Trouble with Scan Alignment – Registration Errors in Multi-Scan LiDAR Data

Posted by: @subhabrata

↑

Hey everyone,

Currently, I am working on a project in which I need to combine scans from different positions for a LiDAR scan. As a result, the final point cloud has gaps and misalignments.

I have tried both target-based and targetless registration methods on a terrestrial LiDAR scanner, but drift and misalignment persist.

Has anyone else faced this issue? Any tips on improving registration accuracy?

Thanks in advance!

Least squares analysis.

When we need a tight registration we use LSA on scanner observations, in this case we use a Trimble SX12 and Topcon GLS scanners, all observations including scanners and differential levelling are analysed with Salsa.  

Is this a linear project?   Whats the overlap percentage. Linear project are rough for terrestrial scanners especially if moving vehicles and stop lights etc are involved and the road way doesn’t have much change for different planes etc. power poles all look the same etc. I would go back and really look at the point clouds scanner station to scanner station in pairs. Any gaps overlaps try and add another scan station or two in those areas maybe set a box or other items in as well or targets where geographically are needed to help aid in the registration.  Also might set some additional control for geo reference purposes. Might have to break it up in sections along the route as you register and then over lap with the next section to keep the alignment staying within specs instead of one end to the other. Also for sure make sure you account for the control you are geo reference to as small uncertainty can add up the same for the scan registration.  Sometimes we’ll often dealing with scan registration math vs visual inspection. I always say scanning sometimes even when the math looks good sometimes it just flat out is wrong.  Visual inspection of point clouds is my go to for sure. I have had great numbers that I could see feet off from cloud to cloud between scan stations. Blunders where a box truck moved up in spots to a light and 3d was like .003 m yet be a meter off visually. 

That's a classic LiDAR headache! Combining scans from different positions and ending up with gaps and misalignments is super frustrating. You're definitely not alone in this one.

It sounds like you've already tried the usual suspects (target-based and targetless registration), which is good. That shows you're tackling it systematically. Drift and misalignment are the bane of our existence sometimes!