans and drones from cloud points through the utilization of frequency-modulated continuous wave radar. The dynamic density-based spatial clustering of applications with noise (Dynamic-DBSCAN) algorithm was employed to classify cloud points into separate groups corresponding to the number of objects within the tracking area. Compared to the original DBSCAN algorithm, this method increased accuracy by about 16.8%, achieving an accuracy of up to 93.99%. Subsequently, a trio of deep learning algorithms— long short-term memory, deep neural network, and residual network (ResNet)—were harnessed to ascertain the categorization of each group as either human or drone. According to the results, ResNet achieved the best accuracy rate of 97.72%. Overall, this study underscores the efficacy of the proposed method in accurately and efficiently distinguishing between human and drone entities for effective monitoring and management.