Detection and Tracking of Drones Using Deep Learning and Computer Vision

Abstract

The rapid proliferation of unmanned aerial vehicles (UAVs), commonly known as drones, has introduced significant challenges related to airspace security, public safety, and privacy protection. While drones are widely utilized in civilian, commercial, and industrial applications, their unauthorized operation in restricted areas such as airports, military installations, and critical infrastructure poses serious security threats. Traditional drone detection approaches based on radar, acoustic sensing, and radio frequency analysis often suffer from high deployment costs, environmental sensitivity, and limited effectiveness against small, low-altitude drones. To address these limitations, this research paper presents a deep learning–based computer vision framework for drone detection and tracking, developed from an empirical dissertation study. The proposed system employs a convolutional neural network (CNN) to automatically learn discriminative visual features from drone and non-drone images, eliminating reliance on handcrafted feature extraction. A systematic methodology encompassing dataset preparation, image preprocessing, CNN model design, training, validation, and comprehensive performance evaluation is adopted to ensure robustness and reliability. Drone tracking is integrated using detection-based temporal association to enhance stability and reduce false alarms. Experimental results demonstrate an overall classification accuracy of 89.20 percent, with balanced precision, recall, and F1-score values across both classes. Confusion matrix analysis shows strong diagonal dominance, while training and validation curves confirm stable convergence and effective generalization. The findings validate the effectiveness of deep learning–based computer vision for reliable drone detection and tracking in real-world surveillance environments.

Introduction

Drone Detection and Tracking using Deep Learning (Summary)

The study addresses increasing security risks caused by unauthorized UAV usage in sensitive areas, where traditional radar, acoustic, and RF-based detection systems fail due to limitations like low radar cross-section, environmental noise, and communication dependence. To overcome these issues, the research proposes a deep learning–based computer vision framework using CNNs for drone detection and tracking.

Unlike conventional methods relying on handcrafted features, CNNs automatically learn hierarchical visual patterns, making them effective for identifying small, fast-moving drones in complex environments. Tracking is integrated with detection to improve reliability by maintaining temporal consistency and reducing false alarms.

The dataset contains 4,000 balanced images (drone and non-drone), including challenging cases such as birds, aircraft, and other similar objects. Preprocessing and data augmentation improve model robustness.

The system pipeline includes image acquisition, preprocessing, CNN-based classification, tracking, and alert generation, designed for real-time surveillance applications.

Performance evaluation uses accuracy, precision, recall, F1-score, confusion matrix, and learning curves. The model achieved 89.2% accuracy with balanced precision and recall, showing strong generalization and stable training behavior. Errors mainly occur with small or occluded drones, but overall performance is reliable.

In conclusion, integrating CNN-based detection with tracking improves accuracy and reduces false alarms, making the system suitable for real-world UAV surveillance, though challenges remain in detecting very small or obscured drones in complex environments.

Conclusion

The present research has comprehensively investigated the problem of drone detection and tracking using deep learning and computer vision, addressing a critical and increasingly relevant challenge in modern airspace surveillance. The rapid growth in the use of unmanned aerial vehicles across civilian, commercial, and industrial domains has introduced new security, safety, and privacy concerns, particularly in restricted and sensitive environments such as airports, military installations, government buildings, and critical infrastructure facilities. Traditional drone detection technologies, including radar-based, acoustic, and radio frequency sensing methods, while effective under certain conditions, exhibit significant limitations related to cost, environmental sensitivity, limited discrimination capability, and reduced effectiveness against small, low-altitude drones. In response to these challenges, this research proposed and evaluated a vision-based framework grounded in deep learning, with the objective of developing a robust, scalable, and practical solution for drone surveillance.The proposed framework employs a convolutional neural network to automatically learn discriminative visual features from raw image data, eliminating the reliance on handcrafted feature engineering that characterizes many traditional computer vision approaches. By leveraging the hierarchical feature learning capability of CNNs, the system effectively captures both low-level visual cues, such as edges and textures, and high-level semantic characteristics, including drone shape, structural symmetry, and spatial configuration. This data-driven learning process enables the model to distinguish drones from visually similar non-drone objects such as birds, aircraft, and airborne debris, which are common sources of false alarms in surveillance environments.Experimental evaluation of the proposed system demonstrated strong and reliable performance across multiple quantitative metrics. The CNN model achieved an overall classification accuracy of 89.20 percent, reflecting its ability to correctly classify a large majority of drone and non-drone instances. More importantly, the model exhibited balanced precision and recall values, indicating that it successfully minimized both false positives and false negatives. This balanced performance is particularly critical in security-sensitive applications, where excessive false alarms can undermine system credibility, while missed detections may result in serious safety or security risks. The F1-score further confirmed that the model achieved an effective trade-off between detection sensitivity and reliability, reinforcing its suitability for real-world deployment. A notable contribution of this research lies in the integration of detection and tracking within a unified framework. Rather than treating detection and tracking as independent processes, the proposed system incorporates detection-based tracking to exploit temporal consistency across successive video frames. This integration enhances system robustness by reducing sporadic or transient detections that may arise due to noise, illumination changes, or momentary visual artifacts. Tracking enables continuous monitoring of drone movement, allowing the system to maintain object identity over time and estimate trajectories. Such capabilities are essential for threat assessment, situational awareness, and timely response in surveillance applications. The successful integration of tracking with CNN-based detection demonstrates that temporal information plays a crucial role in improving the reliability and operational effectiveness of vision-based drone surveillance systems.The analysis of training and validation performance curves further validated the robustness of the proposed framework. The close alignment between training and validation accuracy and loss curves indicates stable convergence behavior and effective generalization to unseen data. This observation suggests that the adopted training strategy, regularization techniques, and dataset preprocessing steps successfully mitigated overfitting, a common challenge in deep learning models. Stable learning behavior is particularly important for real-world applications, where models must perform consistently across diverse environmental conditions and data distributions. The relatively lightweight architecture of the CNN model also contributes to its practical applicability, as it balances representational capacity with computational efficiency, making it suitable for near-real-time operation.Beyond quantitative performance, this research contributes conceptually to the broader field of intelligent surveillance systems by demonstrating the viability of deep learning–based computer vision as a cost-effective and scalable alternative to traditional sensing technologies. Vision-based systems leverage widely available camera infrastructure and can be deployed with minimal additional hardware cost. When combined with deep learning, such systems adapt to visual variability more effectively than rule-based or handcrafted approaches, enabling robust operation in complex and dynamic environments. The findings of this study therefore support the growing consensus in the literature that deep learning–driven vision systems are well-positioned to play a central role in next-generation drone detection and monitoring solutions. Despite the encouraging results, the study also acknowledges certain limitations that provide opportunities for future research. While the proposed framework demonstrated strong performance on a balanced dataset under diverse visual conditions, real-world deployment may involve more extreme scenarios, such as night-time operation, adverse weather conditions, long-range detection, and highly cluttered environments. Addressing these challenges may require the incorporation of additional contextual information, advanced feature extraction techniques, or specialized training strategies. Furthermore, the current framework focuses on binary classification, distinguishing between drone and non-drone objects. Extending the system to support multi-class classification could enable differentiation between various drone types, sizes, or threat levels, enhancing situational awareness and response planning.Future work may also explore the integration of multimodal sensing approaches, combining visual data with complementary modalities such as acoustic, radar, or radio frequency signals. Multimodal fusion has the potential to improve robustness under conditions where visual information alone is insufficient. Additionally, advances in deep learning architectures, including attention mechanisms and transformer-based vision models, offer promising directions for improving small object detection and feature representation. Optimizing the proposed framework for deployment on edge devices and embedded platforms is another important avenue for future research, as real-time processing and low-latency response are critical requirements for large-scale surveillance systems.In conclusion, this research demonstrates that deep learning–based computer vision provides a powerful, flexible, and scalable foundation for drone detection and tracking in real-world surveillance environments. The proposed CNN-based framework achieved strong detection performance, stable learning behavior, and enhanced reliability through integrated tracking. By addressing key limitations of traditional detection methods and highlighting pathways for future improvement, this study contributes meaningfully to the advancement of intelligent drone surveillance systems. The findings support the continued exploration and deployment of deep learning–driven vision technologies as essential components of secure and resilient airspace monitoring solutions.

References

[1] Rahman, M. H., Sejan, M. A. S., Aziz, M. A., & Song, H.-K. (2024). A comprehensive survey of UAV detection and classification using machine learning approaches. Frontiers in Communications and Networks, 5, 1298743. https://doi.org/10.3389/frcmn.2024.1298743 [2] Macedo, S. O., Silva, L. R., & Santos, A. P. (2025). Drone detection in airport environments: A systematic review of challenges and solutions. Safety Science, 174, 106884. https://doi.org/10.1016/j.ssci.2024.106884 [3] Mrabet, M., Sliti, M., & Ammar, L. B. (2024). Machine learning algorithms applied for drone detection and classification: Benefits and challenges. Frontiers in Communications and Networks, 5, 1284012. https://doi.org/10.3389/frcmn.2024.1284012 [4] Tang, G., Chen, Y., Wang, J., & Li, Z. (2023). A survey of object detection for UAVs based on deep learning. Remote Sensing, 15(3), 687. https://doi.org/10.3390/rs15030687 [5] Wu, Y. (2025). Research advances on deep learning-based small object detection in UAV imagery. SCIENCE CHINA Information Sciences, 68(2), 120101. https://doi.org/10.1007/s11432-024-4021-3 [6] Nguyen, P. T., Tran, T. D., & Le, H. M. (2025). LW-UAV–YOLOv10: A lightweight deep learning model for small UAV detection. Expert Systems with Applications, 239, 121713. https://doi.org/10.1016/j.eswa.2024.121713 [7] Li, W. F., & Zhang, Y. (2025). Enhanced small object detection in UAV aerial imagery through attention-gated backbone and context-aware fusion. Scientific Reports, 15, 10421. https://doi.org/10.1038/s41598-025-87011-3 [8] Kaur, D., & Singh, P. (2025). Analysis of vision-based air-to-air UAV detection using deep learning. Aerospace Science and Technology, 146, 108906. https://doi.org/10.1016/j.ast.2024.108906 [9] Maaroof, M. K. A., & Bouhlel, M. S. (2025). Drone image localization and detection using Faster R-CNN under real-world conditions. Journal of Wireless Mobile Computing, Data and Applications, 16(1), 45–58. [10] Alshaer, N., Abdelfatah, R., & Ismail, T. (2025). Vision-based UAV detection and tracking using deep learning and Kalman filtering. IEEE Access, 13, 34122–34136. https://doi.org/10.1109/ACCESS.2025.3339211 [11] Salisu, M. L., Ibrahim, A., & Bello, S. (2025). An efficient CNN-based model with knowledge distillation for real-time drone detection. Procedia Computer Science, 232, 1121–1130. https://doi.org/10.1016/j.procs.2024.12.141 [12] Ciccone, F., Rinaldi, A., & Bianchi, G. (2025). Real-time drone detection for search and rescue using deep learning. Drones, 9(2), 94. https://doi.org/10.3390/drones9020094 [13] Rakshit, H., & Das, S. (2024). A novel approach to detect drones using deep convolutional learning and transfer learning. IEEE Sensors Journal, 24(8), 12845–12856. https://doi.org/10.1109/JSEN.2024.3362147 [14] Tan, Z., Liu, X., & Wang, H. (2025). A drone-chasing-drone framework based on deep reinforcement learning. Robotics and Autonomous Systems, 177, 104662. https://doi.org/10.1016/j.robot.2024.104662 [15] Dumen?i?, O., Kova?evi?, I., & Grgi?, M. (2025). YOLO-powered UAV detection systems for real-world surveillance applications. Sensors, 25(3), 1187. https://doi.org/10.3390/s25031187 [16] El-Latif, E. I. A., Elseuofi, S., & Abd El-Haleem, S. (2024). Detection and identification of drones using LSTM and Bayesian optimization. Journal of Ambient Intelligence and Humanized Computing, 15, 9123–9141. https://doi.org/10.1007/s12652-024-04812-7 [17] Katkuri, A. V. R., Reddy, S. K., & Kumar, M. (2024). Autonomous UAV navigation and detection using deep learning-based vision systems. Neurocomputing, 558, 126507. https://doi.org/10.1016/j.neucom.2023.126507 [18] Poornima, G. (2025). Computer vision and machine learning in modern UAV surveillance systems. Journal of Intelligent & Robotic Systems, 108(2), 41. https://doi.org/10.1007/s10846-024-02011-5 [19] Silva, R., Costa, J., & Pinho, A. J. (2024). Real-time aerial object detection under challenging environments. Sensors, 24(19), 6324. https://doi.org/10.3390/s24196324 [20] Zhang, H., Liu, Y., & Chen, Z. (2024). Robust UAV detection using multi-scale convolutional neural networks. Pattern Recognition Letters, 178, 20–28. https://doi.org/10.1016/j.patrec.2023.11.018 [21] Wang, J., Li, X., & Zhou, Y. (2025). Vision-based drone detection using temporal consistency modeling. IEEE Transactions on Aerospace and Electronic Systems. Advance online publication. https://doi.org/10.1109/TAES.2025.3456219 [22] He, Q., Sun, J., & Wang, P. (2024). Small UAV detection in cluttered backgrounds using attention-based deep learning. Applied Intelligence, 54(1), 203–219. https://doi.org/10.1007/s10489-023-05021-4 [23] Luo, Y., Chen, X., & Fan, G. (2025). Lightweight deep learning models for real-time UAV detection on edge devices. Future Generation Computer Systems, 152, 150–162. https://doi.org/10.1016/j.future.2024.01.015 [24] Islam, M. R., Rahman, M. S., & Hossain, M. A. (2024). Vision-based drone detection and tracking in complex environments. Multimedia Tools and Applications, 83, 25641–25663. https://doi.org/10.1007/s11042-024-18941-2 [25] Xu, L., Zhao, Y., & Li, D. (2025). Deep learning-driven UAV surveillance: Challenges, trends, and future directions. Artificial Intelligence Review, 58(3), 123. https://doi.org/10.1007/s10462-024-10618-9

Copyright

Copyright © 2026 Dr. P. K. Sharma, Mr. Manvendra Singh Divakar, Bharti Sahu. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.