Enhancing License Plate Recognition in Videos Through Character-Wise Temporal Combination

License Plate Recognition (LPR) in videos is a critical task in various domains such as parking management, traffic control, and security. This study focuses on exploring the significance of temporal information in LPR systems to improve their accuracy. Previous research has not fully leveraged temporal information, resulting in multiple prediction results for the same vehicle. Unlike other tasks, LPR generates a sequence of characters per frame, necessitating a distinct approach for combining these outputs. To address this, we conducted a comprehensive investigation of LPR pipelines and different frame combination techniques. Our study introduces a new strategy for character-wise temporal sequence combination, enhancing the accuracy of license plate recognition. The proposed approach was evaluated using a new dataset of Cuban license plates captured in parking lot scenarios showing superior results in most cases. This study can serve as guide for future research and applications in the field of LPR in videos.