集成技术2026，Vol.15Issue(3)：96-111,16.DOI:10.12146/j.issn.2095-3135.20251110001
基于全图搜索的单目标实时跟踪
Single-Target Real-Time Tracking Based on Full-Image Search
摘要
Abstract
Single-target tracking as a pivotal task in the field of computer vision,finds extensive applications in various domains such as security surveillance,human-computer interaction,industrial automation,and consumer electronics.With the rapid advancement of computer vision technologies,deep learning-based target tracking algorithms have achieved remarkable results.However,current mainstream methods still confront numerous challenges when dealing with complex scenarios such as rapidly moving targets or brief occlusions.In view of this,this paper proposes a novel real-time single-target tracking algorithm with full-image search based on deep convolutional neural networks.The algorithm integrates a full-image search strategy and a lightweight Siamese single-target tracking model,as well as an information interaction module with a self-attention mechanism.It can significantly enhance the robustness and continuity of target tracking,and is particularly suitable for real-time applications on edge devices.Experimental validations demonstrate that the proposed algorithm maintains superior tracking performance across a range of complex scenarios.关键词
目标跟踪/深度学习/全图搜索/自注意力机制/轻量级孪生网络/实时跟踪/边缘设备Key words
target tracking/deep learning/full-image search/self-attention mechanism/lightweight siamese network/real-time tracking/edge devices分类
信息技术与安全科学引用本文复制引用
唐文名,谢锐涛,刘军..基于全图搜索的单目标实时跟踪[J].集成技术,2026,15(3):96-111,16.基金项目
深圳市高等院校稳定支持计划项目(20231127143250002) This work is supported by Shenzhen Higher Education Stable Support Plan(20231127143250002) （20231127143250002）
