#002- Is it possible to deliver a point cloud directly from the field?

Wouldn’t it be great to send your scanning data to your customer directly from the site?

Fot. National opera in Oslo. Point cloud scanned with Trimble X7

Let’s continue the Trimble X7 review. As you could read in the first part of the article I had a chance to test out TLS (terrestrial laser scanner) from Trimble. The data suppose to be ready straight from the fieldwork. The surveyor can upload the point cloud directly to some cloud like Google Drive, Dropbox, or any other, and send the link to the customer. In this part of the article, I will focus on the point cloud and picture quality and will check does the connections among scan stations created directly in the Trimble Perspective are good enough to be a final deliverance?

Trimble Perspective deliverance

As I mentioned in Part 1 of this article, Trimble Perspective is the software installed on a Windows tablet and it lets to control acquisition from the Trimble X7 scanner. Users can control the scanner, adjust the scanning parameters, constrain single scans, and export data ready for delivery.

It’s a complex and advanced field software that gives a lot of possibilities for the end-user. Constrained and refined point cloud together with the quality report can be exported to well-known file formats like e57, LAS, PTX, and RCP. Supported formats are also Trimble’s TDX and TZF or Pointools POD. Those formats give the user deliverance flexibility straight from the fieldwork.

Fot. Comparison of the point cloud in intensity colors and RGB colors.

The question that comes to mind, however, whether the quality of the data is sufficient to send them to the client? To investigate it I used two samples of data. Set number one consists of 19 scans collected with parameters: 4 minutes scan time, pictures, and auto white balance.

Fot. Range of a 4min scan — approx. 57.5m

Four scans were collected in the open area — trees, ground, stream, small walk-bridge, and facades of buildings in about 10m to 30m from the scanner positions. The rest of the scans were collected from nearby buildings which were simpler for the algorithm to combine the data together.

Fot. Short distance scans connection failure.
Fot. Registration Failure between scans 16 and 17, about 13m apart.
Fot. Registration Failure between scans on both banks of a river.

Open area scans haven’t been constrained in the field (Trimble Perspective) because of the constraining failures (even though 19% of overlap between scans) so after 10 minutes of trying, I just gave up and decided to complete data processing on a PC. The solution would be to have one more scan station in between but in this particular case, a scan would be done on the bridge which was not stable.

Fot. Comparison of the pictures. The left picture comes from Trimble RealWorks, the right picture comes from Trimble Perspective.

Set number two consists of 11 scan stations with the same parameters as the previous dataset. In this example, I had also trouble combining data. I occur that between scans on opposite sides of the two-line street where distance was approx. 22m, scans couldn’t be connected by the auto-alignment feature. I had to connect them manually but the result was not impressive. An extra scan would be a solution to fix it and give more overlap among scans but then I would have to do it in the middle of the cross-road. I believe the Trimble Perspective would perfectly work with the data come from a narrow area where are a lot of different surfaces, or indoor projects.

Data constrained in Leica Cyclone Core 2021

Both datasets were imported to Leica Cyclone Core software as e57 and PTX formats. I didn’t have any troubles regarding import. Data appear as regular raw colorized point clouds placed in correct spots with each other (constrained in Trimble Perspective). The problematic spot in set no.1 was easily fixed which shows that the overlap was good enough to constrain data. Data from dataset no.2 have been also fixed/constrained in Cyclone.

Fot. Misalignments of the data in Trimble Perspective. Scanned in an open area nearby the river.
  • The first constrain contained the just imported e57/PTX files. They suppose to be placed in the right places as good as Trimble Perspective lets to do that. It is always the reference data.
  • The second constrain contained reproduced single connections (cloud2cloud) as it was done in Trimble Perspective
  • The third one contained a dense network of connections among scan stations which suppose to be the proper final constrained data ready to deliver. Connections were created in Leica Cyclone Core with an auto-add cloud constrains feature which automatically finds overlaps among scans. The data was manually examined in cross-sections.

Comparison methodology

The first comparison was done between the first and second constrain. It let me know if and what is the difference between those constraints in software from the two competitors on the market. In theory, they should be similar or very close to similar.

The second comparison was done between the second and third constrain. This one indicated the quality of constrained, refined, and ready-for-delivery data consists of single connections between scan stations, and the dense network of connections among scans created in Cyclone Core.

Both comparisons were made in Leica Cyclone 3DR 2021 and visualized by colorized distance intervals.

Dataset no.1

Fot. The point cloud of the part of the scanning area.
Fot. The orthographic top view of the scanned area.

Dataset no.1 was collected in a city area. It begins in the open area where are trees and a stream and continued to the more narrow area between 1–2 storage buildings. Below you can see tables show the C2C (cloud to cloud) constrains quality.

  • Number of single connections — 18
  • Number of connections in dense constraining network — 59

The first comparison shows the small differences, especially in the height. The difference oscillates between 0cm and 3cm. The most significant difference is visible in one area where the cobblestone ground differs. The horizontal plane seems to be almost the same except a single spot with shiny, metal wall plates mounted on quite a big wall’s surface. This type of wall can be tricky for the cloud2cloud constraining type. Here we can assume that the Trimble Perspective did a pretty good job.

Fot. Leica Cyclone Core constraining. It is the recreation of the connections made in Trimble Perspective.
Fot. Comparison between the recreated single connections like was made in Trimble Perspective with scans positioned in Trimble Perspective during the fieldwork.
Fot. Position of the cross-section at MP80 of the long section. Both are marked by black thick lines.
Fot. The 10m cross-section — MP80. The difference between point clouds in the first comparison equals about 2cm.

The second comparison presents bigger differences. I again occur height fluctuations which give higher numbers and appear in the same area. The differences oscillate between 0cm and 4.5cm which is more than in the previous comparison and worry a bit. It was only a 19 scans dataset so I wonder what could happen in the case of 150 or more scans project. Do the user can still rely on the algorithm implemented in Trimble Perspective? Can the customer believe that the data is proper and ready to use or should ask for data processed in an office?

Fot. Leica Cyclone Core constraining. It is created by the auto-constrain feature which gave 4mm of accuracy.
Fot. Comparison between the recreated single connections like was made in Trimble Perspective with auto-constrained in Leica Cyclone Core.

Dataset no.2

Fot. Second dataset overview — perspective view. Notice distance approx. 22m between scans on both sides of the street.

Dataset no.2 was collected under the road bridge and continued to the top of the same bridge. It is looped data — the first and the last scan overlapped.

  • Number of single connections — 10
  • Number of connections in dense constraining network — 42
Fot. Trimble Perspective had a problem with registration scans about 12m apart. Overlap equals 44% but not many common features and surfaces which means difficult environment for the app.

Unfortunately, the first comparison wasn’t able to be done. The data from the field wasn’t properly combined because of constraining trouble in Trimble Perspective.

Luckily, the second comparison shows the difference between data based on single connections and dense network of connections among scan stations.

Fot. Long exposition time during taking pictures for panorama. The entire data acquisition took extra 7 minutes. Scan station was placed under the bridge during the sunny morning. Trimble X7 firmware didn’t have HDR option.

Summary

As you could read in this article, point clouds that come from Trimble are impressive but not perfect. Can the data collected with Trimble X7 be delivered to a customer directly from the field? The answer is not obvious and depends on several factors. These factors are:

  • Amount of scan stations in a project — more scans give lower quality of entire scans network
  • Overlap quality between scans — higher overlap percentage lower possibility of rotations between scans
  • Loop scanning method or linear scanning method — loop method improve the quality of final data
  • Required accuracy in a project

In my opinion, the quality of the data is very good and comparable to its competitors in the class like Leica RTC360 or Faro S series. It has some disadvantages like pictures white-balance in direct sunlight or improperly leveled single scan (check below) but I believe it can be fixed by the firmware update. There are no problems with data import failures regarding e57, LAS, and PTX formats.

I think every user of X7 should consider post-processing of the data before delivering it but there are for sure some scenarios when a point cloud could be delivered straight from the construction site.

Below some pictures which show the quality of the point clouds and photos. Read the captions :)

Fot. Point cloud presented in Leica Cyclone 3DR.
Fot. Shinny chimneys. Surprisingly good quality of the point cloud. Point cloud presented in Leica Cyclone 3DR.
Fot. Comaprison RGB and BW colors on shinny pipes.
Fot. The cross-section of the shiny pipes.
Fot. Comparison of the low-resolution scan (yellow) and high-resolution scan (red).
Fot. The panorama picture quality in ReCap. It comes from the E57 file exported directly from Trimble Perspective.
Fot. X7 doesn’t have problems with scanning shiny elements.
Fot. Colorized point cloud. The picture was taken in direct sunlight without HDR. HDR was unavailable at that time (April 2020).
Fot. The picture was taken with the built-in camera on the Trimble T10 tablet.
Church in Oslo. The point cloud

--

--

--

I am a geomatician and software developer with over decade experience in reality capturing and BIM. More info read on 3d-points.com

Love podcasts or audiobooks? Learn on the go with our new app.

Using TensorFlow Go to serve an object detection model with a web service

Data Science Quick Tip #003: Using Scikit-Learn Pipelines!

When Graphs Can Help, 8 Ways to do NLP Better, Instant Neural Graphics Primitives, and Jobs

A Complete Guide to Probability and Statistics for Data Science from zero to Hero[Part -1]

Efficient Pandas: Using Chunksize for Large Data Sets

The Journey of Coffee through Data

Common Time Series Data Analysis Methods and Forecasting Models in Python

How one piece of code empowered hundreds of thousands voters in the South African Local Government…

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Lukasz Wiszniewski

Lukasz Wiszniewski

I am a geomatician and software developer with over decade experience in reality capturing and BIM. More info read on 3d-points.com

More from Medium

Quadridox is working to develop and validate the use of physics-based synthetic data

Walk Softly and Carry an Appropriately Sized Stick

AI-ML : a Tool, not Technology

AI moving to the Edge of the Cloud