From 2D information to 3D representation

The Pointscene Diaries
The Pointscene Diaries
6 min readNov 25, 2017

Photogrammetry and laser scanning, although completely different approaches, both are able to provide a similar result - 3D representation of real world. In recent years we’ve been observing how both technologies improve and develop. We can’t say that one method will eventually replace the other, but we should rather consider which method will help us achieve the effect we want. Does the fact that laser scanning is so widely used in construction, engineering, and architecture to gather 3D data, means it should always be the first choice? Absolutely not! Let’s not forget the alternative - photogrammetry.

Photogrammetric point cloud obtained with drone. Scene uploaded to Pointscene by Groupe HBG ARPENTEURS-GÉOMÈTRES.

In many cases, photogrammetry is more affordable solution to generate a point cloud. It doesn’t require expensive equipment and highly specialised skills. A simple camera to capture images and special software to process them is enough. So before you spend money on your own scanner, check if, for your purposes, you could use photogrammetry instead. Very often it’s the quickest and easiest solution. Especially when it comes to aerial imagery from UAVs, no doubt it’s currently more cost-effective and accessible solution than laser scanning. You can cover large areas faster and it’s great for measurements that need to be repeated regularly. If you plan to use 3D models for visualisation purposes, photogrammetry also beats laser scanning. The textures generated from images are often better quality and they’ll appeal more to your clients. Another great thing about photogrammetry is that you can capture anything from small objects through statues and buildings to whole areas.

So whether you consider documenting projects in 3D or you’re just looking for some tips to improve your current workflow, this text is for you.

Equipment

The best thing about photogrammetry, it’s easy to start collecting data, all you need is a camera. Nowadays when technology is developing so quickly, you can use almost any camera. Although it’s recommended that it has at least 5Mpix resolution. Photos can be taken with a smartphone camera, unmanned aerial vehicle or metric cameras for photogrammetry. Remember that the general rule is: the better the camera and the more images captured, the more accurate and detailed your model will be. The amount of pixels is not everything. It can mean more detail, but the size and quality of the sensors (and shutter) is more crucial. The choice of the camera depends on how do you plan to use generated point cloud.

Archpriest Tabernacle- point cloud generated entirely from terrestrial photos. Scene created by agisense.

Software

There are multiple commercial photogrammetry and 3D modeling software available on the market. With so many different pricing and licensing models available, everyone should find something for himself. Some programs have trial versions and some are open source. Before you make up your mind on one, test them and choose the best for you. Below I listed the most popular programs for processing images and producing point clouds.

  • Recap Photo: It’s Autodesk solution for drones and UAVs based projects. Using Recap Photo you can create photo-based point clouds, textured meshes, and orthophotos.
  • 3DF Zephyr- The software allows for automatic reconstruction of 3D models from photos. There are separate versions for terrestrial and aerial drone-based images. The free version limits uploaded photos to 50.
  • Agisoft Photoscan- Photoscan has a wide range of uses, starting from 3D modeling of terrestrial objects to drone-based aerial surveying. There are two versions available, Standard and Pro. There is a monthly trial for the software available.
  • Regard3D- Free and open source photogrammetry software. It uses structure-from-motion algorithm to produce a 3D model of an object from series of photos.
  • RealityCapture- The software extracts 3D models from a set of images or laser scans. An interesting feature is that it can combine images and laser scans for 3D reconstruction.
  • Pix4D- It’s a drone-based photogrammetry software generating 3D models from images. To test it out you can start a free trial or buy a monthly license. For more information about Pix4D, check my review.
  • Dronemapper- It is a Windows software for photogrammetric imagery processing and simple GIS analysis. They also provide a free version, which is limited to a maximum input of 150 images, also more advanced paid packages available. Check the new text about Dronemapper here.
Bouquet of roses reconstructed from 257 photos taken with Samsung WB150 and processed in RealityCapture. Scene created by Pointscene user cga.

Workflow

Image capture

There are a few guidelines that are good to follow when shooting photos for 3D modeling purposes.

There are some limitations regarding object or scene to be captured. First of all, reflective, transparent or featureless objects should be avoided, because if the algorithm doesn’t find enough matching points between photos, the object might not be fully reconstructed. Shortly speaking, objects with a lot of visual details or textured are a better choice. Fortunately, you can always add small target-like features. Moving object or background causes noise and might also prevent successful reconstruction.

While capturing images, most important thing is, the images must be taken from different viewing positions. To get accurate results, remember to take into account high overlap between pictures. Make sure to capture the object on multiple photos. It is always better to take too many photos than too few. The more pictures the object has been captured on, the better algorithm will work. Don’t zoom in between taking pictures When shooting outside remember that direct sunlight casts shadows that will be visible on the model, so don’t forget to take into account shooting conditions and the environment.

Image processing

Although there are some software specific differences when it comes to settings, the workflow will be similar in each one of them. The first step is always loading the photos into the software. Typically, in the next step software aligns photos and generates a sparse point cloud. The last step is dense point cloud generation.

Point cloud of excavation- result of processing images in Agisoft Photoscan. Scene created by Landsker.

Export and sharing

After creating point cloud and exporting the results, you can upload your work to Pointscene platform and store it there. That way, you’ll be able to manage all your projects from one place and access them from any device. If you want, you might display your works in Pointscene public gallery. This a great way to present your skills and the quality of your projects to potential customers. Similarly, you could embed a scene on your own website. On top of that, Pointscene gives you the possibility to share your data with multiple users, these can be your co-workers, clients or friends. All you need to do is to type their email addresses. So no more tedious data compression, transfer, and download. In Pointscene you can access, view and handle point clouds with ease.

Point cloud of Laukaa center created using nadir RGB images captured from a drone. Data produced by Ympäristonsuunnittelu Oy and published in Pointscene by Olli Nevalainen.

What to read next

  • For more information about photogrammetric software, check the texts about Pix4D’s machine learning algorithm- here and about Dronemapper- here
  • To find out how to use Pointscene, start with data upload instructions: Part1 and Part2

Test Pointscene today

Want to know how to include point clouds in your projects? Visit www.pointscene.com to explore many examples in gallery or start free trial and upload your own data within minutes.

--

--