Description

The are two folders in the package

1. lib for segmentation implemented by CUDA
2. Sample code can show the lib usage and also be used to check perf and accuracy by comparing its output with PCL
3. A point clound: sample.pcd which has 119978 points

Prerequisites

1. Install Jetpack4.4.1 by SDKManager

2. install PCL (Eigen included)

$sudo apt-get update
$sudo apt-get install libpcl-dev

Build

$ make

Run

Please boost CPU and GPU firstly

sudo nvpmodel -m 0
sudo jetson_clocks 

Usage:

./demo src.pcd  dst.pcd

How to check output

We can get output as below:

------------checking CUDA Approx nearest search---------------- 
CUDA costs : 2.55582 ms.
Point distance AVG: 0.721553

------------checking CUDA radiusSearch---------------- 
CUDA costs: 0.083172 ms.
Points selected: 4751

------------checking OC-Tree creating ---------------- 
PCL(CPU) create oc-tree by Time: 4.35747 ms.

------------checking PCL(CPU)  Approx nearest search ---------------- 
PCL(CPU) costs: 11.6701 ms.
Point distance AVG: 2.75023

------------checking PCL(CPU) radiusSearch ---------------- 
PCL(CPU) costs by Time: 1.29362 ms.
Points selected: 4751

Perforamnce table

Approx nearest          GPU             CPU
points count of tree    7000            7000
points count of target  7000            7000
Distance Error          0.721           2.75
cost time(ms)           2.55            11.67

radiusSearch            GPU             CPU
points count of tree    7000            7000
points selected         4751            4751
cost time(ms)           0.083           1.29

How To Check the Version of the Lib

$ strings lib* | grep version | grep lib<br>
lib* version: 1.0 Jun  2 2019 09:30:19<br>

Test Enviroment

Jetson Xavier AGX 8GB
Jetpack 4.4.1
CUDA 10.2
PCL 1.8
Eigen 3