recognition symbols on the dashboard

[Predseda]

Hi there,

Is there anybody, who can help me with my issue in LV, Image Development?

Sources:

• I have a video in MPEG-4 (from GoPro camera).
• I have pictograms from manual (ISO standards)

What I need:

• An algorithm which assign a picture from the dashboard to the design of the manual.

Problem:

• The camera can’t be mounted always in the same place at the same angle and the same distance from the screen.

Does anyone have an idea which algorithm could be used? Or at least where I can get some resources?

Do you have an idea which modules could be used or which method is the best for my issue?
I found only some basic tutorials like THIS. But I can’t use it, because it doesn’t work when I change angle or distance from dashboard picture.

Thank you in advance

P.

[Martiko]

Hello Predseda,

may I ask You what is the purpose of this application? Are You trying to read values from screen and process them somehow afterwards? 

I am not asking to doubt Your approach but to understand the purpose.

If I understand it right, You are basically looking for some online optical character recognition (OCR) tool? Or is the aim to recognize the icon of the cruise control?

I believe that You can use some kind of OCR where You will teach the shape of icon as a character and later represent that character according to Your application needs.

If You provide more information, we can take a closer look.

[Predseda]

Hi,

thanks for response.

Yes, it looks like OCR. (or Machine Learning?)

I have set of symbols (in pdf/jpeg) which are randomly switched on and off (not all in one time, maybe more in one time).

I need app which find them in the ROI in video. If it finds symbol in ROI it will write name of symbol to the table.

If you have some idea how to make it, it will be great.

Thanks

P.

[Martiko]

Hi Predseda,

the easiest way how to find what is on the video for me seems to be via Vision Builder for Automated Inspection - VBAI (http://www.ni.com/vision/software/vbai/).

You can take collection of pictures (can be printed pdf if You can not light them in car) from different angles and then teach them with few clicks in OCR Training application (part of VBAI). Based on what You teach the program will later be able to recognize the pictures as characters.

You can download VBAI and try to play with it. If You need more detailed instructions, please contact Your local technical support.

[Brad]

If the pictograms will be at different angles and distances from the camera, I would recommend using geometric pattern matching. There is even a way to learn multiple templates and have the algorithm find all the objects and tell you which template they most closely match. OCR doesn't work well if the object can be rotated or are at different sizes because the camera distance isn't consistent. Try out vision assistant to experiment with the geometric pattern matching algorithm and see how it works in your application.

Hope that helps
Brad