| Abstrakt: |
Due to the huge rise in the number of vehicles, manual tracking has become a complex task. To deal with this issue, Automatic License Plate Recognition systems have been developed to identify vehicles in real-time situations. This paper proposes a novel automatic license plate recognition approach based on multi-attribute data fusion. The proposed approach's architecture consists of three steps: localization, segmentation, and recognition. For localization, the Yolo object detector is used. The segmentation step is accomplished following two ways: the first method relies on the Yolo object detector, while the second method relies on edge detection. Finally, the recognition step targets the extraction of descriptive attributes from the previously segmented plate images. These attributes are then fused and combined using the mathematical Dempster–Shafer theory and Parzen-Rosenblatt windowing for belief mass building. The proposed system underwent testing and validation using an Algerian license plate dataset, yielding impressive accuracy rates of 98.90% for localization, 98.10% for segmentation, and 97.20% for recognition, outperforming other existing approaches. Furthermore, a comparison is conducted on the OpenALPR-EU, OpenALPR-BR, and SSIG datasets, demonstrating a competitive accuracy rate of 92.23%, 92.45%, and 90.84%, respectively for the proposed approach, comparable to both alternative methods and commercial systems. This good performance can be attributed to the robustness of Yolo in localizing and segmenting characters, as well as the strategic selection and the evidential fusion of significant attributes, leading to better character distinction. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink