1 Introduction

1. multiple echoes are acquired for a single emitted pulse and the echo caused by the object of interest is picked.
2. Generally in single-echo approaches, only the strongest echo (red) is available.
3. Thus, crucial information about the object of interest can be lost.
4. I and R denote intensity and range, respectively.

LiDAR

But most echoes do not represent any physical object, but rather are artifacts from reflections or refractions.

How to identify the echo represent a object rather than noise?

2 Method

1. re-position each query point, according to its knn neighbors of strongest echo points
   1. query point can be strongest echo point or Last/2nd-strongest echo point;
   2. all knn neighbors of the query is from the strongest echo points;
   3. ”Exclude self – Include neighbors” simply discards KNN queries and keeps the values. From this, the coordinate learner predicts the new positions of queries.
2. The point cloud on the right illustrates the final output of our method, where substitutes are points from alternative echoes. ???
3. favor strongest echo + not favor isolated points?
   1. is the 3d knn ok? I think 2d knn is better.

2.0 Terms

1. single-echo point cloud:

   

   Np=HxW: the number of points

   Nc: the number of points and channels.

2. A multi-echo point cloud

   

   Ne: the number of echoes per emitted laser pulse.

3. Γ: spherical projection of the point cloud