search

UMich-BipedLab / LiDARTag

This is a package for LiDARTag, described in paper: LiDARTag: A Real-Time Fiducial Tag System for Point Clouds
https://ieeexplore.ieee.org/abstract/document/9392337
GNU Affero General Public License v3.0
255 stars 50 forks source link

Limitations of tag extraction method #4

Closed plumewind closed 4 years ago

plumewind commented 4 years ago

hello，

I test the lidartag in a corridor, the tag is placed in the center, but the extracted cluster often contains points on the wall, like the picture (I displayed the clusters one by one).

2020-04-27 23-40-31屏幕截图

how to change some paramters?

I think the _updateCluster function, will make a cluster bigger and biger in a bad trend，I want to improve it, have you some advice?

Thanks a lot!

brucejk commented 4 years ago

I would be happy to help!

First of all, it would be better if you place the tag around 45 ^\circ, even though it is not relevant to the problem.

Secondly, I have not tested the package in a narrow aisle; I tested in a clustered laboratory. However, it is indeed a scenario that we should consider and the package should also be able to handle. I noticed there are only two rings on the wall got included. What is the distance from the wall to the tag and the size of the tag?

Thirdly, if you could upload the bagfile to a google drive folder, I can further look into this and get back to you asap! You can email the link to me if you like!

Lastly, this is an on-going project and we are currently improving the clustering algorithm to make it more robust. Please bare with us!

plumewind commented 4 years ago

Sorry, I can not send my data to you because of the wall.

But I find a big problem, I carefully considered the idea of cluster expansion algorithm. I found that the tag plane must be parallel to the lidar coordinate system yoz plane, so that the generation of clusters is reasonable. However, if it is not, there will be a big problem in this cluster generation.

brucejk commented 4 years ago

It is okay that you can't share the data; I will still try to help based on what you described.

Could you illustrate more on ' the tag plane must be parallel to the lidar coordinate system yoz plane'? The cluster algorithm works fine even if the tag does not being vertical to the ground plane (x-y plane, as z-axis pointing up to the air). Could you tell me what the tag size is and what the distance from the wall to the tag?

plumewind commented 4 years ago

hello，a pcd is ok？

small-lidartag.zip

the tag is 16h11, size about 0.3*0.3, but no rings , I add a function:

lidar_param.zip

plumewind commented 4 years ago

hello,

I make a rosbag for lidartag, about 2.7s .(too big to send successfully)

lidar_1smalltag_rosbag.zip

Thanks a lot.

brucejk commented 4 years ago

Hi,

I have discussed it with my labmate. We concluded that you might need to make the labeling process more robust. For example, 18 index:177 x: 2.60129 y: -2.81091 z: -0.0231521 18 index:178 x: 2.58906 y: -2.81539 z: -0.0231215 18 index:198 x: 2.40824 y: -2.97797 z: -0.0231521 18 index:199 x: 2.402 y: -2.98942 z: -0.0231826 18 index:1809 x: 2.58659 y: -2.80341 z: -0.0441436 18 index:1810 x: 2.58115 y: -2.81521 z: -0.0442021 18 index:1830 x: 2.39094 y: -2.96561 z: -0.0440851 18 index:1831 x: 2.38786 y: -2.98092 z: -0.0442021 These points are labeled as ring 18 according to the rslidar32_addRing function. However, we can tell by the z-axis that they do not belong to the same ring. This causes a lot of issues for the following clustering algorithm because we heavily rely on the correctness of the ring number.

The reason why some part of the wall is included is related to the labeling process. In the case of ring 18, these points actually contain all the points from this ring and the adjacent ring. All the points indexed between 177 and 1831 are included in the cluster.

Please let me know if you have other questions!

plumewind commented 4 years ago

Thanks a lot !

I will check it right now.