Object localization

[hankm]

Hello,

I am trying to localize objects ( a mobile robot, for example ) in a 8.68m x 8.68m sized space as below. I was wondering if someone could help me with my questions below. 

1) There will be walls and other objects that could occlude the tracking device. Can I still robustly localize the tracker in the presence of such interference?
2) Buying more base-stations would solve the occlusion problem?
3) How many base-stations do I need to buy in order to cover the entire space in a single coordinate system densely? 
4) Could you please let me know the tracking accuracy?  
5) Is there any open source SW/ API support for reading the pose of the tracking device?

 

HTC vive seems very nice solution compared to other camera based motion tracking systems in terms of both price and performance.

 Best regards,

 

 

 

 

 

 

 

[HackPerception]

@hankm

• Basestation tracking isn't really meant for multi-room deployments. Everything is currently designed and optimized for usage in a single room. The SteamVR
• For anything over 5x5m, you'd need to use SteamVR v2.0 tracking. 2.0 tracking supports upto 4 base-stations into an array. This helps with occlusion but again, it's meant for single room setups. While you could probably get some basic results by mounting a base-station in each of the four corners, the inability for the tracked device to see all 4 stations at once would make it extremely challenging for SteamVR's algorithms to create a coherent "tracking universe"/roomsetup.
• A single instance of the SteamVR compositor (required for the pose estimation) supports upto 4 base-station. It can technically pull additional data from more than 4 stations but the way it does this is not documented and it cannot be used as a reliable system.
• Tracking accuracy is sub-millimeter and true accuracy depends on your deployment parameters. Every installation environment is different and things like reflective surfaces affect the accuracy.
• SteamVR is an entirely proprietary tech stack - there are no open source ways to interface with SteamVR tracked devices. You must use the OpenVR SDK and the corresponding SteamVR runtime to decode the data from a tracker and translate it into a pose estimate.

Kinect V2 is the other low-cost option for this kind of stuff since you can array multiple V2 sensors into an array. The SDKs are optimized for skeletal tracking but you may be able to use a marker based system for your use-case. More advanced systems like Opti-track would be $25,000-$60,000 for a tracking volume this size.

[hankm]

Thanks for the kind reply.
I did some research about the Opti-track system. But, it also doesn't guarantee to solve the occlusion problem.
Never heard of the Kinect V2 array setup. Could you kindly let me know any links or related technical documents for it?

Best,

 

[HackPerception]

@hankm https://azure.microsoft.com/en-us/services/kinect-dk/

There are only two solutions for occlusion

-Put tracking cameras on the device (which has a different set of occlusion limitations)

-Use more external sensors/cameras.

[hankm]

Thanks for your kind reply.

Could you please tell me if the configuration below is doable? Let's suppose that I have three sets of vive-baselines and three vive trackers. Is it possible to uniquely calibrate each vive tracker with respect to each set of base-station? For example, can we make the tracker VT1 responds only to the base-station B1 while VT2 responds to B2 only? If so, all we need to know is the coordinate transformation between B1 and B2.  This may sound crazy for you, but I would like to find a way to use Vive system...

Best,

 

[smsmsmsm]

On 7/1/2020 at 8:37 AM, hankm said:

Thanks for your kind reply.

Could you please tell me if the configuration below is doable? Let's suppose that I have three sets of vive-baselines and three vive trackers. Is it possible to uniquely calibrate each vive tracker with respect to each set of base-station? For example, can we make the tracker VT1 responds only to the base-station B1 while VT2 responds to B2 only? If so, all we need to know is the coordinate transformation between B1 and B2.  This may sound crazy for you, but I would like to find a way to use Vive system...

Best,

 

Were you able to do it??
I have a large area i wanted to localize 
was wondering if it worked out for u