Video uploads containing inaccurate trajectories such as tunnels

—Original Japanese text at the bottom—

I did my first street photography yesterday.
However, after uploading, I noticed that the trajectory where I passed through the tunnel was significantly different from its original position. I predict that this is because GPS does not work, so it automatically completes it.
Since such trajectories are useless, I thought they should be deleted and requested the deletion of several sequences that were significantly off.
The photo was taken in the following environment.

Camera: GoPro Max HERO mode (360 degree mode not yet available due to lack of proper mount)
Vehicle: Motorcycle
Shooting: 24fps video
Upload: Desktop uploader

I would like to know if it is possible to do something like the following.

  1. Select and trim the inaccurate section
  2. Prevent GPS from uploading inaccurate segments.
  3. Avoid shooting if GPS is inaccurate

Some Mapillary users seem to be able to accurately track GPS data even in tunnels, but how do they record this? ?
Please let me know if you have any information.

(The red line is the actual location.)
スクリーンショット 2024-01-29 090734

—Original Japanese text below—

カメラ:GoPro Max HEROモード(適切なマウントが無いため360度モードはまだ使用できていません)
撮影:24fps ビデオ


  1. 不正確な区間を選択して、トリミングする
  2. GPSが不正確な区間をアップロードしないようにする。
  3. GPSが不正確な場合、撮影しないようにする



At some point, in tunnel (and urban canyon), you will need to combine GPS/GNSS with IMU (inertial measurement unit) (and ideally wheel tick sensor). I’m using an U-Block F9R sensor.

1 Like

This is my sequence. The image in the tunnel is also in the correct position.

I will explain the process of recording while I am driving my automobile.

1: Shoot a video in mp4 format using the “Camera” app on my Android smartphone.
2: I use the “Tasker” app to periodically tap the “Camera” app’s record button.
3: Use the Android “GPS Logger” app to record the trajectory in gpx format.
4: The “RouteConverter” app is used to correct the trajectory. This app is written in Java and runs on Windows.
5: Use “mapillary_tools” to associate images with locations and upload them. This set of scripts is written in Python and runs on Windows.

GoPro MAX has a GPS receiver. The trajectory is embedded in the video file. Uploading it with the “Mapillary Desktop Uploader” will require little effort. On the other hand, it is difficult to correct inaccurate data.

I recommend you to use the Windows application “LosslessCut” to split the video file recorded by Gopro into tunnel and other parts. Upload only the non-tunnel videos using the “Mapillary Desktop Uploader”.

If you wish to discuss this in Japanese, I welcome it.


Dear @drivephotograph ,

Thank you for drawing my attention to the LosslessCut application.

EDIT : One apparent drawback can only find Linux and mac versions on Releases · mifi/lossless-cut · GitHub : doesn’t seem useful for a Windows user, then?

Second EDIT : need to expand the dozen listed download links to see the windows version :bowing_man:.

Met vriendelijke groet,

How Waze Beacons work - Waze Partners Help.

the Craeybeckxtunnel has it

Although it is attractive to be able to record position even in environments without GNSS using an external IMU, the cost of the module is high and knowledge to use the module seems to be required. I wish it could take advantage of my smartphone’s accelerometer or integrate more easily with my GoPro.

2024年1月29日(月) 17:37 Eric_S via Mapillary Community Forum <>:

I think it’s a good idea to use your smartphone as a GPS logger. I usually use a new iPhone, but fortunately I can still use my old Android device. Can this method be used in environments such as tunnels where GNSS is not available?

It’s a shame that the desktop uploader doesn’t have the ability to modify or trim routes, but I’m glad to know that RouteConverter and LosslessCut are available. I thought that editing the video file would completely destroy the position tag. I plan to try it out on my next street shoot.

(Whether or not to use Japanese on forums always bothers me. I’m a Japanese speaker so I’d like to use Japanese, but I also want more people to participate in the discussion. Do you have any ideas?)

2024年1月29日(月) 20:20 drivephotograph via Mapillary Community Forum <>:

Android smartphones use GNSS for positioning, and so does the iPhone. In tunnels, GNSS signals cannot be received, so positioning will be inaccurate.

To upload images taken in a tunnel to mapillary, you can choose from three different methods.
1: Use an IMU or similar to measure the accurate location.
2: Modify the trajectory manually using RouteConverter or other tools.
3: Exclude images shot inside the tunnel. Upload only images with accurate location information.

I also recommend as a tool to manually modify the trajectory.

This is available in your browser; it allows you to edit GPX files. The function to make the trajectory follow the road is very useful.
Press the “Load GPX” button at the top of the screen to load the GPX file. Press the “Edit track points” button (pencil icon) on the left side of the screen and check the “Routing(follow roads)” checkbox to follow roads.

It is better to communicate in English so that more people can read it.

Thank you everyone for the information. I tried various steps based on the information you provided and I think the steps below work for me.

  1. Extract GPX from mp4 using TELEMETRY EXTRACTOR for GoPro
    (I succeeded in removing inaccurate GPS information by setting GPS Precision to 144. Above 144, noise is mixed in, and below 144, good information is also lost.)
  2. (If there is a curved tunnel etc.) Use to match the trajectory to the road
  3. Upload your video with mapillary desktop uploader and add GPX file

Figure 1: When the video is uploaded as is

Figure 2: Exporting GPX with GPS Precision set to 144 using TELEMETRY EXTRACTOR

Figure 3: Using to match the trajectory to the road

However, with this method, if the speed of movement inside the tunnel changes due to traffic congestion, etc., it cannot be properly mapped. In this case, the only option seems to be to delete the video or use IMU.

Dead reckoning using an IMU is very attractive, but the information processing and device preparation are very demanding. It would be interesting to be able to use GoPro’s log data or sensor data from smartphone (gyro sensor or accelerometer), but it seems difficult.

1 Like