LabVIEW
extract numbers from image
I don't have the vision toolkit installed right now, but I know that there is a function for reading a 7 segment display, and there is an example in the example finder. The trick I remember using this is that the first time you use it all segments must be on. Then the next read can have some of them off and it will tell you the number from the image.
Of course in the end this solution didn't work for me. I couldn't control my display enough to set all segments to on. Instead I took a baseline image when all segments are off (I turned the power off to them) then turn the device on and subtracted the two images. Then setup a threshold to a binary image. Then setup edge detection on 3 lines, one going vertical (right down the center to find the 3 segments) and two going horizontal (each finding 2 segments). For a segment to be on the edge detection needed to see both a low to high, and a high to low in the space expected. Then there was a conversion from what segments were on to what the number was. The requirement was that we could detect if a segment was out which was more important then the actual number which is another reason why I ended up making my own routine.
@Hooovahh wrote:
I don't have the vision toolkit installed right now, but I know that there is a function for reading a 7 segment display, and there is an example in the example finder. The trick I remember using this is that the first time you use it all segments must be on. Then the next read can have some of them off and it will tell you the number from the image.
Of course in the end this solution didn't work for me. I couldn't control my display enough to set all segments to on. Instead I took a baseline image when all segments are off (I turned the power off to them) then turn the device on and subtracted the two images. Then setup a threshold to a binary image. Then setup edge detection on 3 lines, one going vertical (right down the center to find the 3 segments) and two going horizontal (each finding 2 segments). For a segment to be on the edge detection needed to see both a low to high, and a high to low in the space expected. Then there was a conversion from what segments were on to what the number was. The requirement was that we could detect if a segment was out which was more important then the actual number which is another reason why I ended up making my own routine.
Wouldn't the number "8" turn them all on?
It was a consumer electronic device that was all packaged up. End of line testing type of thing.
Still the technique was quite robust and only relied on the fact that you had an image before the device was turned on, then take a picture after it was turned on, and that the UUT hadn't moved between pictures. The lens, mirror and lighting were all mediocre but the technique was so good it over came these short comings.
