Log in or register to post comments

False positive

January 23, 2018 - 6:09am #3

Hi,

I face a big problem with object detection. My object is a Pepper robot (this guy). My robot is detected correctly but sometimes it detects random things (for exemple some glass wall or a QRCode or a chair and many many more...) before I have the chance to look at the robot, this is very annoying :(

I've attached some screenshots:

1) Screenshot when still nothing is detected (so you can see the 2D guide view)

2) Screenshot of when it detects the robot correctly

3) Screenshot of when it detects a QRCode instead of the robot

4) Screenshot of when it detects a glass wall instead of the robot

5) Screenshot of when it detects another glass wall instead of the robot

6) Screenshot of when it detect a chair instead of the robot

 

I've tested in the editor and on device, it's the same result.

As you can see false positives doesn't look like the robot at all, if someone could give me some clue on what I should do to avoid having those false positives I would be very grateful.

 

Thanks,

Pierre

AttachmentSize
Image icon 1.png989.7 KB
Image icon 2.png768.1 KB
Image icon 3.png696.07 KB
Image icon 4.png732.78 KB
Image icon 5.png601.89 KB
Image icon 6.png846.85 KB
pflamant
Joined:
December 7, 2017
Posts: 24
Top

False positive

February 13, 2018 - 11:52am #2

Hello Pierre,

This target contains many rounded/curved shapes with very few hard edges. This makes it not an ideal Model Target target and will likely result in false positives such as those you are experiencing. Please review this document for more information related to Model Targets: https://library.vuforia.com/articles/Solution/model-targets-user-guide.html

Thanks,

Vuforia Support

Strasza
Joined:
June 13, 2017
Posts: 1721
Top

False positive

February 14, 2018 - 2:46am #1

Hello Strasza and thanks for your answer,

I've read in details the documentation you attached and more importantly this part which is relevant in my case:

"3. Sufficient geometry. Geometric complexity is useful to distinguish an object from other shapes in the environment. Simple shapes such as cuboids, or very elongated simple shapes may be easily confused with other objects in the user’s setting."

 

There is no word of "rounded/curved shapes" in the doc and I thought that CAD based recognition wasn't based on features detection so curved shapes wasn't a problem with that kind of detection method, but apparently I'm wrong ^^

 

I'll try @saiunity astuce:

  • use different colors in the 3D model for different materials (e.g. plastic, chrome, metal, glass). The specific color is not relevant (i.e. the 3D model's colors do not need to exactly match the Model Target's), just the differences in colors.

 

 

G'day to you,

Pierre

pflamant
Joined:
December 7, 2017
Posts: 24
Top
Log in or register to post comments