Welcome to Francis Academic Press
Login | Register

Academic Journal of Computing & Information Science, 2024, 7(12); doi: 10.25236/AJCIS.2024.071210.

TSFPN-YOLO: Object Detection under Low-Altitude Traffic Surveillance Using a Twin-Stream Feature Pyramid Network

Author(s)

Shengjie Feng, Kaixuan Cui, Shu Guo, Bingcheng Jiang

Corresponding Author:
Bingcheng Jiang
Affiliation(s)

Guilin University of Electronic Technology, Guilin, China

Abstract

With the swift advancement of unmanned aerial vehicle (UAV) technology, low-altitude traffic surveillance has become crucial for intelligent traffic management and is extensively used in military reconnaissance, police suspect tracking, and large-scale target searches. Twin-Stream feature fusion methods can enhance detection accuracy and system robustness. However, current methods face challenges in detection accuracy and real-time performance in dynamic, low-altitude scenarios. This paper introduces a novel object detection approach using a modified TSFPN_Concat structure to optimize YOLOv8 for these tasks. We present an improved Twin-Stream Feature Pyramid Network fusion mechanism that boosts detection accuracy and adaptability in complex scenarios.  To validate the effectiveness of the proposed model, we refined and curated a portion of the dataset [1] published in 2020. Compared with mainstream methods such as YOLOv5, YOLOX, Faster R-CNN, and Mask R-CNN, our experimental results demonstrate that the improved TSFPN-YOLO outperforms the original YOLOv8[2] in multiple metrics. Notably, it achieves a 45.2% mAP@50%, surpassing other comparison models and demonstrating remarkable performance improvement. Research indicates that the improved TSFPN_Concat structure can effectively address the insufficient multi-scale feature fusion problem in low-altitude object detection, thereby enhancing the accuracy in low-altitude traffic surveillance tasks.

Keywords

Object Detection, Twin-Stream Feature Pyramid Network (TSFPN), Low-Altitude Traffic Surveillance, YOLOv8

Cite This Paper

Shengjie Feng, Kaixuan Cui, Shu Guo, Bingcheng Jiang. TSFPN-YOLO: Object Detection under Low-Altitude Traffic Surveillance Using a Twin-Stream Feature Pyramid Network. Academic Journal of Computing & Information Science (2024), Vol. 7, Issue 12: 73-78. https://doi.org/10.25236/AJCIS.2024.071210.

References

[1] Bozcan I, Kayacan E. Au-air: A multi-modal unmanned aerial vehicle dataset for low altitude traffic surveillance[C]//2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020: 8504-8510.

[2] Sohan M, Sai Ram T, Reddy R, et al. A review on yolov8 and its advancements[C]//International Conference on Data Intelligence and Cognitive Informatics. Springer, Singapore, 2024: 529-545.

[3] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2014: 580-587.

[4] Dong, Y.-B., Zeng, H., & Hou, S.-J. LMUAV-YOLOv8: A Lightweight Network for Low-Altitude UAV Visual Object Detection [J/OL]. Computer Engineering and Applications, 1–21 [2024-12-11]. Retrieved from http://kns.cnki.net/kcms/detail/11.2127.tp.20241030.1319.008.html.

[5] Xu, T.-N., Gao, A., Chen, Y.-C., et al. A Lightweight Recognition Method for Low-Altitude Battlefield Targets [J/OL]. Journal of Ordnance Engineering, 1–12 [2024-12-11]. Retrieved from http://kns.cnki. net/kcms/detail/11.2176.TJ.20240816.1339.002.html.

[6] Shi Q, Li J. Objects detection of UAV for anti-UAV based on YOLOv4[C]//2020 IEEE 2nd International Conference on Civil Aviation Safety and Information Technology (ICCASIT. IEEE, 2020: 1048-1052.