基于正态分布的三维建筑物变化检测高度差阈值确定方法

A method for determining height difference threshold of 3D building change detection based on normal distribution

针对现有三维建筑物变化检测中高度差阈值确定方法存在主观性强、阈值通用性差的问题,本文提出了一种基于正态分布的高度差阈值确定方法。借鉴二维变化检测中变化和未变化像元差值服从正态分布的理论,首先,假设2期DSM高度差服从或近似服从正态分布,并分析了变化和未变化区域高度差的分布;然后,基于拉依达准则提出了利用期望和标准差计算高度差阈值的方法;最后,利用滑动窗口动态计算待检测区域高度差阈值。在2个试验区内,基于本文阈值方法的建筑物变化检测结果较经验阈值方法在完整率、正确率和检测质量上分别平均提高了7.2%、1.5%和7.8%,说明该方法具有较好的通用性。

At present,the height difference threshold determination method in 3D building change detection is highly subjective and poorly generalized.In this paper,an adaptive method for determining height difference thresholds based on normal distribution is proposed.Firstly,based on the theory that the difference between changed and unchanged image elements in 2D change detection obeys normal distribution,the method assumes that the height difference of 2-phase digital surface model(DSM)obeys or approximately obeys normal distribution,and analyzes the distribution of height difference between changed and unchanged regions.Then,the method of calculating the height difference threshold using expectation and standard deviation is proposed based on the Laida criterion.Finally,using sliding window to dynamically calculate the height difference threshold.In two experimental areas,the building change detection results based on the proposed method improve the completeness,correctness and detection quality by 7.2%,1.5%and 7.8%on average,compared with the empirical threshold method.This indicates that the method has good generality.