摘要
针对摄像机较远距离拍摄目标的情形,提出一种利用人体行走时的投影周期性特征进行人体目标检测的方法,首先通过运动分割获取每一帧的运动目标,然后通过计算运动目标的投影相似性对目标进行检测。为了简化计算,利用Hausdorff距离对运动目标的投影进行相似性计算,同时为减少存储空间,利用码书作为存储相似性特征的数据结构。
To handle the situation that objects are acquired by a still camera and the distance between objects and the camera is long,a human detection method based on the motion projection periodicity feature is presented.This method first obtains the moving objects via the motion segmentation algorithm,and then detects humans by calculating the projection similarities of the moving objects.To reduce the requirements of memory space and CPU time,this method uses the Hausdorff distance as a measure to calculate the projection similarity,and stores the calculated results in codebooks.
出处
《计算机工程与应用》
CSCD
北大核心
2010年第9期169-172,共4页
Computer Engineering and Applications
基金
广东省科技计划项目(No.2006B60155)