Novatek MP4 detection not working out-of-the-box; New FPS logic is buggy

[GPSMapper]

1. Novatek MP4 detection not working
   

   random-mapillary-stuff/ts_processor.py

   Line 259 in 53896f8

   if box.type.decode("utf-8") == "gps": #has Novatek-specific stuff

   
   this will not work with MP4 novatek (gps box is not detected, as it is a subatom under moov)
   Quick workaround: change "gps" to "moov"
   if box.type.decode("utf-8") == "moov": #has Novatek-specific stuff

2. new FPS calculation approach is buggy.
   

   random-mapillary-stuff/ts_processor.py

   Line 562 in 53896f8

   fps = length/packetno

   
   In MP4 Novatek files you could expect 1-2 additional GPS location packets which are carried over between the files.
   So it's typical to have 61 or 181 gps packets ('freeGPS') for 60 or 180 sec video
   As a result, fps is calculated incorrectly and image sampling gets broken.
   For 180 sec video with 1 sec sampling interval you will get 187 frames

How to reproduce:
Sample video: https://drive.google.com/file/d/14AXSVYlToIxUmoGPPmq-lGHsb2S8PWTH/view?usp=sharing
Script parameters: ts_processor.py --input 20210426103432_001156.MP4 --sampling_interval 1

Output:

20210426103432_001156.MP4
Novatek MP4 T
FPS : 30.0; LEN: 5400
GPS data analysis ended, no of points  181
Frames per point:  29.834254143646408
Video extraction started
20210426103432_001156.MP4  processed,  187  images extracted

So, at first FPS is detected properly by cv2

random-mapillary-stuff/ts_processor.py

Line 541 in 53896f8

fps = video.get(cv2.CAP_PROP_FPS)

but later it get's overwritten by

random-mapillary-stuff/ts_processor.py

Line 562 in 53896f8

fps = length/packetno

and while packetno=181 (contains one extra gps packet), resulting FPS is not correct and this breaks the sampling
Frames per point: 29.834254143646408
Extra images are produced (not @ 1fps interval defined) and coordinates are assigned incorrectly.
Similar issue will occur if some of there will be a temp gpsfix loss. (i.e. in the tunnel) Packet will not be captured and counted and fps calculation will be completely erroneous

IMO, original approach works excellent (fps extracted by cv2) and there is no need to redefine it.
Quick workaround: delete or comment lines 559-561

    #if packetno > 0:
        #print ("Frames per point: ", length/packetno)
        #fps = length/packetno