IDEA: create a geo-tag assistant!

The weather is getting better and I did another attempt working with my 360 camera. To my regret it failed to link the geo-tags yet again… To manually tag and orient 312 images is quite a painstaking piece of work…

Instead of throwing the images away yet again and try to figure out again on how to get this to work… The web-uploader could be fitted with a geo-tag assistant!

As a test I uploaded 28 of the 312 images. I Get the message:

We found 28 images
28 images are missing EXIF data and need to be placed on the map.
When reviewing, you can drag the markers to their appropriate places.

Then I can manually position each image one by one on the map… what if you guys& girls created the awesome “geo-tag assistent” ™ ! :muscle:

geo-tag assistent:
Based upon the image names and image dates you can easily determine the first and last image

  1. In the ‘review’ screen you add button with: "use geo tag assistant" (hence: GTA)
  2. The GTA askes you to place the first image on the map (hence: position 1)
  3. The GTA askes you to place the last image on the map (hence: position 28)
  4. => AUTOMATION! => GTA distributes the other 26 images evenly between the two points!

Then the fun begins:

To make things more understandable (I hope!) I created an image to show my intention graphically:

Position 1 and 28 get fixed (color RED) for their position, the other 26 have status “flex”. When you click the little round marker (with the orientation indicator) you will see the information beloning to that image. One can choose to (un)fix the position and/or orientation and alter the orientation (note, the first and last position must remain fixed! One can off-course always drag any position to an other spot) When the orientation is manually altered, then the orientation will become fixed also. Off-course one can manually alter the “fix/flex” setting.

And then the real fun begins:
When a selected position (15) is dragged to an other spot (fat gray arrow), that position becomes fixed also. And all the “flex positions” move to new positions along the lines between the fixed positions!

Also, Ussualy images get taken in the direction one is moving, this is also true for 360 images, so:
The last location (28) does not have a fixed orientation, so that orientation follows the line between fixed position 28 and fixed position 15 (15-28). Also position 16 through 27 will have that same orientation.
Position 15 also has “flex orientation”, that one will have the average orientation of the two lines 1-15 and 15-28.
Interestingly the first location (1) has a fixed orientation in this example (the red color for the orientation indicator). So pos 14 will have the orientation of the line 1-15 and the orientation of the positions 2 through 13 will evenly change from orientation 1 to orientation 14.

In the example shown it will still be necessary to re-position multiple points. But it will go much, much faster. And when one has a set of images in a straight line (of a road…) then one can work really fast to very nicely geo-tag the images!

I hope you understand my idea, if you have questions, please ask them :slight_smile:

PS: “geo-tag assistent” ™ No trademark, just to emphasize that I think this is a great idea :wink: (I do expect that the resulting image, tools etc, will (remain) be used and available for the whole internet community)


easiest way to fix this is for you to record a gpx file with your phone (OSMtracker app) and then use Geosetter to inject the coordinates into the jpeg. make sure to sync the clock of the camera to your phone clock. if you need help with that let me know.

1 Like

I have once installed the GPX Tracker app (iPhone icon: yellow square with black arrow to the top right).

But I have never used it… if you could tell me how to use it @nunocaldeira … (do I need to install to combine date?)

Still think my idea is a great way to edit image sets though :wink:

What’s so frustrating is that I see the mapillary app show all the right dot’s on the map when I take the images… (because the terrain was a bit rough I took every image manually) and when I exit the image making it asks if I want to append the EXIF data… I click 'yes"… but it does “no” :cry: If the app could be updated somehow to get this to work, it would make my life so much easier… or if one could tell me hoe to correctly attach the EXIF data… that would help massively also… (I have an iPhone 6 with LG 360 camera)

considering how busy the dev team is already, I wouldn’t expect a full blown geo tag workflow to come out of mapillary any time soon. You might want to search around for a piece of software that might be able to do it (I have a feeling there might be something similar to what you are suggesting).
Also, if you’ve got LG 360 and an iPhone, why not use the mapillary app integration which geotags images based on the phone’s gps?

Thanks @4004 for replying… As I said I have been trying all sorts of things to get the geo-tagging to work… but to my dismay I can’t get it to work :sob:

I have been searching for something to get it to work… but alas… I couldn’t find anything…

I’m afraid I need to start fiddling around with an external app to reccord the geo-tag… but geez… it’s so intensive to figure this out… Can’t seem to find a good “howto”

there is the wiki on geotagging here, and I think there was a tutorial on merging gpx tracks with images based on timestamps in the forums - if not just post a thread asking for help.

When you say you can’t get geotagging to work, do you mean the link between LG and the app? Are you using the latest LG firmware? I didn’t have issues with mine

Hello i created this Geotag imagery on action cams or 360 cameras that don't have GPS
Hope it helps

1 Like

The GeoSetter solution is great, I am using it a lot also. (I just added a new section to the instructions @nunocaldeira wrote.)

However, there really needs to come a solution also for photos that have already been uploaded to Mapillary. Take for example this hideous sequence that I took recently:

Currently this sequence doesn’t look as bad anymore as it did after uploading, because Mapillary has broken it up in maybe 1000 little sequences, but the original sequence forms a 5-pointed star zapping back and forth over a 20km area around the airport. (Zoom in to the map to see all the spread-out and cut-off photos.) There is no way I am going to be able to correct these images with the current Mapillary editor.

Editing sequence after uploading is the worst workflow, if we manage to do it before uploading is the best option. we all have incorrect sequences when we started out using Mapillary. from my side Chrome is the best browser to edit a sequence to fix those sequences already uploaded.

Have you ever tried evaluating a sequence in the Mapillary app before uploading it? It’s impossible. You just have to upload it, and hope for the best.

I think Chrome is the only browser that actually works with the editor. But it’s still impossible with sequences like the one I linked to, since Mapillary afterwards broke that up in about 1000 single-image sequences, and the Mapillary editor only lets you edit one sequence at a time.

No, we will need a proper editor for editing uploaded sequences. I just hope the API allows proper geodata editing, so that it’s even possible to build something.

didn’t meant on the app. other workflow, if you notice the gps is way off like in the example you shown, is to send the photos to yoour computer and fix on GeoSetter. Yeh i know its painstaking, but its the only way around i know :grinning:

It doesn’t matter what workflow you use.

If there exist situations where many photos with bad GPS information can get uploaded to Mapillary (which is clearly the case as you can see), then there is a need for a good editor of uploaded images.

I would love to hear it if you have a good method of checking the GPS quality of photos taken with the Mapillary app before uploading them.

I red about Geosetter, also about the instabillity…

I have done several walks with my dog in my village, after an (half) hours walk I need about two hours of work to do a decent upload to mapillary…

I just started with my village project (and I am still hoping to find an affordable sollution to get rid of the massive stitching problems with my LG360 cam). And I don’t like the workflow, one hour of image taking and two hours of post processing…

I red up on the possibility to create my idea. Seriously impressed with what Mapillary created with the webuploader (especially the reading image data before uploading part, nice!) but I wouldn’t be me if I didn’t think it could be improved upon :wink:
A pitty they don’t have the time to work on extending this already verry nice piece of software. A pitty that the software isn’t opensource so I can put in the time to extend it. I like leaflet (osm library) and I can manage Javascript… so… this will take some time, but I’ll try to find the time to build something that’ll beat geosetter in usability and help all (not just me) in post processing

1 Like

First step, I saw a great post with a link to “brouter”.

That is a nice first step for my “geo tag assistant”. Check it out >>

This “normalizes” your GPX track and matches it to the very nice route planner of ‘brouter’ !

And all this is done client side, nothing gets uploaded to my server…

Added a OSM map showing the filtered GPS locations and did the first part reading and checking the upload of the geojson

Next: match the geojson with the original GPX file and re-match the date/times and add microseconds…

PS: I have tried this with the few GPX files I have created, I am curious how it’ll match your (larger) GPX files!

PS2: the route must be along known streets, walking through a bush and such will get brouter lost…

is this essentially “snap to road”? i recon it might have the same pitfalls as OSC and similar systems using this - snapping to wrong road, and reliance on good osm road data in the first place.

Yes, but you can edit the path in brouter… so if it goes wrong somewhere, you can correct it.

I’m also planing to add an image uploader so you can match images to the corners/turns. Then it will generate and GPX track matching your images dates so you can match it with an EXIF tool…

I plan to use this with my new (yet to arrive) 360 camera, to match videos (extracted with ffmpeg to images)… I publish this, because it might be usefull for others also…

Ok, did an other update:

  • added processing the brouter-geojson
  • added option to alter the geojson-track

Check it out >>

next up: matching the time… (now the harder part really begins…)

I’ve been thinking, maybe the simpler approach will work best? Both of the track distances should match quite well. I could match the time based upon distance…

Added a dot on the map on mouseover when you go over the GPX positions (helps me find the start of my movie sequence)