Is it possible to define the image position in 6 degrees of freedom?
As far as I’m aware, Mapillary supports latitude, longitute, and heading. Does it support altitude, roll and pitch?
Roll would be especially useful - there are countless images that are not perfectly level, and all modern smartphones have an accelerometer that the Mapillary app could use to save the roll value and level the image without any cropping.
Roll and pitch data is not uploaded by mapillary_tools. The general rule is that the scene should be horizontal and not more than (say) 2m above ground. That statement is in the FAQ somewhere.
Altitude is uploaded last time I checked. I don’t know if it is used or even stored. Accuracy may be an issue though as a GPS device has roughly double the VDOP against HDOP. That’s just how it is and a result of orbital maths etc.
I think Mapillary would greatly benefit from including roll and pitch data. It would help with photos taken on bikes, as they often suffer from not being level.
Yes, this is something we’d like to do - there are also ways to adjust for level with computer vision without having the roll/pitch data (because not all cameras provide it) - in the longer term this is something we’d like to do, but don’t have on the immediate roadmap for this year.
Another solution:
DJI action cameras have a horizontal lock option as a mode of image stabilization. When enabled, this completely cancels out vibrations while walking and ensures that the image is horizontal. It would also be effective when mounted on a bicycle handlebar.
sample sequence
Japan Railway Numazu Station
I assume this works electronically, which would result in cropping the image. Something I considered was using Gyroflow to lock the horizon but having Mapillary handle it would allow me to upload the footage without reducing the field of view.
You’re right - using the image stabilization on a DJI action camera narrows the angle of view.
When I get a chance, I would like to provide a comparison with and without image stabilization.
I am not sure what you are specifically looking for but Mapillary currently does not do horizon leveling on the backend. However, Mapillary does compute horizon deviation and makes it available in the computed_rotation field of the Image Mapillary API endpoint.