Thanks, if you can get someone from the core team to comment more on this it would be very helpful to elaborate on the scanning/detected/tracking chain of events and perhaps influence how we react to them. I have a good grasp of this already except when it comes to that 'degree of confidence' value I was hoping could be exposed. Having that number would also still be useful, but having it with some info on how it is used in the tracker would be better.

Thanks,
-liam m-

I can inquire w/ the dev team, and I'm not sure of the precise confidence thresholds required to activate a DETECTED state under the various Camera Device Modes. Also keep in mind that detection and tracking requirements are a bit different. Detection involves identifying and differentiating a target from a search set, while tracking involves determining the pose of the target once detected, so the estimates from which you'd derive a confidence interval are different.

For instance, you'll notice that once a target is detected it can be tracked through conditions that would render the target undetectable. You can observe this when you approach a target from a distance. Once detected, you're able to increase the tracking distance well past the original detection point.

Would it be possible to expose this to the APIs? Though not always useful on every project we have a use case where knowing this would improve our feedback to the user as well as give us the option to (possibly) show proxies until the track has a high degree of confidence.

When a pose is reported, does this mean that the SDK has a 'full degree of confidence' or has it reached a threshold?

I suppose there are some thresholds we could build into app to prevent large transforms in the pose from taking place (ex. complete direction inversion) but knowing a bit more about how the tracker is fairing could help us quite a bit if we're willing to put in the extra work.

There is no measurement of that sort exposed by the API. You could analyze the trackable's pose data. Errors in pose estimation increase as the quality of tracking decreases, this will appear as random changes in the pose values (i.e. noise), but this 'jitter' is not directly comparable to detection confidence as there are other sources of error.