Glitch with Active Track Sent Drone Toward Trees

[Basspig]

Testing Active Track in the Trace mode, I made this flight path twice. The first time it executed without any problems.. except that I'd fogotten to record video. So I repeated it. Only this time, while recording video. About halfway down the road, the P4P suddenly veered to the left, toward some pines that were lining the road. I had to fight it with the controls to prevent it from hitting something and then drop out of Active Track mode.

I also note that the video feed was breaking up rather badly during this second test run.
There was nothing else in the frame to confuse the AT mode. If a bird had flown across my path and the drone locked on to it, that would explain the sudden veering off locked object, but no bird flew across my path.

All firmware is current, and software is current, running on the Crystal Sky Ultrabright.

This sudden misbehavior makes me not want to risk using Intelligent Flight mode again, especially in tight surroundings with a lot of trees.

 

[Fly Dawg]

You don't suppose that the OA happened to detect anything do you? Take a look at your flight log and you will be able to tell.

 

[Basspig]

You don't suppose that the OA happened to detect anything do you? Take a look at your flight log and you will be able to tell.

I would have thought if a bird flew across my path that could explain it, but I didn't see any other object move near me.
Where are these flight logs? Are they in the Crystal Sky or are they embedded somewhere on the SD card in the drone?

 

[Fly Dawg]

They are in both. ( The Aircraft has an internal card, seperate from the SD card) The easiest one to obtain is on the device ( I use device to refer to the Crystal Sky or another device). Follow the link below and the instructions are there. If you would like others to take a look as well if you are not familiar with these, once you upload the file, place a link back here to the upload.

DJI Flight Log Viewer

 

[Basspig]

Thanks for that handy link! I was able to identify which of the three log files from today's flight was the correct one by the map with the paths drawn on it.

Here's the data for the pertinent flight segment:

2m 2.6s ActiveTrack 11satellites 8.2ft 9.8ft 1.2mph 58.3ft 93% 16.422V 4.106V 4.106V 4.101V 4.109V 0.008V Subject Lost. Searching..


So somehow the drone lost track of me. Could it be because I turned my back to the drone and started jogging? Was it going by face recognition? I thought it was going by contrast and the fact that the subject was selected manually at the start.
The fact that the drone didn't stay put when it lost the subject is unsettling.

Out on the soccer field downtown, I've followed my daughter from much greater distances without losing lock. I found that the only time it lost lock was if the drone was more than 75' from the person being tracked. But then it didn't do anything but hover. So why it veered off the the left at relatively high speed is what raises concern. It should have hovered.

Granted, I am in a wooded area and had 11 satellites.

I will note that the video feed was breaking up a lot during Intelligent Flight mode. If the software on the Crystal Sky is doing the recognition, then it probably depends on a clean video feed to recognize the subject, which it can't do if half the picture is a block of frozen pixels, as was the case all through this test until I cancelled IFM. Then the video feed was perfect after that (P-GPS mode).

 

[Fly Dawg]

So somehow the drone lost track of me.

Can't say for sure from the limited information. Will need to see the full log file to confirm or deny that point. Post a link back here to that uploaded file and can take a look see to confirm.....there could be other explanations for that.

 

[teccer]

((Could it be because I turned my back to the drone and started jogging?)))


you do realize that it could be that if the controller was facing away from the AC the controls are opposite, the same as if you are flying towards yourself

 

[Fly Dawg]

you do realize that it could be that if the controller was facing away from the AC the controls are opposite, the same as if you are flying towards yourself

That would not influence active track.....as this is automatically adjusted by the tracking algorithm. RC controls have nothing to do with that.

 

[Basspig]

Plus I was only 12' from the aircraft and I've turned away, even with the AC 100' away and didn't lose radio contact.

The thing that continues to bother me is that the video feed freezes as soon as I enable one of the Intelligent Flight Modes and clears up to perfection immediately upon cancelling IFM. DJI has been tight-lipped about this, but I have a suspicion the CPU in the CS is too weak to handle both video decoding and IFM at the same time.

 

[Basspig]

Here's a link to that log file: Dropbox - DJIFlightRecord_2018-06-10_[13-24-12].txt

 

[Fly Dawg]

The thing that continues to bother me is that the video feed freezes as soon as I enable one of the Intelligent Flight Modes and clears up to perfection immediately upon cancelling IFM. DJI has been tight-lipped about this, but I have a suspicion the CPU in the CS is too weak to handle both video decoding and IFM at the same time.

Many devices have this issue, however this has little to do with your incident in question. Pixels are read in software, not from the display itself. This is the basics of tracking. The video feed gives you the reference of the tracked pixels, not the pixels within the algorithm that tracks those selected pixels. The visual reference may be slow but the tracking is done in the algorithm. Agreed that visually and within the software itself, it could "drop out" due to the pixel reference mismatch.

 

[Fly Dawg]

Here's a link to that log file

Obviously you misread Post#4......This will need to be re-uploaded or offline converted....Use the link from the file @phantomhelp....The .txt files can be directly uploaded here, but have to be converted......

 

[Basspig]

I didn't believe that link would work outside the machine that requested it. Well if it does, then here it is:

DJI Flight Log Viewer - PhantomHelp.com

 

[Fly Dawg]

I didn't believe that link would work outside the machine that requested it. Well if it does, then here it is:

Bingo....thats the way it works.....

 

[Fly Dawg]

Ok, just a quick look at this you actually lost tracking twice.......Just a few questions before I go through a full detailed look. What was the contrast between what you were tracking and the area itself? The tracking area (or subject) needs to be very well defined for the tracker to maintain visual. Being that you were in a wooded area, this is not a great place to use this function, unless the subject has far greater contrast than the surrounding area. For example, if you were attempting to track yourself, you would need to have a very bright contrasting color that would differentiate yourself from the surroundings. A bit like an orange hunters vest in the woods. Sometimes not that drastic, but you see the point. The tracker algorithms in these drones are not designed to differentiate hunter green from olive green in that much detail. So that is my first thought. It will take some time to look deeper than that.

 

[Meta4]

This sudden misbehavior makes me not want to risk using Intelligent Flight mode again, especially in tight surroundings with a lot of trees.

Glitches happen.
Using automated flight modes in a tree rich environment is not a good idea.
Whenever you need to quickly cancel any automated flight, simply flick the flight mode switch to atti and back to P-GPS again and it's cancelled and you have full control.

 

[Basspig]

Ok, just a quick look at this you actually lost tracking twice.......Just a few questions before I go through a full detailed look. What was the contrast between what you were tracking and the area itself? .....

I was wearing a white T-shirt, so that was highly visible against the brownish earth tones behind me. And I was only a dozen feet from the drone. This thing has tracked my daughter from 100' away without losing lock, while on a soccer field.

 

[Fly Dawg]

This thing has tracked my daughter from 100' away without losing lock, while on a soccer field.

This could very well be, as that is a much more open space with less interfering contrast as both myself and Meta4 mentioned, a highly wooded area is not a great spot to utilize the automated features, and this includes active track. There is too much clutter for the algorithm to select the proper pixels within the track box. You will have much better success in open areas.

 

[Basspig]

This could very well be, as that is a much more open space with less interfering contrast as both myself and Meta4 mentioned, a highly wooded area is not a great spot to utilize the automated features, and this includes active track. There is too much clutter for the algorithm to select the proper pixels within the track box. You will have much better success in open areas.

The thing was, she was wearing black, not a very good contrast to the grass, and I was wearing white, a huge contrast to the background, plus I was on a 30' wide road, walking backwards as the drone followed me. On the first run, it followed perfectly, but I forgot to record video, so I did a second run while recording and that's when it happened.

 

[Fly Dawg]

On the first run, it followed perfectly, but I forgot to record video, so I did a second run while recording and that's when it happened.

That is quite plausible in fact. On the second run, you may have "appeared" to be in the same spot visually, but the tracker's pixel recognition changed from that of the first run. It doesn't take much to break lock under these circumstances. This could be anything from the tracker box size to, sun, shade, etc..etc..