These are samples of the license plates I’m trying to extract the characters from.
I am trying to figure out a way to detect, the characters inside the plate. I first extracted the plate using YOLO, and then I thought of using multi-label classification to get the characters, but then I won’t be able to get the sequence the characters came in. I had another idea of localizing the area where the characters reside and applying a sliding window classifier where the model here will be trained on Arabic characters. However, this is computationally expensive where the plate will undergo lots of predictions to extract the characters. I need help with a better approach than these.
Sometimes the only solution is one that requires a lot of computation.
Yes, that is true, but I am unsure if this approach is the right way to tackle this problem.
@ahmedemam little out of my league here and don’t have enough experience to know if this pattern holds-- But one quick observation from the three examples is the characters to the left all seem to be numeric and on the right, alphabetic.
If you could split the image of the plate in this way then you have two smaller subset models you could run separately in parallel (or it would decrease your search space).
You are correct; however, this is not consistent because some plates had a mix of both letters and numbers on both sides.
You’ve completed a literature search for prior art, right? And read papers like…
Deep Machine Learning Based Egyptian Vehicle License Plate Recognition Systems
A low cost IoT-based Arabic license plate recognition model for smart parking systems
https://www.sciencedirect.com/science/article/pii/S2090447923000679
Arabic Vehicle Licence Plate Recognition Using Deep Learning Methods: Review
What approaches did they take? What worked well? What limitations?
1 Like