Missing data detection and imputation for urban ANPR system using an iterative tensor decomposition approach

Abstract

The automatic number plate recognition (ANPR) system has been widely implemented as an important part of intelligent transportation system (ITS). However, similar to other traffic monitoring devices, missing data is a common and critical problem in the ANPR system. To solve the missing data problem, numerous tensor-based methods have been proposed in previous studies. Most of them, however, assume that where and when missing data occur in the dataset are known. This would be impractical, because missing data may occur randomly. In this study, we propose a novel tensor-based algorithm, specifically, an iterative tensor decomposition (ITD) approach, that utilizes multidimensional inherent correlation of traffic data to detect and impute missing data in the ANPR system. The proposed algorithm is tested with a real-world ANPR system dataset. The experimental results show that missing data from the ANPR system can be classified into three cases, i.e., no missing, random elements missing, and extreme missing. The proposed ITD can accurately detect and correct missing data under different missing cases. Furthermore, ITD is also compared with other state-of-the-art methods and the results show that ITD outperforms the existing methods.