基于运动信息与Hog特征的伪装色移动目标跟踪  

Camouflaged moving object tracking algorithm based on motion information and Hog feature

作　　者：侯文迪 王宁[1,2] 李金屏[1,2] 

机构地区：[1]济南大学信息科学与工程学院,济南250022 [2]山东省网络环境智能计算技术重点实验室,济南250022

出　　处：《计算机工程与应用》2014年第24期187-190,共4页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60873089);山东省高等学校科技计划(No.J12LN19)

摘　　要：针对复杂背景下伪装色运动目标的跟踪问题,提出一种基于运动信息与梯度方向直方图的跟踪方法。通过金字塔光流法获取视频序列中目标运动矢量范围;利用目标区域的灰度特征以及Hog特征构建特征向量,沿着运动主方向在速度大小范围内进行二分法匹配,进而确定目标的准确位置。实验结果表明,针对平缓变化的视频序列,对于颜色色调范围较大或者边缘结构较为明显的伪装色运动目标具有良好的跟踪效果。In order to track the camouflaged moving object under complex background, a method based on multi-resolution optical flow and Histograms of Oriented Gradients（HOG）is proposed. The speed vector range of object is obtained by using the multi-resolution optical flow. The gray feature and HOG feature of object area are used to construct the feature vector,which is a standard of bipartite matching in the range of speed value along the main direction of object, and then the object area is obtained accurately. The experimental results demonstrate the proposed approach is effective to track the camouflaged moving object in slowly changing background when the color of camouflaged object changes greatly or the moving object has obvious edge structure.

关 键 词：伪装色 运动目标跟踪 金字塔光流法 梯度方向直方图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]