Vision-based Anti-UAV Detection and Tracking (2205.10851v1)

Abstract: Unmanned aerial vehicles (UAV) have been widely used in various fields, and their invasion of security and privacy has aroused social concern. Several detection and tracking systems for UAVs have been introduced in recent years, but most of them are based on radio frequency, radar, and other media. We assume that the field of computer vision is mature enough to detect and track invading UAVs. Thus we propose a visible light mode dataset called Dalian University of Technology Anti-UAV dataset, DUT Anti-UAV for short. It contains a detection dataset with a total of 10,000 images and a tracking dataset with 20 videos that include short-term and long-term sequences. All frames and images are manually annotated precisely. We use this dataset to train several existing detection algorithms and evaluate the algorithms' performance. Several tracking methods are also tested on our tracking dataset. Furthermore, we propose a clear and simple tracking algorithm combined with detection that inherits the detector's high precision. Extensive experiments show that the tracking performance is improved considerably after fusing detection, thus providing a new attempt at UAV tracking using our dataset.The datasets and results are publicly available at: https://github.com/wangdongdut/DUT-Anti-UAV

Vision-Based Anti-UAV Detection and Tracking: A Comprehensive Analysis

Unmanned Aerial Vehicles (UAVs) present significant opportunities and challenges across various fields, including logistics, surveillance, and transportation. However, their increasing prevalence poses potential threats to security and privacy, necessitating robust detection and tracking solutions. This paper introduces a vision-based approach for UAV detection and tracking, leveraging the advancements in computer vision and deep learning methodologies.

Development of the DUT Anti-UAV Dataset

The authors propose the DUT Anti-UAV dataset, which is a critical contribution to the field. This dataset comprises 10,000 images for detection purposes and 20 videos for evaluating tracking algorithms. Each image and video sequence is meticulously annotated, providing a rich resource for training and testing state-of-the-art detection and tracking models. This dataset addresses the need for more comprehensive, varied, and challenging data for anti-UAV tasks, compared to existing UAV datasets like MAV-VID and Drone-vs-Bird.

Methodology and Experimentation

The research employs several standard object detection and tracking algorithms, including Faster R-CNN, Cascade R-CNN, ATSS, YOLOX, and SSD for detection, and SiamFC, Eco, TransT, and LTMU for tracking. Each algorithm's performance is meticulously evaluated using the proposed dataset. Notably, Cascade R-CNN with a ResNet50 backbone achieved the highest mean Average Precision (mAP) of 68.3% in detection tasks, while YOLOX with ResNet18 demonstrated superior speed with 53.7 FPS, albeit at a lower accuracy.

In terms of tracking, LTMU stood out with a success rate of 60.8%, demonstrating robust performance in handling small object sizes and complex environmental conditions typical for UAV scenarios. The notable advancement in this paper is the integration of detection algorithms with tracking frameworks, enhancing tracking performance significantly. For instance, coupling Faster R-CNN (VGG16) with LTMU improved success from 60.8% to 66.4%.

Key Challenges and Contributions

The paper identifies several challenges in UAV detection and tracking, such as the small size of UAVs relative to their surroundings, complex backgrounds, and dynamic environments. These challenges necessitate models capable of high precision in detecting and tracking UAVs under varying conditions.

The introduction of the DUT Anti-UAV dataset is a significant advancement, as it not only provides a platform to test current state-of-the-art models but also encourages the development of novel approaches. The dataset's availability promotes transparency and reproducibility in research, allowing others to build upon the findings presented.

Implications and Future Directions

This research underscores the importance of integrating detection with tracking to enhance UAV tracking systems' robustness and accuracy. The proposed fusion strategy can serve as a model for future developments, providing a framework that future algorithms can adopt to improve performance.

Going forward, further work could explore integrating additional sensor data, such as LiDAR or infrared, to supplement vision-based approaches. Refinement of algorithms to handle extreme conditions, such as poor lighting or adverse weather, will also be critical. As UAV technology continues to evolve, future datasets must evolve accordingly, incorporating new UAV models and potential use-case scenarios to maintain relevance in dynamic operational settings.

In summary, this paper effectively addresses current limitations in UAV detection and tracking through the development of the DUT Anti-UAV dataset and a performance-enhancing algorithmic fusion strategy, setting a foundation for ongoing research and development in autonomous UAV surveillance systems.