( read )

Is targetless registration for point cloud processing efficient?

By Charles Thomson
November 8, 2018
Advanced 3D surveys rely on 'point cloud processing'. This article will explain the problems targetless point cloud registration has historically faced and the revolutionary software advances that are changing industry best practices.

Advanced 3D surveys rely on 'point cloud processing'. Although this system has offered an unprecedented level of precision and thoroughness, there have always been lingering manual processes that slow down surveyors. This, however, is changing.

Efficient targetless point cloud processing is finally here, removing a long-standing stumbling block to improving surveyors’ workflows. This article will explain the problems targetless point cloud registration has historically faced and the revolutionary software advances that are changing industry best practices.

The reason surveyors use artificial targets to register point clouds

Laser scanners make line-of-sight measurements. Generating the coverage needed to make a detailed 3D map typically requires scanning from multiple angles. The different scans are joined together using scan overlaps to create a complete model, a process called 'registration'.

Satellite navigation technology such as GPS is used to align datasets of large outdoor areas by referencing the positional data of each scanner. However, making GPS measurements to the needed level of precision requires a clear line-of-sight to satellites. This is difficult or impossible to achieve when surveying urban environments, indoors, or even under tree cover.

Without GPS, a localised method of ‘fixing’ the scans is needed. This is where targets come in. These targets are usually free-standing spheres or chequerboard surfaces, placed within areas of scan overlap. This creates clear points by which to align the scans.  

While reliable, targeted registration is time-consuming. The targets must be placed manually and with care. At least three targets are normally required to align a single pair of scans. The position of each target must be carefully thought out to ensure appropriate overlap. Complex environments create additional challenges to the placing of targets, and any movement of the targets will result in serious alignment issues.

Why targetless registration has historically let surveyors down

The long sought-after solution to the time expenditure of target placement is ‘targetless registration’. Rather than having to place artificial targets, features already present in the scanned environment can be used to align adjacent scans. It's only a matter of finding those common features in adjacent point clouds.

Unfortunately, this is easier said than done. Computer analysis of point cloud scans for cloud-to-cloud registration is slower than identifying and aligning scans based on artificial targets. Traditional processing software also requires human operators to manually input and cross-check each alignment. A poor quality scan can impact the accuracy of the entire model, and any distortions or 'noise' in a scan must be identified and removed. The time saved by avoiding target placement is often completely negated by extra time spent processing the data in the office.

For example, a 130 scan dataset of a building might be completed in a single day, but will require over 28 hours to process using conventional alignment techniques. If manual oversight is required throughout the whole process, that comes out to over three days of work.

The software advance that has made targetless registration the new best practice for point cloud processing

Things have changed, recent breakthroughs in software have massively reduced processing time, enabling rapid, automatic comparisons of datasets without the use of artificial targets. This is achieved through multi-stage vector analysis.

Modern software uses the positional data from the scanner to extrapolate each point within a point cloud into a directional vector. This allows each scan to be condensed into a 'vector sphere' emanating from a single point. The unique identity of each scan is retained by the density and directional characteristics of the vectors. As vector spheres, scans can now be easily overlayed for rotational analysis. This allows for rapid 2D point density alignments to be made on the horizontal and vertical axes in separate and automatic processes.

Using vectors to cut processing into three stages allows for more robust yet faster alignments — cutting processing time by 40%-80% depending on the size of the project. Required scan overlap is also reduced to as little as 30% under ideal circumstances. Most importantly, a more automated process (and the frontloading of the remaining manual inputs) allows scans to be queued for hand-off processing. This enables the processing of large data sets without significant time costs to the surveyors themselves.

Although movement can still compromise scans, modern processing software is capable of using stationary objects to register scans, preventing distortions from corrupting the alignment. Lacking any such features, free-standing targets may still be required. But, under most circumstances, surveyors using the latest point cloud processing technology can gain significant efficiency improvements without compromising accuracy by going fully targetless.

Summary: Multi-stage, vector analysis delivers the future of point cloud processing

Despite technological advancements in many fields, the time-consuming placement of registration targets has remained a staple of surveying. Until now, targetless point cloud processing required too much hands-on activity to be worth it. Time saved in the field was paid back in excess processing the data in the office.

A dramatically faster targetless registration process is now available that requires reduced oversight and more automation. Vector analysis in combination with new algorithms have provided the missing piece needed to move beyond the roadblock of inefficient point cloud processing. An automated procedure for processing targetless point cloud data exists at last.

These improvements are expanding the possibilities for point cloud use. Registration has always been the most time-consuming aspect of point cloud creation. The increases in efficiency created by automated, multi-stage processing are allowing new businesses and industries to access point clouds. Surveyors who take advantage of this developing technology will create a competitive advantage and grow their possible client base. It is an exciting time for the industry.    


Correvate CTA (1)

Tags: point clouds