Automatic Number Plate Recognition Using CNN
- 0 Collaborators
Project covers detecting of number number plate using Haar-Cascade,and then segmenting out individual digits and feeding them to trained CNN ...learn more
Groups
Student Developers for AI,
Movidius™ Neural Compute Group
Overview / Usage
Achieving high accuracy(97%+) on data set MNIST (which contain binary images of handwritten digit) motivated us to train the CNN to classify the number plate letters.We used Haar-Cascade with russiannumberplate pretrained classifier to detect number plates.Then we used Open-CV to find contours and extracted individual number which lie in certain aspect ratio.Last step was to use Caffe by berkeleyvision to classify the detected digits.
We achieved 97% accuracy using the above architecture.
Note:Currently i don't have my lappy with me so i will upload the images soon.
Note:To request source code drop us a mail.