License Plate Detection and Recognition based on Neural Networks in Complex Environments

Now a days due to the rapid advancement of economy around the world the count of vehicles increases day by day. Increase in the number of vehicles causes violation detection, road congestion, accidents at different traffic situations, uneven illumination, lighting and weather conditions. To overcome this issue license plate number is recognized but due to variations in license plate layout, font size of characters, tilted number plates, weather conditions, dirt plate and motion blur license plate recognition becomes difficult. License plate recognition has two main tasks, one is to detect the license plate and the other is to identify the license plate characters. By using region of interest license plate is detected. For recognition first tilted images are corrected using affine transformation and to improve the quality of a low-resolution image super resolution CNN is employed and connected component analysis, horizontal and vertical projection profile area used for separating each individuals characters. Each individual character image is fed to the Convolutional Neural Network (CNN) for character extraction and for classification and the license plate is recognized using convolutional neural networks. The main aim of this paper is to recognize different plate layout with different conditions with minimum data set and less processing time with maximum efficiency.