工作相关论文集锦。

Contents

[TOC]

Topics

Artificial Intelligence

Network Architecture

Attention Is All You Need / mirror

See for more details.

Calibrate

Camera Only

1. Research on Lane-Marking Line Based Camera Calibration / mirror

论文通过设计坐标系变换顺序巧妙地推导出俯仰角、偏航角关于消失点坐标的解析解，其局限为未考虑滚转角，即假定滚转角为 $0\degree$，此外该推导也难以考虑相机畸变带来的影响。

1.1 infer

如上图所示，$O_CX_CY_CZ_C$ 是相机坐标系；$O_MX_MY_MZ_M$ 是以相机光轴与地平面交点为原点建立的过渡坐标系，该过渡坐标系绕 $X_M$ 轴旋转 $\theta$ 再平移 $S$ 可得到相机坐标系；定义世界坐标系 $O_WX_WY_WZ_W$ 的 $Y_W$ 轴与车道线平行，世界坐标系绕 $Z_W$ 旋转 $\beta$ 可得到过渡坐标系。

上述变换关系可用如下矩阵运算描述：

$$ \begin{align} \left[ \begin{array}{c} u \ v \ 1 \end{array} \right] &= \left[ \begin{array}{ccc} f_x & 0 & c_x \ 0 & f_y & c_y \ 0 & 0 & 1 \ \end{array} \right] \cdot \frac{1}{z_c} \left[ \begin{array}{c} x_c \ y_c \ z_c \end{array} \right] \Leftarrow \left[ \begin{array}{cccc} 1 & 0 & 0 & 0\ 0 & cos\theta & sin\theta & 0\ 0 & -sin\theta & cos\theta & S\ \end{array} \right] \cdot \left[ \begin{array}{c} x_m \ y_m \ z_m \ 1 \end{array} \right] \Leftarrow \left[ \begin{array}{cccc} cos\beta & sin\beta & 0 & 0\ -sin\beta & cos\beta & 0 & 0\ 0 & 0 & 1 & 0\ 0 & 0 & 0 & 1\ \end{array} \right] \cdot \left[ \begin{array}{c} x_w \ y_w \ z_w \ 1 \end{array} \right] \ &= \left[ \begin{array}{ccc} f_x & 0 & c_x \ 0 & f_y & c_y \ 0 & 0 & 1 \ \end{array} \right] \cdot \frac{1}{z_c} \cdot \left[ \begin{array}{c} x_m\ y_mcos\theta+z_msin\theta\ -y_msin\theta+z_mcos\theta+S\ \end{array} \right] \Leftarrow \left[ \begin{array}{c} x_wcos\beta+y_wsin\beta \ -x_wsin\beta+y_wcos\beta \ z_w \ 1 \end{array} \right] \ &= \left[ \begin{array}{ccc} f_x & 0 & c_x \ 0 & f_y & c_y \ 0 & 0 & 1 \ \end{array} \right] \cdot \left[ \begin{array}{c} \frac{x_m}{-y_msin\theta+z_mcos\theta+S}\ \frac{y_mcos\theta+z_msin\theta}{-y_msin\theta+z_mcos\theta+S}\ 1\ \end{array} \right] \end{align} $$

$$ \begin{align} \lim\limits_{point \to vp} \frac{x_m}{y_m} &= \lim\limits_{y_w \to \infty} \frac{x_wcos\beta+y_wsin\beta}{-x_wsin\beta+y_wcos\beta} \ &= \lim\limits_{y_w \to \infty} \frac{\frac{x_w}{y_w}+tan\beta}{-\frac{x_w}{y_w}tan\beta+1} \ &= tan\beta \end{align} $$

$$ \begin{align} \lim\limits_{point \to vp} \left( \left[ \begin{array}{c} u \ v \end{array} \right] \right)

\left[ \begin{array}{c} u_{vp} \ v_{vp} \end{array} \right] &= \lim\limits_{y_m \to \infty} \left( \left[ \begin{array}{c} f_x\frac{x_m}{-y_msin\theta+z_mcos\theta+S}+c_x\ f_y\frac{y_mcos\theta+z_msin\theta}{-y_msin\theta+z_mcos\theta+S}+c_y \end{array} \right] \right) \ &= \lim\limits_{y_m \to \infty} \left( \left[ \begin{array}{c} f_x\frac{\frac{x_m}{y_m}}{-sin\theta+\frac{z_m}{y_m}cos\theta+\frac{S}{y_m}}+c_x\ f_y\frac{1+\frac{z_m}{y_m}tan\theta}{-tan\theta+\frac{z_m}{y_m}+\frac{S}{y_mcos\theta}}+c_y \end{array} \right] \right) \ &= \left[ \begin{array}{c} f_x\frac{tan\beta}{-sin\theta}+c_x\ \frac{f_y}{-tan\theta}+c_y \end{array} \right] \ \end{align} $$

PS: 事实上过渡坐标系和世界坐标系并不一定需要选择在地平面上，将它们的原点平移到相机系原点上述结论同样成立。

2. Pitch angle estimation using a Vehicle-Mounted monocular camera for range measurement / mirror

2.1 estimate pitch angle

1. 想象一块平面，固定于汽车的相机运动于其上，则相机的运动矢量平行于地平面（无穷远处的地平面呈现在相机中的就是地平线）
2. 俯仰（Pitch）角就是相机光轴与地平面的夹角
3. 假设相机光轴与地平面平行，此时相机运动矢量落于与地平面平行的平面
4. 当光轴与地平面不平行时，运动矢量与相机光轴平面的夹角是俯仰角（此处需斟酌，二面角及其平面角、旋转角顺序、俯仰角定义）

2.2 estimate pitch angle rate

Camera Lidar

Lane Line

Fitting

1. End-to-end Lane Shape Prediction with Transformers / mirror

Object

Distance Meansurement

1. Vehicle distance estimation using a mono-camera for FCW/AEB systems / mirror

2. Real-time vehicle distance estimation using single view geometry / mirror

Traffic Sign

FailSafe

3D Reconstruction

basic

1. An efficient solution to the five-point relative pose problem / mirror

The cheirality constraint resolves the ambiguity unless all visible points are closer to one viewpoint than the other

该方法不能消除所有可见点都更靠近一个相机（参见原文第四节），论坛也有人提及

VSLAM

SFM

Appendix

Reference

Archive