In the second part of this post I will show you how to add metadata to images previously extracted from frames.
However, the frames extracted using ffmpeg did not copy over vital metadata like capture time and GPS data used for mapping.
The next step is to re-embed the video metadata into the extracted frames.
I’ll be using the same video to demonstrate this process as the one used last week.
1. Extract metadata track from video file (for telemetry information)
$ exiftool -ee -p gpx.fmt VIDEO_7152.mp4 > VIDEO_7152.gpx
Note, this a simple extraction and only preserves
GPSDateTime values. In many cases the telemetry track also includes other sensor data including pitch and heading.
2. Extract global metadata (for camera info, etc.)
This process is described in more detail in the section of this post titled, extracting metadata for video level:
$ exiftool -G -a VIDEO_7152.mp4 > VIDEO_7152_metadata.txt
3. Add timestamps to frames
Before you can add the GPS track you need to add the time each frame was taken (the
datetimeoriginal EXIF value of the image).
To calculate the time the image was taken, two pieces of information are needed; the start time of the video, and the spacing (time) between frames defined when splitting the videos into frames.
In last weeks example the spacing of frames is 1 second (
$ ffmpeg -i VIDEO_7152.mp4 -r 1 FRAMES/img%d.jpg
Videos are stitched on the GoPro Fusion, and a number of other cameras, after the date the time they shot (using desktop software).
This can be problematic.
CreateDate of the video (extracted during step 1) shows:
[Main] CreateDate : 2020:04:15 09:14:04
This is the date I stitched the file.
But if you look at the first GPS sample time:
[Doc1] SampleTime : 0 s [Doc1] SampleDuration : 1.00 s [Doc1] DeviceName : Fusion [Doc1] TimeStamp : 1066.344245 [Doc1] CameraTemperature : 27.048828125 C [Doc1] Accelerometer : (Binary data 10227 bytes, use -b option to extract) [Doc1] TimeStamp : 1066.365138 [...] [Doc1] GPSMeasureMode : 3-Dimensional Measurement [Doc1] GPSDateTime : 2020:04:13 15:37:22.444
The GPS shows two days previous to the
GPSDateTime is the time I actually shot the video.
Therefore, be careful not to use the
CreateDate before being absolutely sure it’s correct.
End of warning
Using this logic, you know
img0001.jpg, the first frame, was captured within +/-1 second of
2020:04:13 15:37:22.444 because the
0 s and
2020:04:13 15:37:22.444 as the video start data time, you know (given frames extracted 1 second apart using
-r 1) that all following frames increment by +1 second.
2020:04:13 15:37:24.444 …,
2020:04:13 15:37:39 which matches the video
15.98 s (
2020:04:13 15:37:39.444 -
2020:04:13 15:37:22.444 =
00:00:17. Don’t forget to count from 0 (0s to 16s = 17s).
You can use exiftool to do this pragmatically in two steps (as I discovered thanks to this answer on the exiftool forum)
3.1. First set all images to the same date/time
$ exiftool -datetimeoriginal="2020:04:13 15:37:22" FRAMES/
Which will create 18 files with the same
You’ll also notice exiftool preserved the original data creating a set of files with the extension
.jpg_original. This is important to consider for step 2.
3.2. Increment the
Now you need to increase each file by +1 seconds from the previous image:
$ exiftool -fileorder FileName -ext jpg "-datetimeoriginal+<0:0:$filesequence" FRAMES/
On Mac/Linux, use single quotes (‘) instead of double quotes (“) around arguments containisng a dollar sign ($) when uing exiftool.
The response gives:
1 directories scanned 17 image files updated 1 image files unchanged
1 image files unchanged because this is the first file (
img0001.jpg) in the sequence with the correct date (set by making sure files are in ascending alphanumeric order using
-fileorder FileName). The original files were omitted because I only wanted files with
.jpg extensions (
-ext jpg) not the originals with
If you check the
datetimeoriginal in the subsequent files (
img0018.jpg) you should see they all update by +1, with
img0018.jpg ending at
4. Geotag the frames
Now that the frames have
datetimeoriginal values you can geotag the photos, again using exiftool, using the track file extracted during step one.
exiftool -ext jpg -geotag VIDEO_7152.gpx FRAMES/
exiftool loads the GPS track log file, and matches the GPS position time with the time of the image to determine and tag its location.
Other track formats
exiftool also supports the following telemetry track formats in addition to
.gpx, should you have used a separate GPS logger.
- NMEA (RMC, GGA, GLL and GSA sentences)
- IGC (glider format)
- Garmin XML and TCX
- Magellan eXplorist PMGNTRK
- Honeywell PTNTHPR (see Orientation)
- Bramor gEO log
- Winplus Beacon .TXT
- GPS/IMU .CSV
- ExifTool .CSV file
5. Add EXIF/XMP tags
There are a variety of other fields in the original video metadata you will want to include into each image file.
For example, the cameras make or model. For 360 images this will also include XMP tags to define the projection type (so it’s loaded correctly by 360 viewing software).
During step 2, I captured the original metadata from the video. e.g.
[Composite] Image Size : 3840x1920
You can now use these values to determine the tag values to inject.
exiftool -make="GoPro" -model=Fusion -ProjectionType=equirectangular -UsePanoramaViewer=True -CroppedAreaImageWidthPixels=3840 -CroppedAreaImageHeightPixels=1920 -FullPanoWidthPixels=3840 -FullPanoHeightPixels=1920 -CroppedAreaLeftPixels=0 -CroppedAreaTopPixels=0
These EXIF and XMP values provide enough metadata to upload to Map the Paths, Mapillary, and Google Street View… though you can add other tags as you wish.
You might also like the post; Lessons learned when geotagging timelapse photos and video frames