Weakly Supervised Character Detection for License Plate Recognition

• Luis F. A. Zeni UFRGS
• Claudio R. Jung UFRGS

Resumo

Automatic Licence Plate Recognition (ALPR) is an essential task in the context of intelligent transportation systems. In a typical ALPR pipeline, the last stage receives as input a cropped license plate region and outputs the string with the plate characters. This paper presents a Weakly Supervised Character Detection (WSCD) approach that requires only string-level annotations (as in generic text recognition methods) but is able to detect characters individually (as in detection-based methods, which require character-level annotations). The proposed method is evaluated in five distinct datasets and present very competitive results against other state-of-the-art methods.

Palavras-chave: weakly supervised character detection, character detection, licence plate recoginition, neural networks, weak supervison