Joints/Limbs sometimes remain in visualization after a person leaves the Kinect field-of-view

[rossmead]

We also saw the same effect for a second tracked person (in which case, it flickered between colors, as per issue #13).

[ekaszubski]

Screenshot is of out-of-date code due to issue with test build 1 for #11 (now fixed). If the issue still exists, please post a new screenshot. Thanks!

[rossmead]

Unfortunately, we're still getting the same issue, even after the most recent update (see version info below).

Version: 0.7.2-0+pr11-1 with 294d8e1cd1324c7004c1ef9c9924ad31 /usr/lib/x86_64-linux-gnu/libsemio_visualization_ros.so 6a835a840e3bde8e50a99983e9b5dada /usr/lib/semio_visualization_ros/semio_visualization_server_node

[ekaszubski]

Can you please post a screenshot as requested?

[ekaszubski]

Also, if possible, please verify whether any non-empty humanoids are being published by your humanoid source during the time the bug occurs

[rossmead]

Send the command(s) you want us to run.

[ekaszubski]

rostopic echo <humanoids topic>

[rossmead]

^ Updated screenshot (seconds after most recent image in #13).

Tried to verify if there were non-empty humanoids being published, but can't seem to echo Humanoid messages. Thoughts?

[ekaszubski]

Thanks for the updated screenshot. Confirmed using NiTE2 UserViewer sample that this is most likely being caused by NiTE temporarily tracking something else near you in the frame when you walk out.

Options to fix:

1. Require some minimum number/set of joints to constitute a humanoid being returned by HumanoidSourceNiTE
2. Require some minimum number/set of joints to constitute a humanoid being visualized by VisualizationAdapterHumanoid
3. Something else?

I think my preference is (1) because it will reduce confusion in other components of the system. Thoughts?

[ekaszubski]

Tracking in libsemio #35

[ekaszubski]

Needs new repro with libsemio >= 1.16.1 and libsemio-ros >= 0.8.0

[ekaszubski]

Moving this to the CRI: Attention: Nice-to-have milestone since it has had no recent activity. Will close after CRI: Attention: Nice-to-have expires.

[rossmead]

When I step out of the frame, it looks like some of the joints (legs and/or arms?) are popping up with a confidence of 0.5, which must be why they're still showing up. Maybe if the "required" NiTE joints aren't available, set the confidence for all other joints to 0.0 for NiTE output and publish at least once (or just force the empty list there). Weird...

Also, I did somehow manage to get it to be an empty list at one point, but I haven't been able to recreate that.

Hopefully this helps. Can discuss more as needed.

[ekaszubski]

Many thanks for the extra debugging. It seems like one or more nodes are ignoring the tracking state of the humanoid, which should be set to TrackingState::NONE when the confidence requirements in HumanoidSourceNiTE are not met. Can you verify whether the tracking state of the humanoid is actually set to NONE when this issue pops up?

[ekaszubski]

It has been confirmed that the tracking state is being set correctly; looking through other components for incorrect handling of tracking state.

[ekaszubski]

Found multiple places where tracking state was not being checked; these have been fixed in #27. Please try to repro with the latest test build in #27. Thanks!

[ekaszubski]

Appears to be fixed by #27; please re-open if this pops up again.