Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Peng_L_2
Beginner
47 Views

RealSense F200 Depth Map data inconsistency

The depth map provided by Realsense F200 is not consistent along time line. It will gradually change slowly and then at some random time a big jump will happen.https://youtu.be/g6ZqQdHjjns (see the last 10s of this video to see this jump)

 

In this video the F200 is looking at a corner with three plane perpendicular to each other, (as shown below)

IMG_0598.JPG

the green point cloud in the above youtube video represent the depth map at frame 10 while red point cloud represent the live depth map. Ideally, these two point cloud will be perfect aligned (you will see red and green points mixed together) since the F200 is fixed. But as at end of the video shows, these two point cloud is not aligned (you can see lagre red and green parts), which means the calibration of the external params of depth sensor is useless since the depth map is not consistent over time. 

Hopefully someone from intel could elaborate on that

0 Kudos
3 Replies
samontab
Valued Contributor II
47 Views

Probably the gradual changes that you see are thermally generated noise on the camera sensor. It makes sense that it will slowly change over time due to heat. Have you tried leaving it on for a long time?, maybe it will get to a stable operation temperature.

The big jump however is more difficult to explain. Maybe you suddenly changed the environment? (turned on the lights?), or maybe there is an overflow of data, or some other type of data mishandling by any of the software that you are using to render the points.

But all of these things are normal for a real sensor. There's always going to be noise mixed with the signal, and it usually changes over time, that's why calibration cannot be done "once and for all".

0 Kudos
Peng_L_2
Beginner
47 Views

samontab wrote:

Probably the gradual changes that you see are thermally generated noise on the camera sensor. It makes sense that it will slowly change over time due to heat. Have you tried leaving it on for a long time?, maybe it will get to a stable operation temperature.

The big jump however is more difficult to explain. Maybe you suddenly changed the environment? (turned on the lights?), or maybe there is an overflow of data, or some other type of data mishandling by any of the software that you are using to render the points.

But all of these things are normal for a real sensor. There's always going to be noise mixed with the signal, and it usually changes over time, that's why calibration cannot be done "once and for all".

Thanks for such a quick reply :-)

That gradual changes and big jump will happen almost every time you start the program ( the big jump will randomly happen within 1 min after the program start), and it's not caused by any environment change (we have kept the sensor running awhile before we made that video, so the thermally generated noise assumption may not hold), the simple program we developed is as straight forward as the sample, which may not have werid behavior that may cause the big jump (but we will double check on that) . We are going to test other F200 sensors to check whether it's caused by defective unit.

Our guess is that the F200 driver somehow can calibrate its self after around 1 mins running time, and apply its new params, which cause that big jump. But we need someone from intel to confirm that.

0 Kudos
Peng_L_2
Beginner
47 Views

samontab wrote:

Probably the gradual changes that you see are thermally generated noise on the camera sensor. It makes sense that it will slowly change over time due to heat. Have you tried leaving it on for a long time?, maybe it will get to a stable operation temperature.

The big jump however is more difficult to explain. Maybe you suddenly changed the environment? (turned on the lights?), or maybe there is an overflow of data, or some other type of data mishandling by any of the software that you are using to render the points.

But all of these things are normal for a real sensor. There's always going to be noise mixed with the signal, and it usually changes over time, that's why calibration cannot be done "once and for all".

 

We actually have observed the same phenomenon on other realsensor F200 (the gradually change, and the big jump) So we highly suspect that the problem (or internal calibration maybe?) comes from Driver or even the hardware its self. It will be great if someone could confirm that this behaviour is caused by something inside F200 driver or hardware, so we could avoid spending useless time to try testing and solve it.

 

Thanks

 

Peng

0 Kudos