The Kolor team has been experimenting and developing a new solution to estimate accurately lengths and surfaces in a room from two spherical panoramas. Its code name is StereoPano. It is still in a R&D stage, but we are glad to share some first results.
Basic idea is to shoot two 360×180Â° panoramas of the same room from two points of view. The second point of view is shot above the first one, so the shooting setup is very easy using the vertical column of a tripod. The only thing to do is to measure the distance from first point of view to second point of view.Â From these two panoramas, we use the concept of stereoscopy to compute a 3D estimate of the room walls and to extract useful information such as walls edges lengths.Â Let us explain in further detail the whole process of StereoPano.
First you have to shoot two 360×180Â° panoramas from two points of view with a vertical shift. Stitch them into your favorite stitching software (aka Autopano). Here is the top panorama:
And here is the bottom panorama:
Next step is to load the two panoramas in StereoPano and choose â€śCompute disparityâ€ť. This will lead to compute a depth map by comparing locally the two panoramas. The depth map stores in each pixel the depth computed by triangulation between the two points of view and the pairs of matched points. The deeper is the surface, the darker is the depth map, and vice-versa.
Now, one can select surfaces by clicking in the main window (like the Polygonal tool from Photoshop). This will help StereoPano to identify the main flat surfaces. From now, by choosing â€śCompute planesâ€ť, StereoPano will estimate the exact positions of the planes and display a RMS error for each plane. This RMS states for the RMS distance between 3D points and the plane.
StereoPano then intersects planes and displays edges (in red) and their lengths (in white) with their measure accuracy (under brackets). We have checked that lengthsâ€™ accuracy are good.
You can now export some visual results such as floor plan, room wireframe, or 3D point cloud.
Through this one year study, we found out that measurements were quite good if you take precaution during shooting. We are able to measure wall lengths with an accuracy of 2 centimeter at 5 meters, which is an average error really below 1%. We’ll continue to study what is possible to do with panorama to take measurement and perhaps even see if we can create 3D models out of panoramas.Â If you are interested into learning more about this prototype, feel free to contact us.
Morning Alessio, thanks for your message. We know this application is quite interesting, but for the moment, it is just a R&D project. It needs to be fine tuned, some more options to be added, before we can think of putting it in beta.
Rgards. Philippe – Kolor team