Simultaneous Localisation and Mapping for android

Question

I recently started exploring the world of visual recognition and was particularly interested in SLAM. I have tried and tested different SLAM implementations on a laptop from openslam.org I'm still new to the field and still learning about it. I want to implement SLAM on an android device. Can anyone point me to any such implementation. Or can you suggest which implementation of SLAM would work best on an android device. I have access to top of the line android devices such as the Galaxy S6 or the Nexus 5.

Before starting to work on my idea I just wanted to know which implementation would work best in terms of efficiency and accuracy on an android device.

Thankyou

Erbium

Answer 1

Those guys are quite good, the source is available under GPLv3 license (in user friendly form) and a friend of mine was able to run it on android in real-time (It was developed for MAVs and they state that it should reach 50FPS on current embedded computers).

https://github.com/uzh-rpg/rpg_svo

(Check out the video on the homepage as well)

Drogin

Thankyou so much. This would greatly help in getting started. The performance of the MAV is very impressive in the video. – Erbium

gfkri, @ShayanJalil I'm working on this project as well! Can you give me some hints or reference to get inspiration for android porting of SVO? – Pare

Answer 2

orbslam is a open source slam system basing on sparsity method.

on pc orbslam can reach about 35 FPS，on smartphone it can‘t perform like on pc， so you can't move the phone too fast. That is because the feature extraction process of orbslam take so much time.

svo is a good choice as for speed, however, svo is only a visual odometry, it dosen't have a loop closing module, therefore when run for a long time, the drift may be large.

therefore, i recommend orbslam.

many people have migrate the code to android, you can search on github:https://github.com/search?q=orb-slam+android&ref=opensearch

Ventricle

@ Abraham Uribe i've add a more detail answer. however, the link is necessary and it's just a part of my answer. firstly, i recommend orbslam, them there are many implementation open source on github, the link is the search result, it's not the entire content of my answer. – Ventricle