When using 5.13 Android version, I noticed that the view direction is completely off. See for example this photo: Mapillary. The view direction shows about North-East (dark blue area), while the correct direction is North-West (yellow area). How can I solve this? I thought that the angle was calculated by looking at the direction of the next photo compared to current photo.
At the moment we use the raw angle provided by the compass (before it was direction of movement, right). It might be very off, especially while driving. But on the server, after uploading, we will improve the angle using multiple techniques based on a given sequence.
BR, Yaro
What if smartphone don’t have compass?
Then it will be zero. The server will still perform the same number of operations to place the view angle in the correct position.
BR, Yaro
Thanks for the reply Yaro. You mention that on the server, after uploading, you improve the angle using multiple techniques. But apparently that doesn’t work correctly at all, as the dark blue direction still is about 90 degrees off, as can be seen in the screenshot. Is there still a bug in your processing then? Could this be fixed?
@nikola do you remember which angle is which? Is blue calculated angle, or raw one?
BR, Yaro
The blue map markers that appear from hovering on dots are from our vector tiles (API Documentation) that have the original compass angle for now. The yellow marker is showing the corrected compass angle (“computed_compass_angle” from API Documentation).
@Loek, on the attached sequence the default (yellow) compass direction looks correct, and this is a computed on (on the server). Sometimes, if computation fails, it might be wrong, but at least not here. So, I guess there is no issue in processing.
The blue one is the angle from the device compass. The value is wrong, and there are many reasons of why that can happen (mostly because of the metal cage).
@nikola thank you for the clarification.
BR, Yaro
Thanks @Yaro. Clear now. I was incorrectly assuming that the blue one was already computed, and the yellow one was just simply using the direction of the next point to determine orientation. Yellow indeed looks good, so agree that processing seems to work fine in this case. Thanks again!
@Yaro It seems it’s not fully working correctly now. See this sequence for instance: Mapillary. There, the yellow marker is pointing in completely the wrong direction. Is this a known (new) bug? And would it be possible to fix the direction still of those processed sequences?
We have two separate pointers on the map: one calculated on the server based on moving position and terrain features, and another derived from device sensor data. However, during driving, it’s not uncommon for sensors to become unreliable and provide seemingly random readings. Unfortunately, as these libraries are provided by Google, there is limited scope for us to address this issue directly.
@nikola could you please remind which point (blue or yellow) is calculated position and calculated angle?
BR, Yaro
The blue one is the original position and angle and the orange one is the calculated/corrected position and angle.
Thanks Yaro and Nikola. So in this case the processing does not seem to work correctly, as the yellow/orange one is not pointing in the right direction. For instance here: https://www.mapillary.com/app/?lat=51.41805459999998&lng=5.426970200000028&z=19.102212437106672&dateFrom=2025-01-19&pKey=647566214273820&focus=photo. Hopefully the processing algorithms can be improved to fix this in the future.