雙目立體視覺匹配演算法之視差圖disparity計算——SAD演算法、SGBM演算法
一、SAD演算法
1.演算法原理
SAD(Sum of absolute differences)是一種影象匹配演算法。基本思想:差的絕對值之和。此演算法常用於影象塊匹配,將每個畫素對應數值之差的絕對值求和,據此評估兩個影象塊的相似度。該演算法快速、但並不精確,通常用於多級處理的初步篩選。
2.基本流程
輸入:兩幅影象,一幅Left-Image,一幅Right-Image
對左圖,依次掃描,選定一個錨點:
(1)構造一個小視窗,類似於卷積核;
(2)用視窗覆蓋左邊的影象,選擇出視窗覆蓋區域內的所有畫素點;
(3)同樣用視窗覆蓋右邊的影象並選擇出覆蓋區域的畫素點;
(4)左邊覆蓋區域減去右邊覆蓋區域,並求出所有畫素點灰度差的絕對值之和;
(5)移動右邊影象的視窗,重複(3)-(4)的處理(這裡有個搜尋範圍,超過這個範圍跳出);
(6)找到這個範圍內SAD值最小的視窗,即找到了左圖錨點的最佳匹配的畫素塊。
3、程式碼
#include<iostream> #include<opencv2/opencv.hpp> using namespace std; using namespace cv; class SAD { private: int winSize;//卷積核尺寸 int DSR;//視差搜尋範圍 public: SAD() :winSize(7), DSR(30){} SAD(int _winSize, int _DSR) :winSize(_winSize), DSR(_DSR){} Mat computerSAD(Mat&L, Mat&R);//計算SAD }; Mat SAD::computerSAD(Mat&L, Mat&R) { int Height = L.rows; int Width = L.cols; Mat Kernel_L(Size(winSize, winSize), CV_8U, Scalar::all(0)); //CV_8U:0~255的值,大多數影象/視訊的格式,該段設定全0矩陣 Mat Kernel_R(Size(winSize, winSize), CV_8U, Scalar::all(0)); Mat Disparity(Height, Width, CV_8U, Scalar(0)); for (int i = 0; i < Width - winSize; ++i){ for (int j = 0; j < Height - winSize; ++j){ Kernel_L = L(Rect(i, j, winSize, winSize));//L為做影象,Kernel為這個範圍內的左圖 Mat MM(1, DSR, CV_32F, Scalar(0));//定義匹配範圍 for (int k = 0; k < DSR; ++k){ int x = i - k; if (x >= 0){ Kernel_R = R(Rect(x, j, winSize, winSize)); Mat Dif; absdiff(Kernel_L, Kernel_R, Dif); Scalar ADD = sum(Dif); float a = ADD[0]; MM.at<float>(k) = a; } Point minLoc; minMaxLoc(MM, NULL, NULL, &minLoc, NULL); int loc = minLoc.x; Disparity.at<char>(j, i) = loc * 16; } double rate = double(i) / (Width); cout << "已完成" << setprecision(2) << rate * 100 << "%" << endl; } } return Disparity; } int main() { Mat left = imread("Left.png"); Mat right = imread("Right.png"); //-------影象顯示----------- namedWindow("leftimag"); imshow("leftimag", left); namedWindow("rightimag"); imshow("rightimag", right); //--------由SAD求取視差圖----- Mat Disparity; SAD mySAD(7, 30); Disparity = mySAD.computerSAD(left, right); //-------結果顯示------ namedWindow("Disparity"); imshow("Disparity", Disparity); //-------收尾------ waitKey(0); return 0; }
4、結果
左圖:
右圖:
視差圖結果:
二、SGBM演算法
1、SGBM演算法作為一種全域性匹配演算法,立體匹配的效果明顯好於區域性匹配演算法,但是同時複雜度上也要遠遠大於區域性匹配演算法。演算法主要是參考Stereo Processing by Semiglobal Matching and Mutual Information。
opencv中實現的SGBM演算法計算匹配代價沒有按照原始論文的互資訊作為代價,而是按照塊匹配的代價。
對該演算法的具體講解可以參考:https://www.cnblogs.com/hrlnw/p/4746170.html。
參考:http://www.opencv.org.cn/forum.php?mod=viewthread&tid=23854。
2、程式碼
#include<iostream>
#include<opencv2/opencv.hpp>
using namespace std;
using namespace cv;
class SGBM
{
private:
enum mode_view { LEFT, RIGHT };
mode_view view; //輸出左視差圖or右視差圖
public:
SGBM() {};
SGBM(mode_view _mode_view) :view(_mode_view) {};
~SGBM() {};
Mat computersgbm(Mat &L, Mat &R); //計算SGBM
};
Mat SGBM::computersgbm(Mat &L, Mat &R)
/*SGBM_matching SGBM演算法
*@param Mat &left_image :左影象
*@param Mat &right_image:右影象
*/
{
Mat disp;
int numberOfDisparities = ((L.size().width / 8) + 15)&-16;
Ptr<StereoSGBM> sgbm = StereoSGBM::create(0, 16, 3);
sgbm->setPreFilterCap(32);
int SADWindowSize = 5;
int sgbmWinSize = SADWindowSize > 0 ? SADWindowSize : 3;
sgbm->setBlockSize(sgbmWinSize);
int cn = L.channels();
sgbm->setP1(8 * cn*sgbmWinSize*sgbmWinSize);
sgbm->setP2(32 * cn*sgbmWinSize*sgbmWinSize);
sgbm->setMinDisparity(0);
sgbm->setNumDisparities(numberOfDisparities);
sgbm->setUniquenessRatio(10);
sgbm->setSpeckleWindowSize(100);
sgbm->setSpeckleRange(32);
sgbm->setDisp12MaxDiff(1);
Mat left_gray, right_gray;
cvtColor(L, left_gray, CV_BGR2GRAY);
cvtColor(R, right_gray, CV_BGR2GRAY);
view = LEFT;
if (view == LEFT) //計算左視差圖
{
sgbm->compute(left_gray, right_gray, disp);
disp.convertTo(disp, CV_32F, 1.0 / 16); //除以16得到真實視差值
Mat disp8U = Mat(disp.rows, disp.cols, CV_8UC1);
normalize(disp, disp8U, 0, 255, NORM_MINMAX, CV_8UC1);
imwrite("results/SGBM.jpg", disp8U);
return disp8U;
}
else if (view == RIGHT) //計算右視差圖
{
sgbm->setMinDisparity(-numberOfDisparities);
sgbm->setNumDisparities(numberOfDisparities);
sgbm->compute(left_gray, right_gray, disp);
disp.convertTo(disp, CV_32F, 1.0 / 16); //除以16得到真實視差值
Mat disp8U = Mat(disp.rows, disp.cols, CV_8UC1);
normalize(disp, disp8U, 0, 255, NORM_MINMAX, CV_8UC1);
imwrite("results/SGBM.jpg", disp8U);
return disp8U;
}
else
{
return Mat();
}
}
int main()
{
Mat left = imread("Left.png");
Mat right = imread("Right.png");
//-------影象顯示-----------
namedWindow("leftimag");
imshow("leftimag", left);
namedWindow("rightimag");
imshow("rightimag", right);
//--------由SAD求取視差圖-----
Mat Disparity;
SGBM mySGBM;
Disparity = mySGBM.computersgbm(left, right);
//-------結果顯示------
namedWindow("Disparity");
imshow("Disparity", Disparity);
//-------收尾------
waitKey(0);
return 0;
}
3、結果
所用的左右圖同上,所得結果為:
NB:對於使用的其他演算法本次沒有實驗,故沒有介紹,可以參考:https://blog.csdn.net/liulina603/article/details/53302168