Xue-Feng Zhu,Tianyang Xu,Jian Zhao,Jia-Wei Liu,Kai Wang,Gang Wang,Jianan Li,Qiang Wang,Lei Jin,Zheng Zhu,Junliang Xing,Xiao-Jun Wu
Abstract:Unmanned Aerial Vehicles (UAVs) have been widely used in many areas, including transportation, surveillance, and military. However, their potential for safety and privacy violations is an increasing issue and highly limits their broader applications, underscoring the critical importance of UAV perception and defense (anti-UAV). Still, previous works have simplified such an anti-UAV task as a tracking problem, where the prior information of UAVs is always provided; such a scheme fails in real-world anti-UAV tasks (i.e. complex scenes, indeterminate-appear and -reappear UAVs, and real-time UAV surveillance). In this paper, we first formulate a new and practical anti-UAV problem featuring the UAVs perception in complex scenes without prior UAVs information. To benchmark such a challenging task, we propose the largest UAV dataset dubbed AntiUAV600 and a new evaluation metric. The AntiUAV600 comprises 600 video sequences of challenging scenes with random, fast, and small-scale UAVs, with over 723K thermal infrared frames densely annotated with bounding boxes. Finally, we develop a novel anti-UAV approach via an evidential collaboration of global UAVs detection and local UAVs tracking, which effectively tackles the proposed problem and can serve as a strong baseline for future research. Extensive experiments show our method outperforms SOTA approaches and validate the ability of AntiUAV600 to enhance UAV perception performance due to its large scale and complexity. Our dataset, pretrained models, and source codes will be released publically.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to continuously detect and track Unmanned Aerial Vehicle (UAV) targets in complex scenarios without prior information. Specifically, existing anti - UAV systems usually simplify the task into a tracking problem and always provide prior information of UAVs. However, in practical applications, this solution performs poorly when facing complex scenarios, uncertain appearance and reappearance of UAVs, and real - time monitoring requirements. Therefore, this paper proposes a new anti - UAV task, that is, continuously detecting and tracking UAV targets in complex scenarios without prior UAV information. To support this new task, the authors propose a large - scale UAV dataset AntiUAV600 and a new evaluation metric. AntiUAV600 contains 600 video sequences and more than 723,000 frames of thermal infrared images, which are densely annotated with bounding boxes and cover various challenging scenarios, such as small, fast - moving UAVs that randomly appear. In addition, the authors also develop a new global detection and local tracking method based on evidence cooperation, which effectively solves the proposed problem and provides a strong baseline for future research.
### Main contributions:
1. **Proposing a new problem**: Defining a new real - world anti - UAV problem, that is, continuously detecting and tracking UAV targets in complex scenarios without prior information.
2. **Constructing a benchmark dataset**: Proposing the currently largest UAV dataset AntiUAV600, which contains 600 video sequences and more than 723,000 frames of thermal infrared images, which are densely annotated with bounding boxes and cover various challenging scenarios.
3. **Developing a new framework**: Proposing a new global detection and local tracking framework based on evidence cooperation, which can effectively handle the random appearance and disappearance of UAVs and can adaptively switch between detection and tracking.
4. **Performance verification**: Through extensive experiments and analyses, demonstrating the effectiveness of the proposed AntiUAV600 dataset and framework, especially performing well in handling scenarios where UAV targets disappear and reappear.
### Technical details:
- **Dataset characteristics**: The AntiUAV600 dataset contains challenging features such as complex backgrounds, small UAVs, and randomly appearing and disappearing UAVs.
- **Evaluation metric**: Proposing a new evaluation metric Acc, which combines the mean Intersection over Union (IoU) score and a penalty term to evaluate the positioning ability of anti - UAV systems and the ability to predict the absence of UAVs.
- **Framework design**: The framework includes a global detection branch and a local tracking branch, and an evidence - based head module is used to evaluate the uncertainty of tracking results, thereby achieving adaptive switching between detection and tracking.
### Experimental results:
- **Performance on AntiUAV600**: The proposed EDTC method significantly outperforms other methods, achieving an Acc of 0.486.
- **Performance on other datasets**: On the DUT Anti - UAV and Anti - UAV datasets, the EDTC method also performs well, achieving an Acc of 0.606 and 0.634 respectively.
- **Performance in scenarios where UAVs disappear and reappear**: In 63 videos containing UAV disappearance and reappearance, the EDTC method significantly outperforms existing methods, achieving an Acc of 0.387.
Overall, this paper provides important progress for the research and application of anti - UAV systems by proposing new problems, constructing high - quality datasets, and developing innovative algorithm frameworks.