Skip to content

Latest commit

 

History

History
45 lines (35 loc) · 1.88 KB

README.md

File metadata and controls

45 lines (35 loc) · 1.88 KB

1. Goal of this project

Goal of this project is to minimize run time and RMSE of ORB_SLAM2.

2. How to approach to achieve goal

My team focused on Frontend and Backend.

1. Search Bottleneck

First, using easy-profiler, we searched bottleneck of ORB_SLAM2.

Frontend : ORBextraction takes long time in Frontend
image

Backend : LoopClosing takes long time in Backend image

Loopclosing in Backend image

2. Adjust Parameter

Second, we decided to adjust parameter in Frontend and Backend.
In Frontend, we can adjust such as the number of features for ORB detection.
In Backend, we can adjust such as functions for non-linear optimization like Gauss-Newton, Levenberg, dogleg etc, Robust kernel like Huber, Cauchy etc, parameters of RANSAC etc.

3. Use CUDA

Third, we tried to apply cuda wherever it could be used.
We uses this open source.
go to the github for ORB_SLAM2_CUDA

3. Results

1. Frontend

image

2. Backend

image

4. How To Run

Please refer to the sites below.
ORB_SLAM2 : here
ORB_SLAM2_CUDA : here

$ ./build.sh
$ python3 load.py

In "load.py" script, you can modify parameters needed in Backend.
If you need to run CUDA, you should modify the command in "load.py" for CUDA.