基于典型相关的视频运动目标检测  

Video Moving Object Detection Based on Canonical Correlation

作　　者：徐俊[1] 郭春生[1] 沈佳[1] 

机构地区：[1]杭州电子科技大学通信工程学院,浙江杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2015年第1期45-48,共4页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：国家自然科学基金资助项目(61372157)

摘　　要：在智能监控系统中,视频运动目标检测是目标分类和行为理解的基础。由于视频背景的复杂性,考虑检测问题中能量函数的全局优化,提出了基于典型相关的树加权置信传播(TRW)算法用于视频运动目标检测。算法首先将视频图像分成一些图像块,构造有环图模型;然后运用TRW算法将环路分解成生成树的形式,实现有环图模型上的视频运动目标优化检测。仿真结果表明,算法具有较高的准确度,与传统的置信传播(BP)算法相比较,能有效降低迭代次数,提高计算效率。Video moving object detection is the foundation of the behavior understanding and target classification in the intelligent surveillance system. As the complexity of video background,consider the global optimization of the energy function in object detection. In this paper,tree re-weighted belief propagation（ TRW） algorithm based on canonical correlation is proposed to detect the video moving objects. This algorithm firstly separates the video image into some blocks and constructs a loop model. Then it uses the TRW algorithm to decompose the loop into spanning tree and achieve the detection of the video moving objects on the loop model. The experimental results verify the veracity of the algorithm,compared with the traditional belief propagation（ BP） algorithm, TRW can decrease the message iterations effectively and improve the computational efficiency of the algorithm.

关 键 词：置信传播 典型相关 视频运动目标检测 

分 类 号：TN911.73[电子电信—通信与信息系统]