基于改进YOLOX和新型数据关联方式的无人机多目标跟踪方法
Multi-object Tracking of UAV Based on Improved YOLOX and New Data Association Method
付书岗1
作者信息
- 1. 中国科学院空天信息创新研究院,北京 100094;中国科学院空间信息处理与应用系统技术重点实验室,北京 100190;中国科学院大学电子电气与通信工程学院,北京 101408
- 折叠
摘要
无人机视频中的多目标跟踪是一项重要的计算机视觉任务,在各个领域有着广泛的应用.针对无人机视频场景中目标遮挡、小目标、背景复杂多变等挑战,本文提出一种改进的无人机多目标跟踪模型.首先,本文对YOLOX进行改进,将Swin Transformer集成到网络中以增强全局信息提取能力,并增加一个额外的检测头来改善对小目标的检测能力,此外引入CBAM注意力模块来增强聚焦有用信息的能力.数据关联阶段,本文采用一种新型数据关联方式,保留所有检测框,并根据置信度将其划分为高分检测框和低分检测框,对高分检测框与跟踪轨迹进行第一次关联,将未匹配轨迹与低分检测框进行二次关联.在公开数据集VisDrone2021和UAVDT上的实验结果表明,本文方法在无人机多目标跟踪场景中具有较高的优越性和鲁棒性.
Abstract
Multi-object tracking in UAV videos is a crucial computer vision task with extensive applications across various do-mains.To address the challenges of occlusions,small objects,and complex,varying backgrounds in UAV video scenes,an im-proved UAV multi-object tracking model is proposed.This paper improves the YOLOX network by integrating the Swin Trans-former to enhance global information extraction capabilities and adding an additional detection head to boost the detection perfor-mance of small objects.Furthermore,this paper introduces the CBAM attention module to focus on informative features.In the data association stage,this paper adopts a new data association approach that retains all detection boxes,categorizing them into high-scoring and low-scoring detection boxes based on their confidence scores.The first association is performed between high-scoring detection boxes and tracking trajectories,while the second association is performed between unmatched trajectories and low-scoring detection boxes.Experimental results on the public datasets VisDrone2021 and UAVDT demonstrate that the pro-posed method exhibits relatively high superiority and robustness in UAV multi-object tracking scenarios.
关键词
多目标跟踪/无人机视频/注意力机制/数据关联Key words
multi-object tracking/unmanned aerial vehicles(UAV)videos/attention mechanism/data association引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据