Object Detection and Trajectory Prediction of Unmanned Aerial Vehicle Using Deep Learning

The increasing use of unmanned aerial vehicles (UAVs) has led to extensive research in detecting drones. Drones, which resemble birds, pose a challenge in accurately identifying them. This paper examines the detection of objects, specifically discussing how deep learning improves the classification, localization, and segmentation of objects. It emphasizes the importance of convolutional neural networks (CNNs) in recognizing objects and the emergence of efficient YOLO models for real-time applications. The comparative analysis focuses on different versions of YOLO, including YOLOv5, YOLOv6, and YOLOv7 discussing their architectural differences, scalability, ease of training, and performance. Additionally, the paper highlights the role of DeepSORT in object tracking, which aims to achieve both high-performance tracking and resilience. Creating a customized dataset for real-time object detection is an essential factor in this study. The dataset includes 75 videos per category, namely drones, planes, and birds, with a total of 6500 frames. This diverse dataset plays a significant role in enhancing deep learning techniques, enabling applications in various fields such as surveillance, airspace management, and wildlife monitoring. The findings of the research reveal certain difficulties in accurately classifying birds due to inconsistent data. However, the performance in detecting planes and drones is particularly noteworthy. The study highlights the importance of adapting models to real-world situations to address the issues of false negatives and false positives. Moreover, this research makes a valuable contribution to the advancement of object detection, especially in the context of the growing significance of drone detection.