摘要
为确保体育运动训练的科学性,设计体育运动视频关键帧自动提取系统。系统利用内置嵌入式Linux系统实时采集体育运动视频,利用目标跟踪模型从动态注意力和静态注意力两个角度跟踪视频目标,经基于动态优先的特征融合方法融合动态注意力和静态注意力后,将融合后注意力值最大的帧作为提取到的运动视频关键帧。实验结果表明系统可更有效获取视频关键帧,提取视频关键帧的压缩比例高,可有效清除冗余视频帧;且关键帧提取保真度和重构度高。
In order to ensure the scientificity of sports training,an automatic key frame extraction system of sports video is designed.The system uses the built-in embedded Linux system to collect sports video in real time,and uses the target tracking model to track the video target from the perspectives of dynamic attention and static attention.After fusing dynamic attention and static attention based on the dynamic priority feature fusion method,the frame with the largest attention value after fusion is taken as the extracted key frame of sports video.The experimental results show that the system can obtain video key frames more effectively,the compression ratio of extracted video key frames is high,and the redundant video frames can be removed effectively;The fidelity and reconstruction of key frame extraction are high.
作者
黄程斌
HUANG Cheng-bin(Physical Education Department,Guangzhou Huashang College,Guangzhou 511300 China)
出处
《自动化技术与应用》
2023年第7期43-46,73,共5页
Techniques of Automation and Applications
基金
广州华商学院2023年校内导师制科研项目(2023HSDS10)。
