Trouble with object tracking

[Richard_F_2]

I'm running into trouble with object tracking.  As a test, I downloaded the rune for F from wikipedia.  It's a simple set of black lines on a white background.  I loaded that into the object tracking sample, and printed out the run on white paper.  The sample was unable to pickup the image.  I've attached the image.  I also tried to take a picture of the same rune written out on a wood piece, and the tracker was unable to find the wood tile.  I've attached the images in question.

Are there special image requirements to enable the camera to pick up the images correctly?  And could the camera position affect this?  The camera is on my desk and tilted up towards my face.

[Dirk_R_]

Hi Richard,

I've the same problems (https://software.intel.com/en-us/forums/topic/536604)
Simple patterns aren't recognized and detailed patterns like QR codes cannot be clearly identified and separated.

The object tracking features seems to be in a beta state.
Only one simple stream (640x480) is possible. So it is obvious that the recognition cannot be very detailed.

The question: When will be released a new SDK with better object tracking features?

[Colleen_C_Intel]

Tips for object tracking - usually better with some texture.....but in general:
Source Image - needs to be large - at least 75x75 pixels on the camera stream. 
Best to create a map for 3D tracking using the toolbox.  I'm privately emailing Richard F a rune.slam created from metaio toolbox. 
When using 2D - it's very sensitive to lighting, any twists in the paper. Best to capture using live camera stream. 
 

[Dirk_R_]

Colleen Culbertson (Intel) wrote:

Tips for object tracking - usually better with some texture.....but in general:
Source Image - needs to be large - at least 75x75 pixels on the camera stream. 

Do you know if different colors are recognized?
Are only patterns independent of their color detected?
I've realized that simple patterns (number, letter) are not detected.

[Colleen_C_Intel]

Re color: You can use a color image, but it's all converted to grayscale before processing, so you track black and white.
What's detected? : The main detection features are “corners”, where two edges in the image meet.  So shapes need to be complex enough to have many detected features (up to 50).  So - most letters and numbers probably don't have enough features to be accurately detected . But to give you a sense of what works - our RS developers say that it is relatively reliable to track Wingdings characters such as T, ä, ®, Ë even though they are fairly simple shapes.