基于低秩矩阵恢复的目标快速检测方法研究  

Fast object detection based on low-rank matrix recovery theory

作　　者：李俊[1] 周薇娜[1] 

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《微型机与应用》2016年第10期75-78,共4页Microcomputer & Its Applications

摘　　要：为了实现快速运动目标检测,利用低秩矩阵恢复原理进行视频前景检测,主要针对低秩矩阵恢复算法存在的耗费大部分运算时间且运算较为复杂的奇异值分解问题,应用统一计算结构装置(CUDA)第三方库实现加速计算奇异值分解的低秩矩阵恢复算法优化,得到快速且高效的前景检测方法。基于开源视频序列实验,与原有的低秩矩阵恢复算法进行各项参数的比较,其中加速倍数达一倍以上。实验结果证明,经过优化的算法运算时间变短,具有更高效率。In order to implement fast moving object detection,foreground detection is completed by using low-rank matrix recovery algorithm.As the main computation of low-rank matrix recovery algorithm is the singular value decomposition,most of which is time-consuming and complex,a fast and efficient foreground detection method is obtained by applying the compute unified device architecture（ CUDA） and its thirdparty library to realize the low-rank matrix recovery optimized algorithm which implements the faster computing of singular value decomposition.We test our algorithm on open source video and compare it with original low-rank matrix algorithm on various parameters,and the acceleration ratio achieves more than one times. The experiment result indicates that the optimized algorithm obtains less running time and is more efficient.

关 键 词：低秩矩阵恢复 前景检测 统一计算结构装置 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]