Log in or register to post comments

Lowering the Confidence Interval

May 29, 2015 - 10:57am #1

Hi everyone,

Is it at all possible to lower the Confidence level used to detect an image. Say we want to detect an image that looks very similar to the image we have, but we don't care if they look exactly the same. Just generally the same. Is it at all possible to lower the confidence level so that we can detect any images that look like the one we've provided. Same goes for 3d objects. Say we are just looking for a cube with roughly these dimensions.

Thanks,
Eric

Lowering the Confidence Interval

June 11, 2015 - 2:27pm #4

ems316 wrote:

So it can't compare how many feature points it finds against the total feature points in your image and make a decision as to whether it found the image or not?

Using the "stones" target as an example, you can cover up a portion of the target and it can still be recognized and tracked. This will vary depending on feature point quality in the image and if enough unique points can be recognized to identify and track the target.

ems316 wrote:

Also, is it possible to change the amount of frames it analyzes per second...

No.

ems316 wrote:

When we are demoing it, sometimes our teapot will flicker. We want to change it so that once it has it tracked, it decreases the amount of times it re-searches a frame for an image, so that tracking is more consistent.

For tips on improving tracking, see this FAQ:

https://developer.vuforia.com/forum/faq/how-can-i-improve-detection-and-tracking-stability

 

Lowering the Confidence Interval

June 3, 2015 - 4:54am #3

So it can't compare how many feature points it finds against the total feature points in your image and make a decision as to whether it found the image or not? Also, is it possible to change the amount of frames it analyzes per second, so that once you've found the image, the tracking stays consistent. When we are demoing it, sometimes our teapot will flicker. We want to change it so that once it has it tracked, it decreases the amount of times it re-searches a frame for an image, so that tracking is more consistent.

Lowering the Confidence Interval

June 2, 2015 - 7:01pm #2

The recognition engine is looking for the feature points in the image targets. It can also recognize partial regions of points. There's no API option to manipulate the reco sensitivity in regards to feature point similarity.

https://developer.vuforia.com/library/articles/Best_Practices/Attributes-of-an-Ideal-Image-Target

 

Log in or register to post comments