基於正態分佈的多元離羣點的檢測方法（項目需要對點雲離羣點檢測）的C++實現

基於異常點檢測算法（一）概率統計的原理地址是https://blog.csdn.net/yangziluomu/article/details/73528080

下面是用C++實現了基於正態分佈的多元離羣點的檢測方法

/*-------------------------------------
1.基於正態分佈的點雲離羣點檢測算法
2.作者：pcb
3.日期：2018.10.3
--------------------------------------*/
//定義3D點的結構體
struct Point3D
{
	float x;
	float y;
	float z;
};
/*----------------------------
*功能：採用高斯分佈的方法進行離羣點的判別
*-----------------------------
*輸入：InputPointCloud（Piont3D的原始點雲數據）
*輸出：OutPointCloud（除去離羣點之後的Point3D結構的點雲數據）
*/
void GaussianDistribution_OutlierDetection(vector<Point3D> &InputPointCloud, vector<Point3D>&OutPointCloud)
{
	//均值
	double X_Ave = 0;     
	double Y_Ave = 0; 
	double Z_Ave = 0;
	
	//方差
	double X_Var = 0;      
	double Y_Var = 0;
	double Z_Var = 0;
	
	//求均值
	for (int i = 0; i <InputPointCloud.size(); i++)
	{
		X_Ave +=InputPointCloud[i].x;
		Y_Ave +=InputPointCloud[i].y;
		Z_Ave +=InputPointCloud[i].z;
	}

	X_Ave = X_Ave /InputPointCloud.size();
	Y_Ave = Y_Ave /InputPointCloud.size();
	Z_Ave = Z_Ave /InputPointCloud.size();

	//求方差
	for (int j = 0; j <InputPointCloud.size(); j++)
	{
		X_Var += (InputPointCloud[j].x - X_Ave)*(InputPointCloud[j].x - X_Ave);
		Y_Var += (InputPointCloud[j].y - Y_Ave)*(InputPointCloud[j].y - Y_Ave);
		Z_Var += (InputPointCloud[j].z - Z_Ave)*(InputPointCloud[j].z - Z_Ave);
	}
	X_Var = X_Var /InputPointCloud.size();
	Y_Var = Y_Var /InputPointCloud.size();
	Z_Var = Z_Var /InputPointCloud.size();

	//開始判斷是是否是離羣點，中間的閾值可以針對實際情況調整
	for (int k = 0; k <InputPointCloud.size(); k++)
	{
		if (((abs(InputPointCloud[k].z - Z_Ave) / sqrt(Z_Var))<1.5 || (abs(InputPointCloud[k].z + Z_Ave) / sqrt(Z_Var))<1.5)
			&& ((abs(InputPointCloud[k].y - Y_Ave) / sqrt(Y_Var))<1.8 || (abs(InputPointCloud[k].y + Y_Ave) / sqrt(Y_Var))<1.8)
			&& ((abs(InputPointCloud[k].x - X_Ave) / sqrt(X_Var))<1.8 || (abs(InputPointCloud[k].x + X_Ave) / sqrt(X_Var))<1.))
		{
			OutPointCloud.push_back(InputPointCloud[k]);                     //如果不是離羣點，則把該點放入CloudPoint中
		}
		
	}
	return;
}