To address the issues in real-world videos identified across the existing algorithms 1- 13, in this study, we designed a CNN architecture with temporal matching priors between adjacent frames. The main issues are: (a) the vehicle motion blur and geometric distortion, (b) outlier objects such as signboards containing text quite similar to license plates, and (c) the lack of a public training dataset for plate detection and recognition in videos. Meanwhile, despite promising results on several refined still image datasets, the algorithms in 1- 13 still suffer from significant limitations when applied to video data. Although Korean license plates were handled in 8, 12, their data are not available and, in all cases, the amount of data is not sufficient to validate the Korean license plate recognition performance. Existing approaches focus mainly on license plates from specific countries including US, Europe, Taiwan, Brazil, and China. Table 1 provides details of the existing approaches 1- 13 including methods, objectives, datasets, target countries, accuracies, and processing times. Drawn by these increasing needs, state-of-the-art algorithms based on convolutional neural networks (CNNs) have been proposed for license plate detection and recognition 1- 13. During network training, we perform data augmentation based on image rotation to increase robustness regarding the various viewpoints in videos.Īutomatic vehicle license plate detection and recognition are increasingly important in intelligent transportation systems, and they play a key role in various areas such as unmanned parking and traffic control. We also built our own video dataset for the deep training of the proposed network. Specifically, to improve the robustness of detection and recognition accuracy in the presence of motion blur and outliers, forward and bidirectional matching priors between consecutive frames are properly combined with layer structures specifically designed for plate detection. In this study, we focus on these challenges and propose a license plate detection and recognition scheme for videos based on a temporal matching prior network. This stems from several problematic conditions in videos, such as vehicle motion blur, variety in viewpoints, outliers, and the lack of publicly available video datasets.
Many algorithms have been proposed for still images, but their accuracy on actual videos is not satisfactory. In real-world intelligent transportation systems, accuracy in vehicle license plate detection and recognition is considered quite critical.
0 kommentar(er)
