We have recently started supporting 3D reconstruction of 360 panoramas in our processing pipeline and along with that we have updated the navigation in the viewer.

Navigation

The reconstruction of 360 panoramas gives us a lot of new opportunities on the viewer side. Whereas we previously used instantaneous transitions when navigating between panorama images we can now do smooth transitions in the same way as for regular images.

You will notice that you are at a panorama when you see the new navigation arrows at the bottom of the viewer. Once at a panorama, here is how it works:

• Drag with the mouse to look around
• Zoom with the -/+ buttons
• Navigate to surrounding images by clicking the new navigation arrows

Check the video below to get a feel for it.

The looking direction is kept the same when navigating from one panorama to another. When looking around by dragging the navigation arrows are rotated to always point in the correct moving direction.

Maybe you noticed that there were arrows with two different colors in the video above. The colors indicate different image types. A white arrow indicate that you are going to move to another panorama when clicking, while a yellow one will send you to regular, flat image.

Take a look at the video below for a deep dive into the different arrow types.

Notice that the yellow arrows for navigation to regular images are only enabled at certain rotations. This is because you can not look around in a regular image like in a panorama. If you see a disabled arrow and want to move in that direction, all you have to do is look around until it lights up. There can be multiple arrows to regular images in the same direction. They will then be enabled for different rotations as in the video. When looking at a regular image a panorama arrow will appear if there is a panorama nearby.

You can play with panorama navigation in Helsingborg, where WSP Sweden has uploaded fantastic spherical images taken from a van, and also in Paris and Marseille where our power user zimmy has been experimenting with the Ricoh Theta.

Reconstruction

On the server side, we have updated the 3D reconstruction pipeline to connect spherical images to each other and to standard images. Some big changes on OpenSfM have been required since most computer vision libraries assume perspective camera models. Fortunately, we have been able to base our code on the fantastic OpenGV library that supports any kind of camera no matter how odd they are.

How to take pictures

Here are some simple tips for taking pictures with spherical cameras:

• Put the camera on a stick up above your head
• Keep the camera straight up
• Take one picture every few meters. If the street is narrow, take more pictures
• No more worries about which direction to cover ;-)

So that is it. Now go find a spherical camera and give it a try!

As always, you can find us on Twitter. If you have any feedback, let us know through GitHub issues or the Forum!

/Pau, Oscar and The Mapillary Team