高速运动目标特征关联检测模型仿真  

Simulation of Feature Association Detection Model for High-Speed Moving Targets

作　　者：叶裴雷[1] 张大斌[2] YE Pei-lei;ZHANG Da-bin(Faculty of Megadata and Computing,Guangdong Baiyun University,Guangzhou Guangdong 510450,China;College of Mathematics and Informatics,South China Agricultural University,Guangzhou Guangdong 510642,China)

机构地区：[1]广东白云学院大数据与计算机学院,广东广州510450 [2]华南农业大学数学与信息学院,广东广州510642

出　　处：《计算机仿真》2023年第4期208-212,共5页Computer Simulation

基　　金：国家自然科学基金(71971089)。

摘　　要：与静态目标不同,在高速运动过程中采集的视频图像存在阴影,为了实现对运动目标的高精度检测,提出基于深度学习的高速运动目标检测模型设计方法。采用光照评估方法判断图像中是否存在阴影,分割视频图像中的阴影区域,消除图像阴影;利用高斯核函数建立滤波器,对运动图像展开滤波处理,消除图像中存在的杂点,并通过剔除兴趣点中存在的冗余点以提高目标检测的准确度和效率,确定图像中的目标区域;采用深度特征网络提取目标特征,结合余弦距离和DeepSort算法展开特征匹配与数据关联分析,根据分析结果利用匈牙利算法构建高速运动目标检测模型,通过该模型实现目标检测。实验结果表明,所提方法的目标检测质量、目标检测精度和检测效率均具有较高的水平。In order to achieve high-precision detection for moving targets,this paper designed a model of detecting high-speed moving targets based on deep learning.Firstly,the lighting assessment method was adopted to judge whether there was shadow in images,and segment the shadow area,thus eliminating the shadow.Secondly,Gaussian kernel function was used to construct a filter for filtering moving images,thus eliminating the noise from the images.Meanwhile,the accuracy and efficiency of target detection were improved by eliminating the redundant points from interest points,and then the target area can be determined.Thirdly,deep feature network was used to extract target features.The cosine distance was combined with the DeepSort algorithm to match features and analyze the data association.Based on the analysis results,Hungarian algorithm was adopted to construct a model to detect high-speed moving target.Experimental results show that the proposed method has high detection quality,detection accuracy and detection efficiency.

关 键 词：深度学习 光照评估方法 深度特征网络 高速运动目标检测 图像滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]