三维虚拟视频关键帧特征提取算法研究  

Research on Keyframe Feature Extraction Algorithm for 3D Virtual Video

作　　者：兰珂 曹计划[2] LAN Ke;CAO Ji-hua(Guilin Institute of Information Technology,Guilin Guangxi 541000,China;Guilin University of Electronic Technology,Guilin Guangxi 541000,China)

机构地区：[1]桂林信息科技学院,广西桂林541000 [2]桂林电子科技大学,广西桂林541000

出　　处：《计算机仿真》2025年第1期188-191,461,共5页Computer Simulation

基　　金：转设背景下民办高校艺术设计类人才特色化发展路径研究(2021ZJY705)。

摘　　要：由于三维虚拟视频中的运动目标特征难以捕捉,在提取关键帧时极易出现漏提取和误提取的情况,为此,提出一种三维虚拟视频关键帧特征提取算法。通过将三维虚拟视频背景初始化重构和更新后,从手工特征提取和SIFT深度特征提取两方面得到视频帧中的关键帧,并分别计算其相似度;利用加权融合算法将两方面特征提取得到的关键帧相似度融合在一起,获得三维虚拟视频关键帧总相似度,通过与阈值对比,实现关键帧的特征提取。实验结果表明,所提方法提取到的均为关键帧特征,且关键帧提取准确率、查全率和F_1值均保持在95%以上,表明所提方法具有高提取精度,没有出现漏提取和误提取现象。At present,it is difficult to capture the features of moving targets in 3D virtual videos.However,there is a high risk of missing extraction and incorrect extraction of keyframes.Therefore,a keyframe feature extraction algorithm for 3D virtual video was proposed.After initializing,reconstructing and updating the background of the 3D virtual video,keyframes in the video frames were obtained by manual feature extraction and SIFT deep extraction.And then,their similarities were calculated separately.Moreover,the weighted fusion algorithm was adopted to fuse the similarities of keyframes,thus obtaining the total similarity of keyframes in the 3D virtual video.After comparing with the threshold,we completed the feature extraction of keyframes.Experimental results show that the proposed method extracts all keyframe features.Meanwhile,the accuracy,recall,and F1 values of keyframe extraction remain above 95%,indicating that the method has high extraction accuracy,without omissions or erors in extraction.

关 键 词：三维虚拟视频 关键帧特征提取 背景初始化重构 相似度 加权融合算法 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]