Serial Code Capture- Tensorflow+OpenCV+Python+Tesseract
The serial numbers are being read off in the black box to the right. It does get some false readings but any numbers not matching the format of the serials can be easily rejected, and correct serial numbers can be stored.
There are a great many improvements still to be made, for instance I was considering re-training the Tensorflow neural network using the EAST text region detector, as it is I am using a simple RCNN trained to recognize readable tags, unreadable tags, the serial number section of the tag, and whether or not the serial is in view enough to read the number.
The video lag does not effect the results, only what the viewer sees on-screen, and I am confident I can fix it.
I spent a good amount of time programming the processing of the image to get it to an ideal readable state, from binary thresholding and median blurring to resizing, but one HUGE improvement I'm still working on is fixing the image skew, the results will be much better after this. Additionally, the model was only trained on about 1000 pictures, and I feel I could have used a better training set than I did. So a major improvement can be made by retraining the model on at least 10,000 photos in order to better recognize the right "moment" to attempt the serial read.
This is a work in progress and not intended to be a finished product. There are bugs to work out.
Anyone interested in utilizing my skills for the benefit of your business- contact me. There's a good chance I can come up with a way that AI and machine vision can improve your enterprise.
-Charles Niswander, Michigan
In questa pagina del sito puoi guardare il video online Serial Code Capture- Tensorflow+OpenCV+Python+Tesseract della durata di ore minuti seconda in buona qualità , che l'utente ha caricato Charles Niswander 10 gennaio 2019, condividi il link con amici e conoscenti, su youtube questo video è già stato visto 742 volte e gli è piaciuto 9 spettatori. Buona visione!
