Mapillary with Olympus TG-Tracker in 2020

Saturday 25.4.2020 13:46:37, Jakuje

Kategorie: Ostatní

Mapillary is world-wide database of street-level imagery for free use. It is great help for mapping in OpenStreetMap as other street-level imagery might not be available or might not have compatible license. Everyone can submit photos directly from mobile phones mounted on the windshield of a car or from any other camera after some postprocessing. There are bunch of "supported" cameras suitable for usage in car and outdoor, but they are very expensive so I wanted to try to go with budget option.

This was the reason I bought (after years of considerations) Olympus TG-Tracker for around 140 USD/3600 CZK. It has advantage that it has integrated GPS sensor so I do not have to bother with recording GPS location using different device and making sure time is properly synchronized. It has fine sensor resolution (8MP) and simple settings for continuous shooting (I use 1s interval). I was already warned about using 0.5 s interval as it can record only one GPS location per second for whatever reason. It has also advantage that it works fine with gopro mounts so I bought few for my car (inside and outside) and for my bike and went out testing.

The configuration was simple. In menu, I set the 1s continuous shooting. The camera has also wide angle of sight but it can be changed to narrower (which is used for underwater shooting). I use the narrower (14mm equvalent) as it is more than enough for normal use). Recording is started and stopped by pushing the largest button. Simple to start or stop when I get to interesting location. The on-off button has also third position "log", which (I think) keeps GPS log, but saves battery from the live-view.

After collecting some images, the tough part came. Mapillary provides set of python scripts mapillary_tools, which does not work with Python 3. And actually does not work with any updated python libraries. It requires some ancient ones and actually even one that needs to be patched (or installed from different source than PyPi). Trying to hack something in there leads into errors of missing testing library, which is obviusly not open-sourced. Sigh. So lets run all that cruft in something that stopped being supported last year.

Getting the hang of the tools also take some time but during my tests, I settled with the following workflow:

  • Take the card from camera to the computer
  • Navigate to the directory with the images (for example /media/jakuje/9016-4EF8/DCIM/100OLYMP)

  • Run the process: mapillary_tools process --advanced --interpolate_directions --import_path "$PWD" --user_name "jakuje" (I am using the --interpolate-directions because the directions from camera are not right)
  • After the processing is done, I run the upload part: mapillary_tools upload --import_path "$PWD"

Depending on the amount of images and speed of your computer (and card reader), both of the steps will be completed and images will show in mapillary website (and in OpenStreetMap editors) and you can jump to improve the map!

I did not test the outside camera mount for a car, but the inside one is placing the camera up side down. But the camera is clever enough to rotate the images without any intervention.




Podbarvená pole označená hvězdičkou jsou povinná!