Advancing UAV Multi-Object Tracking: Integrating YOLOv8, Nano Instance Segmentation, and Dueling Double Deep Q-Network Article Swipe
YOU?
·
· 2024
· Open Access
·
· DOI: https://doi.org/10.21203/rs.3.rs-4854100/v1
Unmanned Aerial Vehicles (UAVs) have become indispensable for navigating complex terrains, accessing remote or hazardous locations, and capturing high-resolution imagery. This paper presents an innovative approach to object detection specifically tailored for computer vision applications in UAVs. Traditional deep learning models such as RCNN, Fast RCNN, and YOLO often face challenges in detecting occluded, blurred, or clustered objects and struggle with simultaneously identifying and tracking multiple objects. To overcome these challenges, we propose a framework that integrates YOLOv8x, Nano Instance Segmentation (NIS), and Dueling Double Deep Q Network (DDDQN). YOLOv8x exhibits outstanding performance, achieving an Average Precision (AP) of 53.9% on the demanding MSCOCO dataset, outperforming previous versions. The DDDQN algorithm significantly enhances tracking capabilities by effectively estimating state values and state-dependent action advantages independently. The combination of YOLOv8x and DDDQN facilitates proficient management of obstacles, varying object sizes, and unpredictable movements. We simulated the proposed framework using the UAVDT and VisDrone datasets and compared its performance against approximately nine contemporary frameworks from recent literature. The results demonstrate that our framework significantly improves 1 object tracking in densely populated environments, offering a robust solution for real-world applications requiring precise and resilient object detection.
Related Topics To Compare & Contrast
- Type
- preprint
- Language
- en
- Landing Page
- https://doi.org/10.21203/rs.3.rs-4854100/v1
- https://www.researchsquare.com/article/rs-4854100/latest.pdf
- OA Status
- gold
- Cited By
- 1
- References
- 27
- Related Works
- 10
- OpenAlex ID
- https://openalex.org/W4401810344