7

I have a large volume of video from cameras set up to record vehicles registrations. Unfortunately it was necessary to carry out much of the observation during hours of darkness and a large proportion of the footage is "blurred". Consequently, the package usually used for OCR of the vehicle registrations (3M autoplate) is not able to process the data.

It would be inconvenient to repeat the observations. I question whether there are image processing techniques, or opensource/propriety software that would make sense of the images and return a reliable vehicle registration.

I include below a dropbox link to 6 consecutive frames of a vehicle that was recorded. These images were generated using the "take snapshot" tool in VLC from the original *.avi file. I can generate screenshots in a different if helpful, or else provide a brief snippet from the video footage.

https://www.dropbox.com/sh/4ne7m6fk6uerxbw/AAAnYCLL1II-_EMlh359IFs4a?dl=0

Ideally I require a solution that can be effectively applied to 100+hrs of video, returning perhaps 50,000 vehicle registrations, and including a confidence statistic for each.

(There is no data protection issues with the vehicle shown in these images as permission has been granted from its owner.)

  • 1
    +1 for that final statement about permission – Jan Doggen Jan 8 '15 at 13:48
  • Do you have a specific operating system in mind? Must it be a free solution, or do you have a budget, please read How to Ask. The more help that you give us, the more than we can help you. – Mawg Dec 8 '15 at 13:15
1

Software for plate identification is plenty, so I assume the main problem is OCR of the images as appropriately extracted from the footage (the title suggests so).

Presumably in your area you'll have plates of rather similar sources, hence some image cleanup + tesseract training should do the job. Scan Tailor is a commonly used solution to perform both: http://scantailor.org/

Terese might be easier to use programmatically: http://terese.sourceforge.net/

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.