Detection and recognition of license plate using cnn and lstm

Abstract

Due to the world’s rapid economic growth, the motor vehicles’ number has significantly increased which lead to the necessity for the security of vehicles. It becomes difficult to manually track the vehicles which are over speeding and to monitor the traffic so as to avoid congestion problems, tracking stolen cars, detecting the unlawful activities that involve a vehicle. Automatic number plate recognition (ANPR) is an image processing technology which allows users to track, identify, and monitor moving vehicles by automatically detecting the license plate from the vehicle. A convolutional neural network (CNN) is used which slides over the entire image so as to detect the characters from an image. After detecting the characters, the non-plate region must be eliminated from plate region. For that a plate/non-plate CNN classifier is used. After detecting the license plate, LSTM is being used for obtaining the feature sequence. Then the characters are decoded, and the license plate is recognized. By combining both CNN and LSTM methods, it gave an accuracy of 85.0%.