摘要
针对相关滤波跟踪算法开展并行计算研究,首先介绍相关滤波跟踪算法的理论基础,分析其数学机理和并行计算的必要性;其次对最小输出平方误差和(Minimum Output Sum of Squared Error,MOSSE)相关滤波跟踪算法流程进行分解,设计中央处理器(Central Processing Unit,CPU)-图形处理器(Graphics Processing Unit,GPU)异构计算加速方案;最后进行不同大小目标跟踪时并行计算与串行计算的耗时对比,结果表明当目标模板大于一定值时,并行计算耗时小于串行计算,验证了并行加速的有效性。
In this paper,the parallel computation of correlation filtering tracking algorithm is studied.Firstly,the theoretical basis of correlation filtering tracking algorithm is introduced,and its mathematical mechanism and the necessity of parallel computation are analyzed.Secondly,the process of Minimum Output Sum of Squared Error(MOSSE)related filter tracking algorithm is decomposed,and a Central Processing Unit(CPU)-Graphics Processing Unit(GPU)heterogeneous computing acceleration scheme is designed.Finally,the parallel computation time is compared with the serial computation time when the target template is larger than a certain value,and the parallel computation time is less than the serial computation time,which verifies the effectiveness of parallel acceleration.
作者
辛林杰
程勇策
宋煜明
XIN Linjie;CHENG Yongce;SONG Yuming(The 3rd Research Institute of China Electronics Technology Group Corporation,Beijing 100015,China)
出处
《电视技术》
2025年第7期81-85,共5页
Video Engineering
关键词
视觉跟踪
相关滤波跟踪算法
并行计算
统一计算设备架构(CUDA)
visual tracking
correlation filtering tracking algorithm
parallel computing
Compute Unified Device Architecture(CUDA)
