Skeleton tracking using the R200 camera

[GProf]

Hi,

I'm interested in the skeleton tracking feature of the R200 sensor.

As far as I can tell from the https://software.intel.com/sites/landingpage/realsense/camera-sdk/v1.1/documentation/html/jointtype_personjoints_pxcpersontrackingdata.html documentation the sensor is able to track a full skeleton.I'm thinking about tracking a custom humanoid robot using the sensor.Would the module be able to detect humanoid figures around 50cm in height ?

(The proportions can match that of a child).

Is there any possibility to train/update the algorithm to improve detection/tracking for this scenario ?

Thank you,

George

[idata · Accepted solution]

Hi George,

Thanks for your interest in the Intel RealSense Technology. The SDK Skeleton Tracking algorithm tracks the body movement of a person and given its preview condition, it tracks only from the waist to the head, and six joints can be detected. I believe that tracking your custom humanoid robot might not work, however you could try and let us know your results. Additionally, you can create your custom algorithms using the SDK and try to improve them for your application. Regards,-Yermi A.

[MSobr1]

Yermi,

Could you clarify which 6 joints can be detected?

Thanks,

Michael

[GProf]

Hi Yermi,

Thank you for the feedback.

I've done some basic tests and managed to trick the SDK algorithm to track the robot's skeleton, but it is very inconsistent and the Person Tracking demo crashes often.I did try it myself and noticed the upper body joint tracking is pretty jittery and the labels get easily confused if hands are crossed in an X shape, or swapped if a hand is near the head.

It looks like a custom algorithm is the way to go.

[idata]

Hi Michael,

The RealSense SDK R2 exposes only 6 points in the upper body. Other users have mentioned that using the Skeleton tracking only shows 6 joints: JOINT_HAND_LEFT, JOINT_HAND_RIGHT, JOINT_HEAD, JOINT_SPINE_MID, JOINT_SHOWLDER_LEFT, and JOINT_SHOULDER_RIGHT. Hope this information helps. Regards,-Yermi A.

[MSobr1]

Thanks Yermi