Collecting street level imagery in less than ideal conditions


As the days get shorter the conditions to collect street level data deteriorate. There was mension last week in the Forum of glare caused when the sun is low.
I know I do more driving in sub ideal conditions during these short days and long nights.
I keep the cameras rolling. On the darkest night the center line of the road can be picked out. If I’m driving into the sun a slight turn of the steering wheel often clears up the pictures I’m collecting.
This week I will be driving over to Harrisburg Pennsylvania. I hope to take route 30 all the way. I’ll leave before daybreak and hope to get over there before dark. In Pennsylvania no one has recorded route 30. Fog is th he worst. Pennsylvania does fog in spades. No matter. The cameras roll.


Rememeber that a good reputation is easier lost than acquired.

Mapillary Camera Grant Program has launched

You think I should turn off the camers when it gets dark?
I’m asking.


Think about the eyes of the mappers.
Try it for yourself.
What good is making pictures when you cannot read the milestones ?


I think if there is no data already captured anything is better then nothing, and can always be replaced at another time.


Nowadays roads for cars are on the map. More difficult things to read are now needed to refine. Max speeds, direction, access restrictions.
When I map some highways, I tend to filter out my pictures in order not to be bothered by junk.
I am even happy when most new contributors stop contributing because people pursuing quality are scarce.


While I agree having something is better than nothing, there has to be a limit to how far you let the quality slip. Of course, where the threshold lies is pretty subjective. But for instance I’d argue tracks like this one should never have been uploaded in the first place (or should have had 90% of the pictures our filtered out first):


I cant get that link to work.
What was it that was so bad?
I just looked at some tracks I did by headlight light only.
The center of the road was clear in all and some signs were readable.


Fog seems to be the worst, and fog at night is very bad.
Downpours where you can’t see show more than the naked eye picks up.


That track was in a park and half the photos only show the tree tops (the ground is not visible), others are half filled by the back of the person walking in front, all are tilted in one direction or another, a lot are fuzzy due to motion and the others due to bad focus. With such a combination they are unusable to contribute to OpenStreetMap (points of interest are not visible and there would be no landmark to precisely position them anyway).


The question is quality for whom. I don’t disagree that there are images that aren’t pretty to the human eye. But what Mapillary is doing is not a photography project; it’s about AI gathering data, correct?


I wouldn’t be surprised if in a couple of years they have mapillary filtering images based on quality.


Something has to happen. My good pictures are overwhelmed by other people’s rubbish.
Maybe propose them for deletion to the photographer.
A craftsman does not mind a competing craftsman as long as he does not dishonor the craft.


That image has no usable data.
Therefore it needs to be deleted.
I’m not sure how that happened
Maybe my camera fell from the mount.