search

wh200720041 / floam

Fast LOAM: Fast and Optimized Lidar Odometry And Mapping for indoor/outdoor localization IROS 2021
Other
1.07k stars 278 forks source link

Tradeoff between performance and accuracy #1

Closed Ckerrr closed 4 years ago

Ckerrr commented 4 years ago

Hello Han,

Thanks for your great work and sharing with the community.

As is mentioned by you, the performance optimization is one of the highlights from FLOAM. Is there a configurable or easy way by tuning the parameters in order to improve accuracy if we don't expect it to be real-time (eg. change num of features, resolution, num of iterations in optimization, etc)?

Looking forward to your reply!

My best

Calvin

wh200720041 commented 4 years ago

Hi Calvin:

A quick answer to your question, tuning parameters can improve accuracy only by a little, but at huge computational cost, which is not worth.

I have tested different parameters, but it seems that the localization accuracy can improve like 3-5% at most, which is not significant in practice. The parameters you pointed out are correct, you may test with different params (num of features, resolution, num of iterations) if you are interested, but the result shouldn't varies too much.

I think if you would like to improve the performance, it is still a quite challenging research problem. For examples, add dynamic object filtering or semantic recognition to existing work. I think this may takes quite long time.

Let me know if you if you have any questions.

Regards Han