基于微分平滑滤波的特征点跟踪  

FEATURE POINT TRACKING BASED ON SMOOTHING DIFFERENTIATION FILTERING

作　　者：邓义[1] 赵晖[1] 

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《计算机应用与软件》2011年第5期38-41,共4页Computer Applications and Software

基　　金：国家自然科学基金(60962005);新疆维吾尔自治区自然科学基金(200821124);教育部科学技术研究重点项目(208164);新疆高校科研计划青年教师科研启动基金(xjedu2007509)

摘　　要：光流跟踪作为一种重要的二维运动估计技术,在运动目标检测和跟踪中有着重要的作用。L-K光流跟踪算法是一种利用Newton-Raphson梯度下降法进行图像匹配的算法,所以离散图像的偏导数的计算显得尤为重要。由于数字图像离散的方格结构导致在计算各阶偏导数时产生较大的误差。利用微分平滑滤波的方法先利用一个连续的基础函数模拟离散的灰度图像,再利用基础函数取代离散的图像数据来计算偏导数。实验表明,该方法可以有效地降低由图像的不连续性导致在计算偏导数时产生的误差,从而提高特征点跟踪的精度。Optical flow tracking,as an important estimation technique of two-dimensional movement,plays an important role in moving object detection and tracking.Lucas-Kanade optical flow tracking algorithm is an algorithm making use of Newton-Raphson gradient descent algorithm to match the images,so the computation of partial derivatives of discrete image is of paramount importance.Since it is the discrete lattice structure of digital image which results in occurring a bigger error when computing partial derivatives with various orders,we employ smoothing differentiation filtering method to compute partial derivatives in the way of simulating the discrete gray image using a consecutive underlying function first and followed by substituting discrete image data with the underlying function.Experiments show that this method can effectively reduce the error of partial derivatives computation caused by the discontinuity of the image so as to improve the accuracy of feature point tracking.

关 键 词：L-K算法 特征点跟踪 微分平滑滤波 基础函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]