Critical analysis of deep neural network on Malaysia license plate recognition

Lee, Wei Xiang (2021) Critical analysis of deep neural network on Malaysia license plate recognition. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Critical analysis of deep neural network on Malaysia license plate recognition.pdf - Submitted Version Download (2MB)

Abstract

Nowadays, traffic congestion in Malaysia is getting serious. To optimize the traffic flow, creating an automatic license plate recognition system that can collect the traffic flow parameter will be an ideal solution. To create a good automatic license plate recognition model, the effect of variety prefix on neural network needs to be study as there is variety prefix in Malaysia's license plate. Hence, the aim of the project is to analyse the deep neural network on the variety of prefix in Malaysia license plate recognition. To obtain the results, three Malaysia license plate recognition model was train with three different datasets, together with three different distributions of prefix of Malaysia license plate. Besides that, a license plate detector (LPD) was also being train to automate the process of cropping the license plate from the source image captured from overhead poles camera at traffic light junctions. The LPD was trained and achieved 75.8% mAP and 99.8% accuracy in automatic detecting the license plate given a vehicle image. All the LPR models (Model A, B, and C) manage to have 79.58%, 81.25% and 80.42% in mean test full sequence accuracy after training.

Actions (login required)

View Item