[Paper Review] A review of UAV Visual Detection and Tracking Methods

Raed Abu Zitar, Mohammad Azmi Al-Betar+2 more|arXiv (Cornell University)|Jun 8, 2023

Video Surveillance and Tracking Methods52 references4 citations

TL;DR

This paper reviews visual detection and tracking methods for UAVs using multi-sensor fusion, focusing on deep learning techniques like YOLOv3, YOLOv4, and CNNs applied to image, RF, thermal, and acoustic data. It demonstrates that hybrid systems combining radar, vision, and RF detection achieve higher accuracy and real-time performance, with CNN-based models reaching 98.9% classification accuracy.

ABSTRACT

This paper presents a review of techniques used for the detection and tracking of UAVs or drones. There are different techniques that depend on collecting measurements of the position, velocity, and image of the UAV and then using them in detection and tracking. Hybrid detection techniques are also presented. The paper is a quick reference for a wide spectrum of methods that are used in the drone detection process.

Motivation & Objective

• To provide a comprehensive review of visual detection and tracking techniques for UAVs using diverse sensor modalities.
• To analyze the strengths and limitations of individual detection methods including thermal, RF, radar, optical, and acoustic sensing.
• To evaluate hybrid detection systems that combine multiple sensors to overcome individual technology limitations.
• To assess the role of deep learning models such as YOLO and CNNs in improving detection accuracy and real-time performance.
• To identify research gaps and future directions in UAV detection, particularly in sensor fusion and classical machine learning integration.

Proposed method

• Systematic review of UAV detection techniques based on sensor technology: thermal, RF, radar, optical, and acoustic sensors.
• Evaluation of deep learning architectures including YOLOv3, YOLOv4, YOLOv4-tiny, and CNNs for real-time object detection and feature extraction from visual and acoustic data.
• Application of hybrid detection frameworks integrating radar, vision, and RF signals to enhance detection range, accuracy, and robustness.
• Use of advanced signal processing techniques such as FMCW radar, phase interferometry, and Doppler shift analysis for target velocity and range estimation.
• Employment of ensemble learning and late fusion strategies for acoustic signal classification using CNN, RNN, and CRNN models.
• Implementation of preprocessing techniques such as data augmentation to address class imbalance in drone image datasets.

Experimental results

Research questions

• RQ1 How do different sensor-based UAV detection methods (thermal, RF, radar, optical, acoustic) compare in terms of detection range, accuracy, and environmental robustness?
• RQ2 What is the performance of deep learning models like YOLO and CNN in detecting and classifying UAVs across visual, RF, and acoustic modalities?
• RQ3 To what extent can hybrid detection systems improve detection reliability by fusing data from multiple sensors?
• RQ4 How do signal processing techniques such as FMCW radar and Doppler shift analysis enhance UAV tracking and velocity estimation?
• RQ5 What are the limitations of single-sensor detection systems, and how can sensor fusion mitigate these issues?

Key findings

• Hybrid detection systems combining radar, vision, and RF sensors significantly outperform single-modality approaches in detection range and accuracy.
• YOLOv3 and YOLOv4-based models enable real-time UAV detection with high precision, particularly when combined with CNNs for feature extraction.
• CNN-based classification of radar and acoustic signals achieved up to 98.9% accuracy, demonstrating high effectiveness in drone recognition.
• FMCW radar with enhanced signal processing techniques (e.g., phase interferometry) enabled detection of small UAVs at ranges up to 20 nautical miles.
• Acoustic-based detection using CRNN and ensemble models showed strong performance in identifying multirotor drones, especially when combined with optimal microphone array configurations.
• Data augmentation in preprocessing improved model generalization, particularly in low-data regimes for detecting loaded vs. unloaded UAVs.