几何矩在行为分析中的应用

几何矩表征图像的几何特征,一般几何形状分区域和边界两种,因此几何矩分为区域矩和轮廓矩。7Hu不变矩具有平移、尺度和旋转不变性。论文以二维图像上单个人的走、横跳和跑为例,通过寻找当前帧目标的外层最大轮廓,计算对应子区域的一阶轮廓矩来获取人体质心、计算目标整体的运动速度和方向、绘制质心轨迹来作初步的行为分析。之后通过寻找外层最大轮廓点到质心欧式距离的局部极值点来寻找变化频繁的手脚和头,构建人体模型序列图,作进一步的行为分析。

Invariant Moments and Their Application on Behavior Analysis

Geometric moments characterize the geometric properties of image.General geometric shapes divided into two classes:regions and single-pixel structures,thus geometric moments include region moments and contour moments.keeps invariant in translation,scale and rotation,The paper took behaviors of a single person’s walking,cross-jumping and running for example,found the outest contour of current frame,got the centroid of the target by calculating the contour moment of corresponding sub-region,caculated speed and direction of overall,described centroid locus for early analysis.Through searching local extreme points of Euclidean distance from edge to centroid to search frequently changed feet,hands and head,built human model sequence diagrams to further analysis the three behaviors.