摘要
由于当前已有方法未能对目标图像进行去噪处理,导致目标跟踪精度和目标跟踪成功率下降。提出一种多物体遮挡下基于深度学习的目标跟踪方法。采用主成分分析法和局部像素分组方法对目标图像进行去噪处理。对目标图像关键点进行定位,并在目标的分块区域进行特征提取,对提取的目标特征采用卷积神经网络进行分类,获取优化的目标特征提取结果。在此基础上,将粒子滤波和检测器结合,根据提取到的目标样本特征,在多物体遮挡条件下采用SVM分类器进行目标检测,最终实现目标跟踪。仿真结果表明,所提方法可以有效提升目标跟踪精度以及目标跟踪成功率。
This paper presented a method for tracking target covered by multiple objects based on deep learning.Firstly,principal component analysis and local pixel grouping method were used to remove noise from the target image.Secondly,key points in target images were located,and then features were extracted from the blocks of the target.After that,the extracted target features were classified by a convolutional neural network,so as to obtain optimized target feature extraction results.On this basis,the particle filter was combined with a detector.According to the extracted sample features,an SVM classifier was used to detect targets under multi-object occlusion.Finally,the target tracking was achieved.Simulation results show that the proposed method can effectively improve the accuracy and success rate of target tracking.
作者
王莉君
唐骞
李鹏
WANG Li-jun;TANG Sai;LI Peng(School of Software Engineering,Chengdu University of Information Technology,Chengdu Sichuan 610225,China;Chengdu College of University of Electronic Science and Technology of China,Chengdu Sichuan 610051,China)
出处
《计算机仿真》
北大核心
2023年第9期176-179,309,共5页
Computer Simulation
基金
成都信息工程大学科研基金资助成果(KYTZ202281)。
关键词
多物体遮挡下
深度学习
目标跟踪
特征提取
目标图像去噪
Multi-object occlusion
Deep learning
Target tracking
Feature extraction
Target image denoising
作者简介
通讯作者:王莉君(1983-),女(藏族),四川泸定人,博士,副教授,研究方向:数据挖掘、数字图像处理;唐骞(1978-),男(汉族),四川遂宁人,硕士,讲师,研究方向:嵌入式系统;李鹏(1990-),男(汉族),河南邓州人,博士后,研究方向:自然资源评估、生态系统碳汇。
