VETD220: A Visible-Event Benchmark and Baseline for Tracking Drones

View at publisher Summarise with VIRAT
Authors: Jiayu Qiu, Yasheng Zhang, Kaiyuan Zheng, Pengju Li, Jianan Li, Gege Sun, Yuqiang Fang
Cited by: 0
References: 65

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Anti-drone tracking faces challenges from targets’ fast movement, complex backgrounds, and extreme lighting, which often cause overexposure and motion blur in the visible modality, leading to insufficient tracking robustness. Owing to the High Dynamic Range (HDR) and high temporal resolution of event cameras, they can provide sufficient auxiliary information to the visible modality, and their complementary features offer a new solution for anti-drone tracking. In this paper, we first introduce event modality to support visible data. Specifically, we propose the multi-modal anti-drone tracking dataset, providing both visible sequences and event streams, namely VETD220. It contains 220 videos and 68,379 visible-event sequence pairs, covering multiple scenarios and target sizes with fine-grained annotations.The proposed VETD220 provides sufficient data for the anti-drone tracking community, which facilitates model training and fair comparison. Meanwhile, we propose EMFETrack, an event-guided multi-modal feature enhancement algorithm for drone tracking. It extracts spatiotemporal features of events via sparse convolution and designs two key modules: Event-Guided Dynamic Feature Enhancement (EG-DFE) and Motion-aware Position Encoding (MAPE). These modules achieve effective modality interactions and improve feature discriminability in complex scenes like background clustering and target blur. We conduct comparative experiments with 17 mainstream trackers on VETD220, where our method achieves a Success Rate (SR) of 72.7% and a Precision Rate (PR) of 96.8%. It also performs competitively or superiorly on FE240hz and COESOT datasets, verifying its effectiveness. VETD220 is available at https://github.com/QiuuuJY/VETD220.

Fields of Study & Subjects

Video Surveillance and Tracking Methods Infrared Target Detection Methodologies Advanced Neural Network Applications

Publication details

Journal / source
IEEE Transactions on Circuits and Systems for Video Technology
Volume
36
Issue
8
Pages
11488–11502
Publication date
30 March 2026
Language
EN