计算机工程与应用 ›› 2011, Vol. 47 ›› Issue (32): 4-6.

• 博士论坛 • 上一篇    下一篇

压缩感知目标跟踪

郭岩松,杨爱萍,侯正信,何宇清   

  1. 天津大学 电子信息工程学院,天津 300072
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2011-11-11 发布日期:2011-11-11

Object tracking using compressive sensing

GUO Yansong,YANG Aiping,HOU Zhengxin,HE Yuqing   

  1. School of Electronic Information Engineering,Tianjin University,Tianjin 300072,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-11-11 Published:2011-11-11

摘要: 视频分析通常在分类或检测等高级任务之前解码并重构视频序列。但是,有时希望只进行视频分析而不暴露敏感信息,例如人员身份。提出了一个能够跟踪目标而不需要重构视频序列的编码方案。根据压缩感知理论,用每帧的少量伪随机投影编码一个视频序列。解码器利用背景消除图像的稀疏性重构前景目标。以粒子滤波器估计的目标位置作为先验知识,可以改进前景目标位置的重构。该编码方案同时具有隐私保护和安全加密功能。

关键词: 压缩感知, 目标跟踪, 粒子滤波, 重新加权l1最小化

Abstract: In video analysis,it usually needs to decode and reconstruct the video sequence before any higher level processing such as classification or detection.However,sometimes a video analysis needs to be proceeded without revealing sensitive information,e.g.the identity of people.This paper proposes a new encoding scheme which enables object-tracking without reconstructing the video sequence.According to compressive sensing theory,encoding a video sequence into a few pseudo-random projections of each frame is reasonable due to the sparsity in the interested information.Then the decoder will exploit the sparsity of background-subtracted image to recover the foreground target.Taking a prior knowledge as target position which is estimated by a particle filter,it is possible to improve the reconstruction of foreground target.As expected,this encoding scheme has capability of privacy-protection as well as security.

Key words: compressive sensing, object tracking, particle filtering, reweighted l1 minimization