Can surveyors trust targetless registration for point cloud creation?

A 4-point checklist to making sure that your software will deliver when it comes to the targetless registration of point clouds

The simple answer is ‘yes’. The complex answer is ‘only if using the right software and processes’. However, in all cases, the current state of the market is marred by past failures. Early attempts to develop target registration processing created a reputation for wobbly, slow and inaccurate results. A lack of robust verification procedures made it difficult to guarantee quality outcomes even if they were achieved. Even more rare, finding all of these features in one place has been a headache for surveyors for years.

Surveyors should take a fresh look at the technology available to them. It has never been easier to find quality point cloud software that, when paired with robust field techniques, will produce cloud-to-cloud registered surveys that can be trusted to deliver results that rival anything produced using artificial targets. This article will explain the developments that have made this a reality and the features to look out for when processing point cloud datasets.     

1. Point cloud software with multi-faceted verification features

In order to be able to trust your data, verification procedures are a critical asset. Ideally, you want access to both visual and statistical analysis to make sure your data aligns with job specifications. Visualisations (such as the popular ‘bubble view’) give you confidence in the results being correct while statistics tell you ‘how’ correct that data is — you need both. You also need the ability to correct data after coarse registration and review the data again following fine registration.

If processing software has all of these features, you have very little to be practically concerned about. You can check coarse and fine registration and then manually rectify any issues. Verification is also a critical step in processing data registered with artificial targets. In many ways, software advances have levelled the playing field to such a degree that verification is not any more of a concern for targetless registration than when using targets. But, when undertaking a project with exacting specifications, it is a critical stage in the processing journey.   

2. Point cloud software that delivers pre-processing options

Point clouds can be large, they can also be distorted by movement. You need software that offers you clean up features and the ability to thin data so that it can be more quickly and reliably processed. The ability to set a ‘maximum distance’ metric is most important to robust outcomes. Laser scanners have an enormous range. If scanning outdoors, or even in a large indoor space, scans can collect point data that is very far away from the scanner. This often creates unnecessary duplicate data that is not useful for alignment, and can even hinder it. The ability to remove that data by restricting the scope of processing to points within a particular distance from the scanner will help create a clear end product.

The ability to cull excess data around a scanner by setting a ‘thinning’ metric and decimation features that allow you to thin point data throughout scan fields regardless of its proximity to a scanner are also useful.

However, they are more important to handling large data sets than to robust outcomes with targetless data.

3. Point cloud software that processes quickly and frontloads manual tasks

Targetless point cloud processing can take time. This is particularly true when it comes to cloud-to-cloud registrations. It will take any program longer to align two point clouds based on natural features than to overlap three to four artificial targets. However, some programs are better at this than others. A lot of the software on the market is slow to change, focused on delivering end-to-end experiences — processing through to final 3D models. However, startups that deliver just one area of this journey sometimes come out with innovative solutions that can make life much easier.

For example, there have been advances in vector based, multi-stage processing that delivers faster, yet more robust registration with less manual intervention by splitting the procedures into three stages — rotational, vertical and horizontal alignments. For targetless registration, this has produced outcomes that are 40%-80% faster, depending on the size of the project.      

This improvement has been achieved by first subjecting point clouds to rotational alignment based on a ‘vector sphere’ analysis. This spherical analysis then allows for rapid 2D point density alignments on the horizontal and vertical axes. Through looking at scans in this way, there is reduced need to set parameters for scan comparison, decreasing manual involvement. Not only can scans be processed faster, they can be queued up for hands-off processing — greatly improving the efficiency of generating point clouds using targetless registration.

In some software programs, all of the manual involvement is frontloaded, basically limited to the processes of setting scan pairs. That only leaves manual verification. Combined, this faster processing procedure allows operators to focus all of their time on verification, increasing the quality of the outcome while still generating efficiencies throughout the whole procedure.

4. Software and field procedures that can account for propagation of error

The last important aspect of delivering robust targetless alignments is accounting for propagation of error. This is an issue for targeted processing as well. But, it is so critical to creating precise composite point clouds that it is worth mentioning. When you combine point clouds, the inherent errors created by the nature of the scanner will build on top of each other. Each pair of scans is effectively built on the next one, all rooting back to a single ‘home’ scan that ‘fixes’ the entire composite point cloud. The more links between any given scan and its ‘home’ scan, the greater propagation of error it will suffer.

To fix this, you need software that can take two or more composite point clouds already aligned through a coarse registration and combine them prior to a ‘global’ fine registration. However, doing this requires targets. But, these are not the same targets that would be used for ‘targeted registration’. These are targets that are placed and located within a larger site grid using a total station. Total stations are capable of making measurements orders of magnitude more precise than point cloud laser scanners, allowing for the creation of a rigorous control frame for the global placement of scans.

To do this correctly, you need to think about the end requirements of the project while scanning in the field and build a site grid proportional to the propagation of error that you can tolerate. You then need software that can handle this information. This is critical to precisely registering any large dataset.

Summary: The right software can deliver targetless point cloud data that you can trust

Many of the features that make a program capable of producing robust composite point clouds using targetless registration are critical to aligning scans based on artificial targets as well. Under all circumstances, you need software and procedures that can account for the propagation of error and verification features that will guarantee your adherence to job specifications.  

The biggest difference between processing targeted and targetless scans is the requirement for processing software that is fast and automates more of the journey. The extra time it takes to undertake cloud-to-cloud alignments makes this a point of practicality. Realistically, the inability to do this historically is the biggest reason for the lack of targetless registration adoption across the industry.

Automated and faster processing has delivered the biggest impact to hit the industry in recent years. This is freeing up surveyors, giving them the time to go back over the verification process in detail, verification processes and prove to themselves that targetless registration can produce robust outcomes. It is critical to look for these features in processing software in order to make targetless registration practical. All of the other features (verification, site grids, pre-processing features) are needed to create quality point clouds using targets as well.