Max distance range skeleton tracking

Nuitrack SDK
1

Hello everybody!
I was testing the skeleton tracking with the Intel RealSense 415 and over 4 meters, the NuiTrack sample lost the skeleton information.
Is it possible to reach a higher distance range?
Do somebody hat experience on it?

Mariana

1 Like
2

I am also wondering if adding external IR pattern projector will help. Does anyone have experience on that?

3

I tested it with sun light and it wasn’t different … ?

4

Hi Mar,

There are some parameters in the nuitrack.config which control the possible tracking depths - at least in terms of the math itself.

While it may in theory be possible to update these values - there are some real physics reasons while doing so may not produce the desired results.

Firstly the sensor. All sensors have practical limits on their levels of accuracy the further the tracking point is away from the sensor.

The D415 is one of the best in class for longer distances

  • YES - intel claim it can track out to 10meters and beyond
  • BUT the level of (RMS errors) depth accuracy at those distances is starting to get rather high. For skeleton tracking where the quality of the depth data is very important - such RMS errors translate very quickly to very poor and unreliable tracking - to the point of being unusable.

Secondly, the physical realities. When you track an object at the center of the D415 sensor at distances over 4000m - the actual area of the sensor that the body is contained within starts to become rather small.

At close range - the 1280x720 pixels - even when downscaled to 640x480 provide the algorithms in the track with lot of pixel to work with - to help discriminate between what is a body and where its parts are.

However, at longer ranges - the effective area that you are tracking starts to become a window inside this area - something again that will cause the tracker to being to struggle

  • basically the number of pixels of depth data contained within the bound area of the body becomes too small to effectively discriminate where parts of the body should be.

For example - at close range - the head elements can be well identified - as can the hand (open / close)

  • but at distances with much less effective pixels to examine, these elements start to become blobs of less the viable information.

But regardless - I would suggest having a look at these settings below to see how they work for you.

Westa

=======================================================================
Have a look at the Skeletonization setting - it is hard coded to cap out at 4000 mm

"Skeletonization": {
    "MaxDistance": 4000,

Also have a look at the Realsense2Module settting - it is hard coded to cap out at 5000mm

"Realsense2Module": {
    "Depth": {
        "ProcessMaxDepth": 5000,
1 Like
5

Hello,
On the realsense site Intel describes the 435 as having better range than the D415 yet you seem to claim the opposite is true - could you please explain?
Thanks!

6

Hi GOB,

According to the information from Intel:

There are two main differences between the two models—field of view (FOV) and shutter type.

Their maximum range is the same (up to 10 meters). However, if we compare D415 and D435, the RMS error rate with D435 is twice as high as those of D415. With these sensors, noise increases with the distance, so we usually recommend to place RealSense at a distance of 1.5-2 meters from a tracked user.