[Chris] tried his hand at using Optical Character Recognition in his server power monitoring rig. The image above is what the IP camera used in the setup sees. He’s included a bright light to ensure that the contrast is as great as possible. After applying a threshold filter to the captured still, he is able to process the image to test all seven segments of every digit.
He uses Mathematica for the processing. We’re not familiar with the particulars of the language, but it’s easy enough to see the main parts of the program. Line six of his source code applies the image filters and then the program loops through the assigned location of each digit, testing segment combinations to ascertain what number is shown. Things get hairy when it comes to the decimal point. We gather that the meter can show varying degrees of precision based on the total number of digits needed (like a Digital Multimeter). But [Chris'] setup has a difficult time reliably detecting that decimal point because of its size. He uses a shortcut to get around this, knowing that his server never pulls less than 300W so he corrects the output (by multiplying it by ten) if the reading is below that benchmark.
Of course it would be easier to crack open the monitor and glean data electronically (that’s how the Tweet-A-Watt does it) but then [Chris] wouldn’t have had the fun of playing with OCR.