HBA代码解读（1）tools.hpp

其中，函数jr 和 jr_inv 暂时未找到出处。其他函数都已经注释。

#ifndef TOOLS_HPP
#define TOOLS_HPP

#include <Eigen/Core>
#include <unordered_map>
#include <pcl/point_cloud.h>
#include <pcl/point_types.h>
#include <math.h>

#define HASH_P 116101
#define MAX_N 10000000019
#define SMALL_EPS 1e-10
#define SKEW_SYM_MATRX(v) 0.0,-v[2],v[1],v[2],0.0,-v[0],-v[1],v[0],0.0

//方阵
#define PLM(a) vector<Eigen::Matrix<double, a, a>, Eigen::aligned_allocator<Eigen::Matrix<double, a, a>>>
//矢量
#define PLV(a) vector<Eigen::Matrix<double, a, 1>, Eigen::aligned_allocator<Eigen::Matrix<double, a, 1>>>
#define VEC(a) Eigen::Matrix<double, a, 1>

#define G_m_s2 9.81
#define DIMU 18
#define DIM 15
#define DNOI 12
#define NMATCH 5
#define DVEL 6

typedef pcl::PointXYZ PointType;
// typedef pcl::PointXYZI PointType;
// typedef pcl::PointXYZINormal PointType;
using namespace std;

Eigen::Matrix3d I33(Eigen::Matrix3d::Identity());
Eigen::Matrix<double, DIMU, DIMU> I_imu(Eigen::Matrix<double, DIMU, DIMU>::Identity());

// 根据XYZ值生成 hash位置
class VOXEL_LOC
{
public:
  int64_t x, y, z;

  VOXEL_LOC(int64_t vx = 0, int64_t vy = 0, int64_t vz = 0): x(vx), y(vy), z(vz){}

  bool operator == (const VOXEL_LOC &other) const
  {
    return (x == other.x && y == other.y && z == other.z);
  }
};

namespace std
{
  template<>
  struct hash<VOXEL_LOC>
  {
    size_t operator() (const VOXEL_LOC &s) const
    {
      using std::size_t; using std::hash;
      // return (((hash<int64_t>()(s.z)*HASH_P)%MAX_N + hash<int64_t>()(s.y))*HASH_P)%MAX_N + hash<int64_t>()(s.x);
      long long index_x, index_y, index_z;
			double cub_len = 0.125;
			index_x = int(round(floor((s.x)/cub_len + SMALL_EPS)));
			index_y = int(round(floor((s.y)/cub_len + SMALL_EPS)));
			index_z = int(round(floor((s.z)/cub_len + SMALL_EPS)));
			return (((((index_z * HASH_P) % MAX_N + index_y) * HASH_P) % MAX_N) + index_x) % MAX_N;
    }
  };
}

//矩阵元素和
double matrixAbsSum(Eigen::MatrixXd mat)
{
  double sum = 0.0;
  for (int i = 0; i < mat.rows(); i++)
    for (int j = 0; j < mat.cols(); j++)
      sum += fabs(mat(i, j));
  return sum;
}

// 误差柯西函数
double sigmoid_w(double r)
{
  return 1.0/(1+exp(-r));
}

// 向量变成旋转矩阵
Eigen::Matrix3d Exp(const Eigen::Vector3d &ang)
{
  double ang_norm = ang.norm();
  Eigen::Matrix3d Eye3 = Eigen::Matrix3d::Identity();
  if (ang_norm > 0.0000001)
  {
    Eigen::Vector3d r_axis = ang / ang_norm;
    Eigen::Matrix3d K;
    K << SKEW_SYM_MATRX(r_axis);
    /// Roderigous Tranformation
    return Eye3 + std::sin(ang_norm) * K + (1.0 - std::cos(ang_norm)) * K * K;
  }
  else
  {
    return Eye3;
  }
}
// 向量变成旋转矩阵
Eigen::Matrix3d Exp(const Eigen::Vector3d &ang_vel, const double &dt)
{
  double ang_vel_norm = ang_vel.norm();
  Eigen::Matrix3d Eye3 = Eigen::Matrix3d::Identity();

  if (ang_vel_norm > 0.0000001)
  {
    Eigen::Vector3d r_axis = ang_vel / ang_vel_norm;
    Eigen::Matrix3d K;

    K << SKEW_SYM_MATRX(r_axis);
    double r_ang = ang_vel_norm * dt;

    /// Roderigous Tranformation
    return Eye3 + std::sin(r_ang) * K + (1.0 - std::cos(r_ang)) * K * K;
  }
  else
  {
    return Eye3;
  }
}

// 旋转矩阵变成向量
Eigen::Vector3d Log(const Eigen::Matrix3d &R)
{
  double theta = (R.trace() > 3.0 - 1e-6) ? 0.0 : std::acos(0.5 * (R.trace() - 1));
  Eigen::Vector3d K(R(2,1) - R(1,2), R(0,2) - R(2,0), R(1,0) - R(0,1));
  return (std::abs(theta) < 0.001) ? (0.5 * K) : (0.5 * theta / std::sin(theta) * K);
}

// 构建v的反对称矩阵
Eigen::Matrix3d hat(const Eigen::Vector3d &v)
{
  Eigen::Matrix3d Omega;
  Omega <<  0, -v(2),  v(1)
      ,  v(2),     0, -v(0)
      , -v(1),  v(0),     0;
  return Omega;
}

Eigen::Matrix3d jr(Eigen::Vector3d vec)
{
  double ang = vec.norm();

  if(ang < 1e-9)
  {
    return I33;
  }
  else
  {
    vec /= ang;
    double ra = sin(ang)/ang;
    return ra*I33 + (1-ra)*vec*vec.transpose() - (1-cos(ang))/ang * hat(vec);
  }
}

Eigen::Matrix3d jr_inv(const Eigen::Matrix3d &rotR)
{
  Eigen::AngleAxisd rot_vec(rotR);
  Eigen::Vector3d axi = rot_vec.axis();
  double ang = rot_vec.angle();

  if(ang < 1e-9)
  {
    return I33;
  }
  else
  {
    double ctt = ang / 2 / tan(ang/2);
    return ctt*I33 + (1-ctt)*axi*axi.transpose() + ang/2 * hat(axi);
  }
}

// 主要用到了p和R
struct IMUST
{
  double t;
  Eigen::Matrix3d R;
  Eigen::Vector3d p;
  Eigen::Vector3d v;
  Eigen::Vector3d bg;
  Eigen::Vector3d ba;
  Eigen::Vector3d g;
  
  IMUST()
  {
    setZero();
  }

  IMUST(double _t, const Eigen::Matrix3d& _R, const Eigen::Vector3d& _p, const Eigen::Vector3d& _v,
        const Eigen::Vector3d& _bg, const Eigen::Vector3d& _ba,
        const Eigen::Vector3d& _g = Eigen::Vector3d(0, 0, -G_m_s2)):
        t(_t), R(_R), p(_p), v(_v), bg(_bg), ba(_ba), g(_g){}

  IMUST &operator+=(const Eigen::Matrix<double, DIMU, 1> &ist)
  {
    this->R = this->R * Exp(ist.block<3, 1>(0, 0));
    this->p += ist.block<3, 1>(3, 0);
    this->v += ist.block<3, 1>(6, 0);
    this->bg += ist.block<3, 1>(9, 0);
    this->ba += ist.block<3, 1>(12, 0);
    this->g += ist.block<3, 1>(15, 0);
    return *this;
  }

  Eigen::Matrix<double, DIMU, 1> operator-(const IMUST &b) 
  {
    Eigen::Matrix<double, DIMU, 1> a;
    a.block<3, 1>(0, 0) = Log(b.R.transpose() * this->R);
    a.block<3, 1>(3, 0) = this->p - b.p;
    a.block<3, 1>(6, 0) = this->v - b.v;
    a.block<3, 1>(9, 0) = this->bg - b.bg;
    a.block<3, 1>(12, 0) = this->ba - b.ba;
    a.block<3, 1>(15, 0) = this->g - b.g;
    return a;
  }

  IMUST &operator=(const IMUST &b)
  {
    this->R = b.R;
    this->p = b.p;
    this->v = b.v;
    this->bg = b.bg;
    this->ba = b.ba;
    this->g = b.g;
    this->t = b.t;
    return *this;
  }

  void setZero()
  {
    t = 0; R.setIdentity();
    p.setZero(); v.setZero();
    bg.setZero(); ba.setZero();
    g << 0, 0, -G_m_s2;
  }
};


// 复制四元数 q，t
void assign_qt(Eigen::Quaterniond& q, Eigen::Vector3d& t,
               const Eigen::Quaterniond& q_, const Eigen::Vector3d& t_)
{
  q.w() = q_.w(); q.x() = q_.x(); q.y() = q_.y(); q.z() = q_.z();
  t(0) = t_(0); t(1) = t_(1); t(2) = t_(2);
}

//该结构体主要服务于 downsample_voxel
struct M_POINT
{
	float xyz[3];
	int count = 0;
};

//计算每一个体素中的均值点
void downsample_voxel(pcl::PointCloud<PointType>& pc, double voxel_size)
{
	if (voxel_size < 0.01)
		return;

	std::unordered_map<VOXEL_LOC, M_POINT> feature_map;
	size_t pt_size = pc.size();  // pt_size 在此时是点个数

	for (size_t i = 0; i < pt_size; i++)
	{
		PointType &pt_trans = pc[i];
		float loc_xyz[3];
		for (int j = 0; j < 3; j++)
		{
			loc_xyz[j] = pt_trans.data[j] / voxel_size;  // 计算点所在体素
			if (loc_xyz[j] < 0)
				loc_xyz[j] -= 1.0;
		}

		VOXEL_LOC position((int64_t)loc_xyz[0], (int64_t)loc_xyz[1], (int64_t)loc_xyz[2]);  //计算所在体素索引
		auto iter = feature_map.find(position);
		if (iter != feature_map.end()) 
		{
			iter->second.xyz[0] += pt_trans.x;  //加入该点坐标
			iter->second.xyz[1] += pt_trans.y;
			iter->second.xyz[2] += pt_trans.z;
			iter->second.count++;
		}
		else
		{
			M_POINT anp;  // 第一个加入该体素的点
			anp.xyz[0] = pt_trans.x;
			anp.xyz[1] = pt_trans.y;
			anp.xyz[2] = pt_trans.z;
			anp.count = 1;
			feature_map[position] = anp;
		}
	}

	pt_size = feature_map.size();   // pt_size 在此时是体素个数
	pc.clear();
	pc.resize(pt_size);

	size_t i = 0;
	for (auto iter = feature_map.begin(); iter != feature_map.end(); ++iter) // 求得体素内的点的均值点
	{
		pc[i].x = iter->second.xyz[0] / iter->second.count;
		pc[i].y = iter->second.xyz[1] / iter->second.count;
		pc[i].z = iter->second.xyz[2] / iter->second.count;
		i++;
	}
}

/*
    rr是旋转矩阵，tt是平移矩阵
    pl1 是vector<Eigen::Vector3d>
    对自身进行RT变换
*/
void pl_transform(pcl::PointCloud<PointType> &pl1, const Eigen::Matrix3d &rr, const Eigen::Vector3d &tt)
{
  for(PointType &ap : pl1.points)
  {
    Eigen::Vector3d pvec(ap.x, ap.y, ap.z);
    pvec = rr * pvec + tt;
    ap.x = pvec[0];
    ap.y = pvec[1];
    ap.z = pvec[2];
  }
}

/*
    该函数是通过 IMUST 中的R,P把 porig中的三维点 变换到 ptran中
    PLV 是vector<Eigen::Vector3d>
    IMUST 中的 R和 P表示旋转和平移
*/
void plvec_trans(PLV(3) &porig, PLV(3) &ptran, IMUST &stat)
{
  uint asize = porig.size();
  ptran.resize(asize);
  for(uint i=0; i<asize; i++)
    ptran[i] = stat.R * porig[i] + stat.p;
}

/*
 VOX_FACTOR 

        VOX_FACTOR 的数据结构包含了一个协方差P，一个中心点V和一个点数N。

函数解释：

          1.1 void push(const Eigen::Vector3d &vec) 加入一个点并更新 P V N。

          1.2  cov() 就是计算协方差了

           1.3 VOX_FACTOR & operator+=(const VOX_FACTOR& sigv) 就是P V N 的加

          1.4  transform(const VOX_FACTOR &sigv, const IMUST &stat) 就是R和T对 P V N 的变化
*/
class VOX_FACTOR
{
public:
  Eigen::Matrix3d P;
  Eigen::Vector3d v;
  int N;

  VOX_FACTOR()
  {
    P.setZero();
    v.setZero();
    N = 0;
  }

  void clear()
  {
    P.setZero();
    v.setZero();
    N = 0;
  }

  void push(const Eigen::Vector3d &vec)
  {
    N++;
    P += vec * vec.transpose();
    v += vec;
  }

  Eigen::Matrix3d cov()
  {
    Eigen::Vector3d center = v / N;
    return P/N - center*center.transpose();
  }

  VOX_FACTOR & operator+=(const VOX_FACTOR& sigv)
  {
    this->P += sigv.P;
    this->v += sigv.v;
    this->N += sigv.N;

    return *this;
  }

  void transform(const VOX_FACTOR &sigv, const IMUST &stat)
  {
    N = sigv.N;
    v = stat.R*sigv.v + N*stat.p;
    Eigen::Matrix3d rp = stat.R * sigv.v * stat.p.transpose();
    P = stat.R*sigv.P*stat.R.transpose() + rp + rp.transpose() + N*stat.p*stat.p.transpose();
  }
};

//平面参数计算
const double threshold = 0.1;
bool esti_plane(Eigen::Vector4d& pca_result, const pcl::PointCloud<PointType>& point)
{
    Eigen::Matrix<double, NMATCH, 3> A;
    Eigen::Matrix<double, NMATCH, 1> b;
    b.setOnes();
    b *= -1.0f;   // AX + BY + CZ +D =0； 其中D等于 1

    for (int j = 0; j < NMATCH; j++)
    {
        A(j, 0) = point[j].x;  // A的系数
        A(j, 1) = point[j].y;  // B的系数
        A(j, 2) = point[j].z;  // C的系数
    }

    Eigen::Vector3d normvec = A.colPivHouseholderQr().solve(b); // A B C 参数

    for (int j = 0; j < NMATCH; j++)
    {
        if (fabs(normvec.dot(A.row(j)) + 1.0) > threshold)  // 点到面的距离 大于 阈值（0.1）
            return false;  // 表示不是一个平面
    }

    double n = normvec.norm();   // sqrt(A*A + B*B + C*C)
    pca_result(0) = normvec(0) / n;  // A
    pca_result(1) = normvec(1) / n; // B
    pca_result(2) = normvec(2) / n; // C
    pca_result(3) = 1.0 / n;  // D = 1.0
    return true;
}

#endif