Unified Deep Learning Pipeline for License Plate Recognition: From Detection to Ocr With Faster Rcnn, Fcn, and Crnn

We propose a fully deep-learning-based ALPR pipeline that unifies license plate detection using Faster R-CNN, character segmentation using a fully convolutional network (FCN), and optical character recognition using a convolutional recurrent neural network (CRNN) trained with connectionist temporal classification (CTC) loss. Evaluated on $\mathbf{5, 0 0 0}$ mixed-condition roadway images, the system achieves a mean average precision (mAP) of 90.5 % at IoU 0.5 for detection, 85.6 % IoU for segmentation, and 95.2 %-character recognition accuracy while operating at approximately 25 frames per second on an NVIDIA RTX 3080 GPU, yielding an 88.7% end-to-end plate accuracy. Compared to a YOLOv5 + Tesseract hybrid, our pipeline achieves up to 8.6 percentage points higher overall accuracy, demonstrating that replacing all three ALPR stages with task-specific deep models significantly improves robustness to blur, low light, and unconventional fonts, without sacrificing real-time performance.