图像拼接（九）：双摄像头实时视频拼接（单应变换模型）

来源：互联网发布：淘宝手机市场编辑：程序博客网时间：2024/05/18 01:58

单应变换相比平移变换，具有更广泛的场景适应性，但同时稳定性会有一定程度下降。

设计到的技术细节有：

特征检测与描述
特征匹配与单应矩阵估计
opencv采集视频
渐入渐出图像融合

这个解决方案的硬件条件包括：有两个USB接口的计算机，两个合理放置的USB摄像头。

合理放置是指：两个摄像头分隔一定夹角，相机中心相距接近，所拍摄场景有足够的重叠部分。以上保证了单应变换的可用性。

代码实现：

#include "opencv2/core/core.hpp"#include "opencv2/highgui/highgui.hpp"#include "opencv2/imgproc/imgproc.hpp"# include "opencv2/features2d/features2d.hpp"#include"opencv2/nonfree/nonfree.hpp"#include"opencv2/calib3d/calib3d.hpp"#include<iostream>using namespace cv;using namespace std;int main(){    VideoCapture cap1(0);    VideoCapture cap2(1);    double rate = 60;    int delay = 1000 / rate;    bool stop(false);    Mat img1;    Mat img2;    Mat result;    int d = 200;//渐入渐出融合宽度    Mat homography;    int k = 0;    namedWindow("cam1", CV_WINDOW_AUTOSIZE);    namedWindow("cam2", CV_WINDOW_AUTOSIZE);    namedWindow("stitch", CV_WINDOW_AUTOSIZE);    if (cap1.isOpened() && cap2.isOpened())    {        cout << "*** ***" << endl;        cout << "摄像头已启动！" << endl;    }    else    {        cout << "*** ***" << endl;        cout << "警告：请检查摄像头是否安装好!" << endl;        cout << "程序结束！" << endl << "*** ***" << endl;        return -1;    }    cap1.set(CV_CAP_PROP_FOCUS, 0);    cap2.set(CV_CAP_PROP_FOCUS, 0);    while (!stop)    {        if (cap1.read(img1) && cap2.read(img2))        {            imshow("cam1", img1);            imshow("cam2", img2);            //彩色帧转灰度            //cvtColor(img1, img1, CV_RGB2GRAY);            //cvtColor(img2, img2, CV_RGB2GRAY);            //计算单应矩阵            if (k < 1 || waitKey(delay) == 13)            {                cout << "正在匹配..." << endl;                ////////////////////////////////                vector<KeyPoint> keypoints1, keypoints2;                //构造检测器                //Ptr<FeatureDetector> detector = new ORB(120);                Ptr<FeatureDetector> detector = new SIFT(80);                detector->detect(img1, keypoints1);                detector->detect(img2, keypoints2);                //构造描述子提取器                Ptr<DescriptorExtractor> descriptor = detector;                //提取描述子                Mat descriptors1, descriptors2;                descriptor->compute(img1, keypoints1, descriptors1);                descriptor->compute(img2, keypoints2, descriptors2);                //构造匹配器                BFMatcher matcher(NORM_L2, true);                //匹配描述子                vector<DMatch> matches;                matcher.match(descriptors1, descriptors2, matches);                vector<Point2f> selPoints1, selPoints2;                vector<int> pointIndexes1, pointIndexes2;                for (vector<DMatch>::const_iterator it = matches.begin(); it != matches.end(); ++it)                {                    selPoints1.push_back(keypoints1.at(it->queryIdx).pt);                    selPoints2.push_back(keypoints2.at(it->trainIdx).pt);                }                vector<uchar> inliers(selPoints1.size(), 0);                homography = findHomography(selPoints1, selPoints2, inliers, CV_FM_RANSAC, 1.0);                //根据RANSAC重新筛选匹配                vector<DMatch> outMatches;                vector<uchar>::const_iterator itIn = inliers.begin();                vector<DMatch>::const_iterator itM = matches.begin();                for (; itIn != inliers.end(); ++itIn, ++itM)                {                    if (*itIn)                    {                        outMatches.push_back(*itM);                    }                }                k++;                //画出匹配结果                //Mat matchImage;                //drawMatches(img1, keypoints1, img2, keypoints2, outMatches, matchImage, 255, 255);                //imshow("match", matchImage);                ///////////////////////////////////////////////////////////////////////            }            //拼接            double t = getTickCount();            warpPerspective(img1, result, homography, Size(2 * img1.cols-d, img1.rows));//Size设置结果图像宽度，宽度裁去一部分，d可调            Mat half(result, Rect(0, 0, img2.cols - d, img2.rows));            img2(Range::all(), Range(0, img2.cols - d)).copyTo(half);            for (int i = 0; i < d; i++)            {                result.col(img2.cols - d + i) = (d - i) / (float)d*img2.col(img2.cols - d + i) + i / (float)d*result.col(img2.cols - d + i);            }            imshow("stitch", result);            t = ((double)getTickCount() - t) / getTickFrequency();            //cout << t << endl;        }        else        {            cout << "----------------------" << endl;            cout << "waitting..." << endl;        }        if (waitKey(1) == 27)        {            stop = true;            cout << "程序结束！" << endl;            cout << "*** ***" << endl;        }    }    return 0;}

实验效果：

这里写图片描述

上述视频是用录屏软件录制的，分辨率会有下降。实际测试中，直接观察显示良好。两幅输入的源图像均为640*480分辨率，能够做到实时的实现。在我的具有i3处理器配置的笔记本上运行，拼接图像显示间隔为0.10″~0.12″。

0 0