Again this is a term project for Soft computing course. The results were actually amazing considering I never understood how it worked ūüėČ I guess that is how neural nets are...You only know how to train them and control the learning parameters and apply regularization. But once trained the net is nothing more than a matrix of data giving¬†absolutely¬†no idea on how it actually works. For that reason I don't like many of the soft computing techniques because they may solve a problem at hand but they don't give any insight to how things are working! The ability to acquire license plate information in ASCII text format opens up many diverse and utilitarian applications. License plate recognition (LPR) is a key technique to many automated systems such as road traffic monitoring, automated payment of tolls on high ways or bridges, security access, and parking lots access control. Difficulties result from illumination variance, noise, complex and dirty background. In this paper, an automated license plate recognition system is proposed based on image processing, feature extraction and neural networks. This system utilizes colour based license plate localization for finding and isolating the plate on the picture, which further goes through multiple levels of image pre-processing before final stage of character segmentation and neural network based recognition. Back-propagation neural network (BPNN) is selected as a powerful tool to perform the recognition process. Excerpt from the paper -
Abstract
The ability to acquire license plate information in ASCII text format opens up many diverse and utilitarian applications. License plate recognition (LPR) is a key technique to many automated systems such as road traffic monitoring, automated payment of tolls on high ways or bridges, security access, and parking lots access control. Difficulties result from illumination variance, noise, complex and dirty background. In this paper, an automated license plate recognition system is proposed based on image processing, feature extraction and neural networks. This system utilizes colour based license plate localisation for finding and isolating the plate on the
picture, which further goes through multiple levels of image pre-processing before final stage of character segmentation and neural network based recognition. Back-propagation neural network (BPNN) is selected as a powerful tool to perform the recognition process.
Problem Definition
Automatic recognition of license plate is an essential stage in intelligent traffic system, automated traffic management and law enforcement. Real time LPR plays a major role in automatic monitoring of traffic rules and maintaining law enforcement on public roads. The automatic identification of vehicles by the contents of their license plates is important in private transport applications. License plate recognition is a complex problem. The steps involved in recognition of a license plate can be categorised in brief into image acquisition, candidate region extraction, segmentation, and recognition. Process starts with image acquisition. This can be done using high resolution CCD colour cameras with fast shutter speeds, which will enable the LPR to perform from greater distance and on fast moving vehicles or a CMOS based low resolution camera that can perform well in restricted situations like an automated parking lot where images will be taken from close proximities and in good lighting conditions. Next task is finding and isolating the plate on the picture. Success of further stages solely lies on its result. The task is compounded due to variation in illumination, noise, motion blur and complicated background. Localisation is achieved on the basis of colour and dimension. Based on plate localisation
information the image is cropped and sent to multiple stages of image pre-processing which removes noise, increases contrast and convert the image to binary which assist individual character segmentation. These segmented characters are then resized into a two dimensional grid of 15X10 size. Information thus obtained is fed to a back propagation neural network with 150 neurons (one for each grid element) in input layer, 1 hidden layer with 40 neurons and finally the output layer with 36 neurons, with each neuron corresponds to a number from 0 to 9 or character from A to Z. The neuron which corresponds to input data fires with highest weight.
You can download the paper here Fully Automated license Plate recognition

Stages of operation

Print This Post Print This Post

Looking for something?

Use the form below to search the site:

Still not finding what you're looking for? Drop a comment on a post or contact us so we can take care of it!

Visit our friends!

A few highly recommended friends...

Archives