Moving camera automatic number plate recognition using neural network in android platform

• Main Author: Conny Liew
• Format: Thesis
• Language: English; English
• Published: 2019
• Subjects: TJ Mechanical engineering and machinery
• Online Access: https://eprints.ums.edu.my/id/eprint/25121/1/Moving%20camera%20automatic%20number%20plate%20recognition%20using%20neural%20network%20in%20android%20platform.pdf; https://eprints.ums.edu.my/id/eprint/25121/6/Moving%20Camera%20Automatic%20Number%20Plate%20Recognition%20Using%20Neural%20Network%20In%20Android%20Platform.pdf; https://eprints.ums.edu.my/id/eprint/25121/
• Institution: Universiti Malaysia Sabah

Automatic number plate recognition (ANPR) system has been widely used in many applications such as electronic payment gateway system, parking fee payment system, road monitoring system and traffic control system, as named a few. Till today, there are many methods have been proposed and developed to solve the ANPR related problems. However, most of the researches conducted were focused on fixed camera instead of moving camera. It is easy to localize and recognize the scanned car plate with fixed camera as the car plate position is almost static and can be estimated easily. But images captured using moving camera is very complex due to the background changes rapidly. Conventional ANPR system usually requires high processing power CPU and high resolutions camera, which is usually bulky and highly expensive. Recent technological advancement in smartphone industry have seen that many smartphone devices are now equipped with high end processor and high resolutions camera. Besides that, Android mobile platform is open source and equipped with many useful libraries to easily modify hardware setting programmatically such as camera. This allow the implementation of ANPR system in mobile device is possible. Based on the literature review, there are several researches have implemented ANPR system in mobile platform, but most of the researches detect license plate from static image or directly pointing the phone camera to capture license plate image. Hence, this research is conducted to provide study of the implementation of moving camera ANPR system in real time environment using Android mobile platform. This research aims to (1) design and implement image processing step for moving camera ANPR system in Android mobile platform, (2) design and implement Convolutional Neural Network (CNN) and Backpropagation Feed Forward Neural Network (BPFFNN) algorithms for moving camera ANPR system in Android platform, and (3) test and compare moving camera ANPR using CNN and BPFFNN in Android platform. The proposed localization step includes combination of Sobel edge detection method and morphological based method. Successfully detected license plate image is segmented and each character is bounded with a rectangular bounding box and cropped out. Each cropped character is feed into CNN or BPFFNN model for character recognition process. The NN model is pretrained in desktop computer or notebook and the trained NN model is then exported and implemented in the Android platform. There are five preliminary experiments have been conducted to identify the license plate search area, camera resolution of ANPR system, distance of camera with target license plate and finally, relative speed of moving camera with target license plate and vice versa. The experimental results show that the proposed technique could automatically recognize license plate in real time. Then, the performance of the proposed CNN and BPFFNN had been compared. It shows that the CNN experimental result performed better compared to BPFFNN in ANPR with moving camera approach.