Driver License Field Detection Using Real-Time Deep Networks

We present an automatic system for real-time visual detection and recognition of multiple driver’s license fields using an effective deep YOLOv3 detection network. Driver licenses are essential Photo IDs frequently checked by law enforcement and insurers. Automatic detection and recognition of multiple fields from the license can replace manual key-in and significantly improve workflow. In this paper, we developed an Intelligent Driving License Reading System (IDLRS) addressing the following challenging problems: (1) varying fields and contents from multiple types and versions of driver licenses, (2) varying capturing angles and illuminations from a mobile camera, (3) fast processing for real-world applications. To retain high detection accuracy and versatility, we propose to directly detect multiple field contents in a single shot by adopting and fine-tuning the recent YOLOv3-608 detector, which can detect 11 fields from the new Taiwan driver license with accuracy of 97.5%. Our approach does not rely on text detection or OCR and outperforms them when tested with large viewing angles. To further examine such capability, we perform evaluations in 4 large tilting view configurations (top, bottom, left, right), and achieve accuracies of 93.3%, 90.2%, 97.5%, 94.3%, respectively.