I'd just notice when i put my Spark on the table (solid and flat), the camera move horizontally slowly... movement doenst travel too far just a lil bit.

Does this meant Spark have 3rd gimbal within the camera housing?

And im curious to know why Dji design the camera lens seated on one side of the camera housing.




(Photo source: Google)

Yes mine does this too. it looks like a software bug
The sensor is 4k and the image is being stabilised in yaw however it makes it appear that the spark is yawing around and I have tried to correct this many times with the sticks. When you look at the spark it is not moving at all

Confirmed by DJI bug with software stabilization.

JBSpark Posted at 2017-8-17 20:54

The sensor is 4k and the image is being stabilised in yaw

As far as my logic goes, the digital stabilization has nothing to do with the image sensor being 4K.
The digital stabilization can only use pixels outside of the frame to stabilize the footage, so they reserve some pixels on all 4 edges for this.
You can see how it works on some camera's that allow to switch on / of image digital stabilization. (I can upload a video from my compact if you want)

It just croppes some of the edges and effectively 'zooms in' a bit.

Of course they could use only the center part of the sensor (1920 pixels of the 4000) and use the other 2180 for stabilization, but that would make a very 'zoomed in' image.  (this is digital zoom)
And this is clearly not the case, since the Spark has a rather wide field of view.
So my conclusion is that other pixels are just dropped by the software by downscaling the footage.

No big issue for me, since my workflow will remain on 1080p for some years.

It happens to me too. I often see that when I make a left or right turn and stop, the picture of the camera tends to move in the opposite direction for some seconds. Its annoying ****!

Mine doesnt do that. Maybe this could be used as an extra test to check if yaw gyro bias is well calibrated. Maybe if the Spark tries to compensate electronicaly at camera level is because it believes is under yaw rotation caused by too high gyro bias

I'm also very curious about that?

If the gimbal moves only when you power on the drone, it is common, the gimbal and the drone need some time to finish the self-checking. Or if it moves during flight, please calibrate the compass.

I think its normal, i had before the spark the Parrot Bebop, that drone uses 100% digital stabilization on all AXIS, one thing i learn is that its just works 100% when drone are flying, the digital stabilization works  comparing previous image with the next and catching info at real time from IMU to compensates and keep stabilization , if  drone are perfect steady, theres no previous image to compare and stabilize, so the software make this type off thing like move a little the image and if drone are not flying it dont give real time IMU flight information to digital stabilization, at least, is how Parrot BEBOP stabilization works, and why parrot bebop footages are much better stabilized when drone are moving and not steady in the air.

If a drone uses some type of digital stabilization, it means the sensor size are bigger than final footage resolution, maybe spark sensor are 4K, just take a look at Parrot Bebop example..

FULL sensor photo taken by me with bebop and i will make a frame explaining how it works:

You’re experiencing normal gyro drift.

Gyros do not report absolute position - they report relative movement.
The more still a gyro is, the more noise and less useful signal from that gyro you’ll get.

If a gyro’s noise is biased to one particular direction, it will slowly drift in that direction.

The reason you don’t experience this with the other two axes has nothing to do with digital stabilization and everything to do with the fact it’s the yaw gyro. For pitch and roll, sensor fusion with accelerometer output gives you a solid *absolute* reference, thanks to the gravity (the gimbal always knows which way is ”down” thanks to the ever present gravity vector).

For yaw, there is no reference (you’d have to use a magnetometer for that, i.e. compass, which would be a bad idea on a gimbal, or some kind of an external reference, which is not applicable on a drone).

Therefore, yaw drifts. Pitch and roll do not.

Montfrooij Posted at 2017-8-17 22:53
As far as my logic goes, the digital stabilization has nothing to do with the image sensor being 4K.
The digital stabilization can only use pixels outside of the frame to stabilize the footage, so they reserve some pixels on all 4 edges for this.
You can see how it works on some camera's that allow to switch on / of image digital stabilization. (I can upload a video from my compact if you want)

"Of course they could use only the center part of the sensor (1920 pixels of the 4000) and use the other 2180 for stabilization, but that would make a very 'zoomed in' image"

You noticed the size change between video/photo mode on the display?

I think, they do use only the 1920/1080 from the middle and have the "rest" do the digital stabilizing.

And no, no digital zoom.

I saw a YouTube video on the weekend and the guy explains the difference between the camera sensors of the Mavic to the Spark.  Turns out it's the same sensors (Sony) for both Mavic Pro and Spark.  on the Mavic Pro the sensor uses the full range of the 4k sensor because of the three axis gimbal.  On the Spark, they use the 4k sensor but reduce the footprint to allow for the 3 axis to stabilize the image due to the two axis gimbal.  This was a pretty interesting video and very informational.

Sparky_17 Posted at 2017-8-22 06:16
I saw a YouTube video on the weekend and the guy explains the difference between the camera sensors of the Mavic to the Spark.  Turns out it's the same sensors (Sony) for both Mavic Pro and Spark.  on the Mavic Pro the sensor uses the full range of the 4k sensor because of the three axis gimbal.  On the Spark, they use the 4k sensor but reduce the footprint to allow for the 3 axis to stabilize the image due to the two axis gimbal.  This was a pretty interesting video and very informational.

DJI could make a gift to us spark users and let us in future choose between 4k videos with just 2 axes and 1080p with 3 axis (1 digital).

- 2 axis is not that bad for some type of footages, whe need to remember Phantom 2 2d gimbal ages.

On my S7 Edge i have the choice to use 4K record with just optical stabilization or 1080p with digital stabilization + optical stabilization.

STech - Hathder Posted at 2017-8-22 06:22
DJI could make a gift to us spark users and let us in future choose between 4k videos with just 2 axes and 1080p with 3 axis (1 digital).

- 2 axis is not that bad for some type of footages, whe need to remember Phantom 2 2d gimbal ages.

That is a very good point / idea

Sparky_17 Posted at 2017-8-22 06:16
I saw a YouTube video on the weekend and the guy explains the difference between the camera sensors of the Mavic to the Spark.  Turns out it's the same sensors (Sony) for both Mavic Pro and Spark.  on the Mavic Pro the sensor uses the full range of the 4k sensor because of the three axis gimbal.  On the Spark, they use the 4k sensor but reduce the footprint to allow for the 3 axis to stabilize the image due to the two axis gimbal.  This was a pretty interesting video and very informational.

link, please?

Here is the link to the video

Sparky_17 Posted at 2017-8-22 10:01
Here is the link to the video

To my believe this guy is missing one important factor.
The digital stabilization can only use the OUTSIDE of the frame to compensate for movement.
Other pixels (inside the frame) won't help since they have to stay inside the frame and won't compensate for movement.
Digital stabilization is just about making changes to the crop location every frame and so you can only use the pixels outside of the frame for it.
I sometimes use a cheap compact camera (underwater camera from Olympus) that shows how it is done when you activate the digital stabilization. It just crops a bit of the edges.
So coming back to this video, the difference between 4K and 1080 is far too much to cut off.
If they only use the center 1920 x 1080 of the sensor, you would see the effect of digital zoom and a much more narrow field of view. This is obviously not the case, as the Spark has a rather wide field of view (even wider than the Mavic).
So I believe this 1080 only is mainly a marketing trick to justify the price difference.
And maybe caused by slower hardware (cheaper) like the CPU that could not handle the large data stream.
Since some of it's features rely on analyzing the video stream 'on the fly' (no pun intended) which does require serious calculating power at 1080 and even more with a 4K stream.
I did also comment his video, but no response...

Montfrooij Posted at 2017-8-22 12:16
To my believe this guy is missing one important factor.
The digital stabilization can only use the OUTSIDE of the frame to compensate for movement.
Other pixels (inside the frame) won't help since they have to stay inside the frame and won't compensate for movement.

I think the limited processing power (and thermals) is the main reason.
Marketing reasons are just a bonus.

By the way, there’s a very simple way to find out how much of the image is cropped for stabilization - just compare a frame of video to a photo.

0rcinus Posted at 2017-8-22 17:13
I think the limited processing power (and thermals) is the main reason.
Marketing reasons are just a bonus.

I agree.
So cost efficiency it is.
BTW, no problem for me, I'll be fine with 1080 for some years.
My workflow is not optimized for rendering 4K images and a new editing PC + new TV would be a bit too much for me.

Ya i'm satisfied with 1080p too... watching it on a Full HD tv look promising to me.

Btw i will put hope that the next firmware update DJI will give Spark a 2k video upgrade... since there are more than enough pixel for their Digital Stabiliser purposes.

Alagapas Posted at 2017-9-13 18:56
Ya i'm satisfied with 1080p too... watching it on a Full HD tv look promising to me.

Btw i will put hope that the next firmware update DJI will give Spark a 2k video upgrade... since there are more than enough pixel for their Digital Stabiliser purposes.

Alagapas, thanks for your advice. Our engineers are working on it to provide users a better drone.

Alagapas Posted at 2017-9-13 18:56
Ya i'm satisfied with 1080p too... watching it on a Full HD tv look promising to me.

Btw i will put hope that the next firmware update DJI will give Spark a 2k video upgrade... since there are more than enough pixel for their Digital Stabiliser purposes.

Having that would be great. It would also be nice to have an 1080p60 mode as an option for sports videos.

I've noticed this slow YAW evolution even without digital stabilization in photo mode too and it didn't allow me to take photos with long exposure. Sometimes Spark starts to slowly rotate by YAW axis. Maybe because of wind? All sensors are in green zone...

I see the same behavior as the OP shows in their video.  My assumption is that the horizontal electronic stabilization is just centering itself again after there is no further movement to be stabilized.

Yes, the same thing happens to me! The image slowly turns to the side with my Spark on the floor standing!

The problem appear after .500 fw upgrade. I take photos at night with spark ... and is very annoyng. in video mode the movement can not be seen.

djiuser_xLE6O03 Posted at 2017-9-15 00:28
The problem appear after .500 fw upgrade. I take photos at night with spark ... and is very annoyng. in video mode the movement can not be seen.

It didnt just appear, but I think it became more noticeable in the latest firmware.

DJI Elektra Posted at 2017-9-14 01:12
Alagapas, thanks for your advice. Our engineers are working on it to provide users a better drone.

wow~ Does "working on it" meant working on upgrading Spark video quality to something higher than 1080p? Thats a "WOW" !

Looking forward to it !!!