Multi sensor fusion positioning (I) -- 3D laser odometer

Catalog

One 、 The overall process of point cloud map

  Two 、 Laser odometer scheme

2.1 ICP Point to point

2.1.1 ICP deduction

2.1.2 ICP improvement

2.2 NDT

2.2.1 NDT deduction

2.2.2 NDT improvement

2.3 LOAM system  

2.3.1 LOAM

2.3.2 A-LOAM

2.3.3 LEGO-LOAM

 2.4 Data set and evaluation index

2.4.1 KITTI brief introduction

2.4.2 indicators

One 、 The overall process of point cloud map

  Two 、 Laser odometer scheme

2.1 ICP Point to point

The initial posture is required to be better

 

2.1.1 ICP deduction

  The core idea ： Solve the rotation and translation separately , First get the rotation, and then get the translation value according to the rotation

Trace Is the value of the diagonal ,  Exchange multiplication order , Trace invariant

  After solving , To iterate

2.1.2 ICP improvement

ICP It's a hypothetical model , Point to point itself is a fuzzy solution , A pile of point clouds hitting the same place cannot completely coincide . 

2.2 NDT

NDT What is calculated is the difference between the distribution in each grid of two point clouds , High robustness  

2.2.1 NDT deduction

Just calculate the characteristic distribution of the first point cloud , Instead of calculating the second point cloud every iteration , You only need to calculate that the probability of the second point cloud on the first point cloud grid is the greatest . 

 

2.2.2 NDT improvement

2.3 LOAM system  

2.3.1 LOAM

1、 The extracted features  

1） Split by harness

2、 Inter frame matching

1） Define point to line , Distance from point to face

2） Look for and Rt The connection of

The most perfect situation is through rotation and Translation , Click on the line , Point on the surface .

3）LM Optimize

3、 Merge map points

 

2.3.2 A-LOAM

F-LOAM

2.3.3 LEGO-LOAM

 2.4 Data set and evaluation index

2.4.1 KITTI brief introduction

2.4.2 indicators