wrong bearing on thight curves

[Stefal]

Basic information

• Release version: 0.1.0
• System: Windows
• Capture Device: Action Camera

Steps to reproduce behavior

1. Add bearing to the pictures with some tight curves

Expected behavior

The bearing should follow the curve

Actual behavior

There is too much bearing as you can see on the screenshot:

A correct angle would be 190° instead of 174°

A correct angle would be 20° instead of 352°

[jernejaMislej]

Hi, sorry for the late reply, I m on vacation until 10th August with limited access to internet, @ybkuang will try to address the issues while I am gone.
Just to make sure, the directions were interpolated with the processing tools?

[Stefal]

Yes, with the mapillary_tools 0.1.0

[jernejaMislej]

Ok thanks.
If the direction is missing in EXIF or the tools are run with --advanced --interpolate_directions then the direction is derived based on latitude and longitude here and then here
Unfortunately it is very likely that the formula is not accurate for tight curves since this information can not be derived from latitude and longitude only. If you know of a better formula, you are welcome to contribute.

[Stefal]

This "bug" seems related to #272 as images are not time-sorted correctly.

[jernejaMislej]

hi,
why do you think the images are not time-sorted correctly? can you give some examples?

[Stefal]

It's a partial mistake.

1. As the script add the subsecond in the datetimeoriginal, if you process an image with a subsectimeoriginal exif date, the subsecond value will be read 2 times and added. An image with these data:
   -datetimeoriginal = 2018-08-21_15h00mn59s
   -subsectimeoriginal = 9
   After mapillary_tools process, it becomes
   -datetimeoriginal = 2018-08-21_15h00mn59s9
   -subsectimeoriginal = 9
   And will be read as
   -2018-08-21_15h01mn800ms
   Why?
   Because in https:/mapillary/mapillary_tools/blob/master/mapillary_tools/exif_read.py There is this code in the extract_capture_time method:
   

   mapillary_tools/mapillary_tools/exif_read.py

   Lines 168 to 171 in 2715b5f

   	capture_time = format_time(capture_time)
   	sub_sec = self.extract_subsec()
   	capture_time = capture_time + \
   	datetime.timedelta(seconds=float("0." + sub_sec))

So, you really could have a wrong sorting.

2. After a new check, I didn't encounter this case.

Perhaps it could be better to computer the bearing with the previous image too, and return the average value.

[jernejaMislej]

thanks for the feedback, i will have a look at that part of code and check this