ch7 pose_estimation_2d2d essential matrix 计算结果不一致

[yq-90]

这个问题可能重复 #277
跟着书上的代码，使用cv的orb特征提取，获得500个匹配，筛除hamming distance过小的剩余79个，与书上相同。但是最大的hamming distance我的cv库算出来是94而非95，图片显示如下，肉眼观察无问题。

到了pose estimation这里，代码如下，

void pose_estimate_2d2d(
        const std::vector<cv::KeyPoint> &kp1, const std::vector<cv::KeyPoint> &kp2,
        const std::vector<cv::DMatch> &matches, const cv::Mat &K, cv::Mat &R, cv::Mat &t)
{
    std::vector<cv::Point2f> p1;
    p1.reserve(4096);
    std::vector<cv::Point2f> p2;
    p2.reserve(4096);
    for (const cv::DMatch &m : matches) {
        p1.push_back(kp1[m.queryIdx].pt);
        p2.push_back(kp2[m.trainIdx].pt);
    }

    cv::Mat F;
    F = cv::findFundamentalMat(p1, p2, cv::FM_8POINT);
    std::cout << "Fundamental Matrix: " << std::endl << F << std::endl;

    cv::Mat E;
    E = cv::findEssentialMat(p1, p2, K);
    std::cout << "Essential Matrix: " << std::endl << E << std::endl;

    cv::Mat H;
    H = cv::findHomography(p1, p2, cv::RANSAC);
    std::cout << "Homograph Matrix: " << std::endl << H << std::endl;

    cv::recoverPose(E, p1, p2, K, R, t);

    std::cout << "K ^ -T * E * K ^ -1 = " << std::endl << K.inv().t() * E * K.inv() << std::endl;
}

参数列表中相对书上的代码多了一个相机内参K，数值来自ORB SLAM3的TUM FREIBURG2数据集：

K = 
520.909               0 325.141
        0       521.007 249.702
        0             0        1

算出来的Fundamental matrix

Fundamental Matrix: 
[4.54443750398184e-06, 0.0001333855576992603, -0.01798499246479044;
 -0.0001275657012964255, 2.266794804645652e-05, -0.01416678429206633;
 0.01814994639971766, 0.004146055870980492, 1]

与p175的结果误差不大，而essential matrix为

Essential Matrix: 
[-0.008457490262424407, 0.05444442307766559, 0.1546250154919883;
 -0.008221501658750767, 0.03351078406299837, -0.6896501808694672;
 -0.1153973780662328, 0.6945965301729999, 0.02159972069358522]

与书上结果毫无关联，甚至正负号都不一致，而计算homography则又与教材上近似：

Homograph Matrix: 
[0.9261214238501467, -0.1445322040811318, 33.26921163455131;
 0.04535424241276671, 0.9386696657491596, 8.570980719920279;
 -1.006198263331389e-05, -3.008140673707998e-05, 1]

而且，p175的验证E = t^R并不成立，t^R与E也有误差。
最后我的计算结果为：

R: 
[0.9956597556244213, -0.05614218770142942, 0.07422739245059702;
 0.052677474871967, 0.997465004120366, 0.04783982855603914;
 -0.07672505895097613, -0.04372208040859218, 0.9960931909282955]
t: 
[-0.9726822505199769;
 -0.2153697392331127;
 0.08663206649996272]

请问这些误差是正常的吗？