- 双目估计深度;
- 基于深度得到三维点云XYZ;
- 基于单目相邻帧求解摄像头位姿:
- 创建样本集合:X,Y。
\(
X^T = \left[\begin{aligned}x_1\\y_1\\z_1\\1\end{aligned}\right] \\
Y^T = \left[\begin{aligned}x_2\\y_2\\z_2\end{aligned}\right] \\
AX^T=Y^T \rightarrow XA^T=Y\rightarrow (X^TX)A^T=X^TY\rightarrow A^T=(X^TX)^{-1}X^TY\\
\text{let } P = {(X^TX)}^{-1}X^T\\
A = (PY)^T=({(X^TX)}^{-1}X^TY)^T.\\
\text{let } S = \left[ \begin{aligned} A && \mathbf{0}\\ \mathbf{0} && I \end{aligned} \right]
\)
- 对齐坐标系到初始启动位姿:
\(
S_0\leftarrow S\cdot S_0
\)
- 为了实时画出摄像头的轨迹,需要获得摄像头的位姿相对于初始坐标系的状态:
\(
\text{let }\mathbb{N} \text{ be the basis of camera under coordinate of } A_0 \\
\mathbb{N}_0 = \left[\begin{aligned} 1 && 0 && 0 \\ 0 && 1 && 0 \\ 0 && 0 && 1 \\ 1 && 1 && 1 \end{aligned}\right]\\
\text{then }\\
\quad \mathbb{N} = {S_0}^{-1}\mathbb{N}_0
\)
posted @
2022-06-10 12:42
xchk138
阅读(
167)
评论()
收藏
举报